DEV Community: MillionMiner

The Bitcoin Mining Death Spiral Everyone Fears: Does It Actually Exist?

MillionMiner — Wed, 22 Jul 2026 09:03:21 +0000

The theory sounds airtight: price crashes, miners quit, the network dies. Here is why it has never happened in sixteen years, and the one mechanism that makes it almost impossible.

Right now, in mid-2026, the whispers are back. Bitcoin has fallen from its late-2025 high to around $64,000, miners are switching off machines faster than at any point since China banned mining in 2021, and the network's difficulty just dropped more than 10 percent in a single adjustment, over 20 percent below its peak. Every time this happens, the same fear resurfaces on timelines and in group chats: is this finally the death spiral?

It is a genuinely frightening idea, and on paper it looks unstoppable. It is also something that has never once happened in Bitcoin's sixteen-year history, through crash after brutal crash. What you are watching in 2026 is not the beginning of a death spiral. It is, almost exactly, the opposite. Here is why.

The theory, stated fairly

Let us give the death spiral its strongest form, because it deserves one. The logic goes like this. Bitcoin's price crashes hard. Mining becomes unprofitable, so miners switch off their machines to stop losing money. The network's hashrate, the total computing power securing it, falls. With fewer miners, the fear continues, blocks are found more slowly, the network looks weak and less secure, and confidence erodes. That falling confidence pushes the price down further, which makes even more miners unprofitable, so they leave too, which weakens the network more, which drops the price again. Round and round, faster and faster, until the whole thing collapses to zero.

Stated that way, it sounds airtight, a perfect self-reinforcing loop with no obvious way out. If mining and price only ever pushed each other in the same direction, Bitcoin really would be fragile. The entire theory rests on that assumption. And that assumption is where it falls apart, because of one mechanism built into Bitcoin from the very first block.

The one mechanism that breaks it: difficulty adjustment

Bitcoin has a thermostat. Roughly every two weeks, precisely every 2,016 blocks, the network automatically re-tunes how hard it is to mine a block, with one goal: keep blocks arriving about every ten minutes no matter how much or how little computing power is online. This is the difficulty adjustment, and it is the single reason the death spiral does not work.

Watch what it does to the spiral. Miners switch off, so blocks start arriving more slowly than every ten minutes. At the next adjustment, the network sees those slow blocks and lowers the difficulty to compensate. Suddenly, mining is easier, and every machine still running earns more Bitcoin for the same work. The exact event the death spiral treats as the beginning of the end, miners leaving, is what makes it more profitable for everyone who stayed.

This is not theory, it is the live data. When difficulty fell 10.09 percent in June 2026, the miners still online immediately began earning roughly 9 to 11 percent more Bitcoin per machine, and hashprice, the revenue a unit of computing power earns, jumped about 13 percent in response. The network did not weaken. It handed its survivors a raise.

Here is the idea underneath it, and it is worth keeping. A death spiral requires a positive feedback loop, where leaving causes more leaving. Bitcoin's difficulty adjustment is a negative feedback loop, where leaving makes it profitable to stay. You cannot build a runaway collapse on a mechanism that pushes back harder the more it is stressed. The death spiral, in effect, eats itself.

The proof, in real time

If that still sounds like a comforting story, the 2026 data tells it plainly. Earlier this year, Bitcoin's difficulty dropped around 11 percent as miners powered down under margin pressure. Within two weeks, at the very next adjustment, it snapped back with a record upward move of nearly 15 percent, as miners flipped their machines right back on the moment conditions eased.

Sit with that for a second, because it is the whole argument in one data point. Hashrate did not drain away and keep draining, the way a spiral demands. It fell, the network made mining more profitable, and it came straight back. The operators who survive a downturn are committed and well capitalized, and they scale up the instant it pays to, which is exactly what a negative feedback loop looks like from the outside. The system is not fragile. It is springy.

Why there is always a miner left standing

The spiral also assumes that when mining becomes unprofitable, it becomes unprofitable for everyone at once. It does not, because miners do not share the same costs.

Some miners run three-year-old machines on expensive residential-grade power. Others run the latest sub-15 joules-per-terahash hardware on industrial electricity at a few cents per kilowatt-hour. When the price falls, the high-cost, inefficient operators hit their break-even first and switch off, while the efficient, low-cost fleets keep mining profitably long after. The floor is set by whoever has the cheapest power and the most efficient machines on the planet, and someone always does. So hashrate does not fall to zero. It falls until only the leanest operations remain, and then it stops falling, because those operators are still making money. Efficiency, and the electricity rate behind it, is what decides who is standing when the dust settles.

It has never happened, and we have watched it get tested

The strongest evidence is simply history. Bitcoin has lived through repeated crashes of 50 to 80 percent and multiple mass miner exits, and the death spiral has not arrived once.

The clearest stress test came in mid-2021, when China banned mining outright and roughly half of the entire network's hashrate went dark in a matter of weeks, the largest drop in Bitcoin's history. By the death spiral's logic, that should have been fatal. Instead, blocks slowed for a couple of weeks, the difficulty adjusted sharply downward to compensate, the remaining miners earned more, and displaced miners relocated and plugged back in elsewhere. Within months the network had fully recovered. The 2018 and 2022 bear markets told smaller versions of the same story: prices cratered, inefficient rigs were purged, difficulty fell, and the network kept producing blocks the entire time and recovered when capital returned. The mechanism has been tested under the worst conditions anyone could design, and it has held every time.

The honest part: where the fear has a kernel of truth

None of this means the concern is stupid, and it is worth being precise about what is real, because two things genuinely are.

First, between adjustments, blocks really do slow down. Difficulty only re-tunes every two weeks, so if a large chunk of hashrate leaves suddenly, transactions can confirm more slowly until the next retarget catches up. The June 2026 epoch ran about 15.6 days instead of the usual 14 for exactly this reason, and after the China ban blocks were noticeably slow for a couple of weeks. It is a temporary inconvenience, not a collapse, and the adjustment always resolves it, but it is real.

Second, and more importantly, the death spiral is a genuine risk for small proof-of-work coins. A cryptocurrency with a thin, low-value hashrate does not have the deep bench of profitable miners that Bitcoin does, and it can suffer a real spiral, or a cheap attack, when miners leave. This is precisely why Bitcoin's enormous scale is not a vanity metric but its core defense. The thing that makes the death spiral impossible for Bitcoin is the same thing that makes it possible for a tiny altcoin: the sheer amount of committed hashrate and the global spread of cheap power behind it. The theory is not wrong in general. It is wrong for Bitcoin, specifically because of its size.

What is actually happening in 2026

So when you read that miners are capitulating and difficulty is falling, reframe it. Capitulation is not the network dying, it is the market clearing. The weakest, highest-cost operators are being flushed out, difficulty is dropping to reward whoever remains, and the network is emerging leaner, more efficient, and cheaper to secure per unit of Bitcoin produced. Historically, this phase has marked bottoms rather than endings, precisely because it clears out the fragile capacity and hands the survivors a bigger share. What looks like the network spiraling down is the network finding its floor.

That is also why seasoned operators treat a deep capitulation as a signal rather than a warning. It is uncomfortable and it is not for everyone, and profitability is never guaranteed, since it still depends on the price and on your own power costs. But the fear driving people out at the bottom is, more often than not, the very mechanism that rewards the people who stay.

The bottom line

The Bitcoin mining death spiral is a real theoretical worry that Bitcoin's difficulty adjustment and its sheer scale have defeated every single time they have been tested. The mechanism that the spiral treats as fatal, miners leaving, is the mechanism that makes it profitable for miners to return. The 10 percent difficulty drop and the capitulation you are reading about in 2026 are not the network breaking down. They are the network doing exactly what it was designed to do, in public, on schedule, for the sixteenth year running.

Frequently asked questions

What is the Bitcoin mining death spiral?
It is a theory that a falling Bitcoin price forces miners to shut down, which reduces hashrate, which supposedly weakens the network and drives the price down further, causing even more miners to leave in a self-reinforcing collapse. It is a real concern in principle, but Bitcoin's difficulty adjustment prevents it in practice, and it has never occurred in Bitcoin's history.

Why does the difficulty adjustment prevent a death spiral?
Because it makes mining easier and more profitable exactly when miners leave. Every 2,016 blocks, about two weeks, Bitcoin re-tunes difficulty to keep blocks near ten minutes apart. When hashrate drops, difficulty falls, so the remaining miners earn more per machine, which gives them a reason to keep mining and gives others a reason to switch back on. It is a self-correcting loop, the opposite of a runaway collapse.

Does a lower hashrate make Bitcoin less secure?
It reduces the raw cost of attacking the network, but Bitcoin's hashrate remains enormous even after large drops, so the practical security margin stays very high. More importantly, a lower hashrate is temporary: as difficulty falls and mining becomes profitable again, hashrate returns, as it did after the 2021 China ban and repeatedly in 2026. The network self-heals rather than decays.

Is the 2026 miner capitulation a death spiral?
No. It is the ordinary, healthy process of the market clearing. High-cost and inefficient miners are shutting down under margin pressure, difficulty is falling to reward those still running, and the network is becoming leaner. Historically this phase has marked cycle bottoms rather than endings, because it clears weak capacity and benefits the survivors.

Can a death spiral happen to other cryptocurrencies?
Yes. Small proof-of-work coins with thin, low-value hashrate are genuinely vulnerable, because they lack Bitcoin's deep pool of committed, low-cost miners and can suffer real collapses or cheap attacks when miners leave. Bitcoin's immunity comes specifically from its scale, which is why the same theory that fails for Bitcoin can succeed against a tiny altcoin.

The takeaway for miners
What looks like the start of a spiral is the network clearing out its highest-cost capacity and rewarding whoever runs efficiently on cheap power. If you want to see where your own break-even sits, model it against your electricity rate, and if home power is the problem, hosting on industrial rates is usually the fix.

Model your break-even: mining profitability
Survive the downturn on cheap power: ASIC hosting

Bitcoin Mining and AI Are Becoming the Same Business, and 2026 Is the Year to Get In

MillionMiner — Tue, 21 Jul 2026 09:02:51 +0000

The same warehouses, the same cheap power, the same racks of hot chips, now running both Bitcoin miners and AI GPUs. Here is why the two industries merged, and every way to get in.

The most valuable real estate in technology right now is not in Silicon Valley. It is a warehouse next to a power substation. It may have been built to mine Bitcoin, and today it can just as easily run the AI models everyone is talking about, because it turns out those two things need the exact same ingredients: enormous amounts of cheap electricity, industrial cooling, and racks of chips running flat out around the clock. Bitcoin mining did not die. It quietly became the blueprint for the AI data center, and that convergence is the most exciting opportunity in this space in years.

If you have been waiting for a reason to get involved, in either Bitcoin or AI, this is the moment the two paths opened up at once. Here is what is actually happening, and every door you can walk through.

Why mining and AI are suddenly the same business

Strip away the buzzwords and a Bitcoin miner and an AI GPU are the same kind of object: a specialized chip that eats electricity and produces heat, running at full load every hour of every day. The differences that matter are not philosophical, they are physical. An AI accelerator like a B200 draws around 1,000 watts, and a single rack of them pulls well over 100 kilowatts and needs liquid cooling. The thing standing between a company and more AI is almost never the chips. It is power, and a place built to run it.

Now ask who spent the last decade securing cheap power contracts, negotiating with grid operators, building out megawatts of cooling, and running warehouses full of hot silicon nonstop. Bitcoin miners. They built, through years of trial and brutal margin pressure, the precise infrastructure that AI is now desperate for. The power agreements, the substations, the cooling, the operational know-how of keeping thousands of chips alive, all of it transfers. That is why the line between a Bitcoin mining farm and an AI data center has basically dissolved, and why serious operators now build facilities that can do either.

The great pivot everyone is watching

This is not a prediction, it is already underway. Across 2025 and 2026, some of the largest public Bitcoin miners began converting parts of their sites to AI and high-performance computing, signing enormous deals to rent their power and space to AI companies. The logic is simple and a little brutal: for a given megawatt of cheap power, AI compute can, in the right conditions, pay more than mining does, so operators who own the power now get to choose what to point it at. The power that used to be pure mining infrastructure is becoming the foundation of the AI build-out, and the miners who pivoted early into AI got a head start on the biggest demand wave in computing.

What makes this exciting rather than threatening is that it does not replace Bitcoin mining, it sits alongside it. The same operator can mine when mining is more profitable and host AI when AI is, and the same building supports both. For anyone with capital, power, or just the interest to participate, that means there have never been more ways in.

Two doors, the same building

Here is the honest map of how you actually get involved in 2026, on either side of the convergence. Start with the path that fits, and use the tools to figure out which one that is.

The Bitcoin path

If you want to mine, the decision is no longer just "buy a machine and plug it in at home," because home power is usually too expensive to profit. The real options are about where and how your hardware runs.

The simplest way in is to own an efficient miner and run it somewhere built for it. ASIC hosting puts your machine in a facility on industrial-rate power with cooling and staff around the clock, which is what makes modern miners profitable when your kitchen outlet would not. If you want that at institutional scale and with compliance built in, regulated US mining farms offer professionally run, above-board facilities, and if you would rather acquire a whole operation ready to run, turnkey mining farms and AI data centers are exactly the converged asset this article is about, built to mine Bitcoin, host AI, or both.

For the efficiency frontier, hydro mining runs the newest sub-10 joules-per-terahash machines on water cooling, the most competitive way to mine that exists today. If you want exposure to mining without owning or hosting hardware at all, cloud mining lets you rent hashrate directly, and wherever your machines run, mining pools are where you point them to earn a steady, shared share of the network instead of waiting years alone.

And before you commit a single dollar, model it. Run your numbers and your power rate through the mining profitability calculator, cross-check with the dedicated BTC mining calculator, and if you would rather see it with your own eyes than trust a spreadsheet, a free 24-hour miner test lets you watch real hardware mine before you buy anything.

The AI path

If the other side of the convergence is what excites you, the entry points are just as concrete. The build-out running short of compute means the people who supply and run AI hardware are in an extraordinary position.

It starts with the hardware itself. AI hardware, the H100, H200, B200, and the GPUs and servers around them, is the raw material of the entire AI economy, and getting the right configuration matters as much as getting it at all. Because the newest chip is not always the right one, the GPU and AI benchmark scores real inference performance and shows what actually fits your workload, so you buy for the job instead of the spec sheet. And when you are ready to run that hardware at scale, the same turnkey data-center capacity that hosts mining hosts AI, on the cheap power and cooling that make the economics work.

The common denominator, and the honest part

Whichever door you choose, the thing that decides whether it pays is identical, and it is worth saying plainly. Both Bitcoin mining and AI compute are real businesses with real costs, and both live or die on the same two things: access to cheap power, and efficient hardware run at high utilization. Neither is free money, neither is guaranteed, and anyone who tells you otherwise is selling a fantasy. The convergence is genuinely exciting because it opens more paths and makes the underlying infrastructure more valuable, not because it removes the need to do the math.

So do the math. The reason the tools above exist, the calculators, the benchmark, the free test, is that the right answer depends entirely on your specific power rate, your hardware, and your goals. Model your situation honestly, pick the path that fits, and the convergence works in your favor. Skip that step and no amount of trend will save a bad setup.

The bottom line

Bitcoin mining and AI have merged into a single question: what is the best thing to do with a megawatt of cheap power and a building full of chips? That question is being answered in real time by the biggest operators in the world, and in 2026 the doors are open to everyone else too. Mine Bitcoin through hosting, a regulated farm, a turnkey site, hydro, or the cloud. Power AI through the hardware, the benchmarks, and data-center capacity. Or do both, from the same foundation.

The infrastructure that Bitcoin miners spent a decade building turned out to be the infrastructure the future runs on. The only real question left is which door you walk through, so figure out which one fits, model it, and go.

Frequently asked questions

Is Bitcoin mining dying because of AI?
No. Mining is being joined by AI, not replaced by it. Many operators now run facilities that can do both, mining Bitcoin when it is more profitable and hosting AI compute when it is, from the same power and cooling infrastructure. The convergence actually makes mining infrastructure more valuable, because the power contracts and data centers can serve two markets instead of one.

Why do AI and Bitcoin mining need the same infrastructure?
Because both come down to running specialized chips at full load, which requires large amounts of cheap electricity and industrial cooling. An AI accelerator can draw around 1,000 watts and a rack of them over 100 kilowatts, so the binding constraint is power and a facility built to handle the heat, exactly what Bitcoin miners spent a decade building. The hardware differs, but the physical foundation is nearly identical.

Can I get into AI compute without technical expertise, like I can with mining hosting?
Increasingly, yes. Just as hosting lets you own mining hardware without running a facility, the AI side offers hardware supply and data-center capacity so you can deploy compute without building your own site. The right starting point is understanding what hardware fits your goal, which is what a benchmark tool is for, and then running it in a facility built for the power and cooling it needs.

Which is more profitable in 2026, mining Bitcoin or hosting AI?
It depends on the current Bitcoin price, network difficulty, AI compute demand, and above all your power rate, so there is no fixed answer, which is precisely why large operators keep the flexibility to switch between them. For an individual, the honest approach is to model each path against your specific costs rather than assume one always wins. Both are real businesses, and neither is guaranteed.

How do I start if I only have a modest budget?
Start by modeling, not buying. Use a profitability calculator to see what a hosted miner would earn at realistic power rates, or a free miner test to watch real hardware mine before committing. From there, hosting a single efficient machine is the most accessible entry into mining, while the AI path generally starts with understanding hardware options through a benchmark before scaling up. The tools cost nothing and save expensive mistakes.

Sources and image credit

Industry context is current as of mid-2026, drawn from public reporting on Bitcoin miners' expansion into AI and high-performance computing and from manufacturer hardware specifications. Power figures for AI accelerators are approximate and vary by model and configuration. Profitability for both mining and AI depends on power cost, hardware, and market conditions and is never guaranteed. Hero image graded to brand navy; credit to be added on publish.

Whichever side of the convergence you're on, the winners buy for the job, not the spec sheet. Score real inference and see what actually fits your workload on the GPU and AI benchmark, and browse the chips themselves in the AI hardware catalog.

The Most Profitable Bitcoin Miners Right Now, and Why 2026 Is the Time to Buy

MillionMiner — Mon, 20 Jul 2026 09:05:17 +0000

There is a strange thing about mining hardware: the best time to buy it feels like the worst. Right now, with Bitcoin well off its late-2025 high and plenty of miners switching off, the most efficient machines ever made are selling at a discount, and the ones still running are earning a bigger share of the network than they were six months ago. That combination does not last. Below is the honest ranking of the most profitable Bitcoin miners you can buy in 2026, what each is actually for, and the real reasons the timing is on your side, along with the one number that decides whether any of it works for you.

What "most profitable" actually means

Before the list, one thing has to be clear, because it is the difference between making money and losing it. The most profitable miner is not the one with the biggest hashrate. It is the one that turns the least electricity into the most Bitcoin, running on power you can actually afford.

That number is efficiency, measured in joules per terahash, and lower is better. In 2026 the line is simple: anything under about 15 J/TH is competitive, the very best machines are under 10, and anything above 20 is a space heater that loses money at most power rates. Every machine below is chosen on that basis. But efficiency is only half the equation. The other half is your electricity rate, and no machine on earth is profitable if you are paying too much per kilowatt-hour. Keep that in mind as you read, because the right machine on the wrong power is still a loss.

The most profitable machines you can buy right now

Here is the current field, ranked by what they are best at rather than a single number, because the right pick depends on your power, your cooling, and your budget.

Antminer S23 Hydro, the efficiency king. At roughly 9.5 joules per terahash, this is the most efficient production Bitcoin miner ever sold, the first to break the sub-10 barrier, delivering around 560 TH/s. It needs hydro cooling and three-phase power, so it is built for serious hosted or facility deployment, not a spare room. If your priority is the best possible margin per watt and you have the infrastructure, nothing beats it.

Antminer S21 XP Hydro, the best pick for most operators. At 12 J/TH and 473 TH/s for around nine thousand dollars, this is where most institutional buyers land. It pairs elite efficiency with a proven hydro form factor and strong resale liquidity, balancing purchase price against operating margin better than anything else in its class. If you are scaling and can run a water loop, this is the safe, high-performance choice.

Antminer S21 XP (air-cooled), the best first miner. At 13.5 J/TH and 270 TH/s, this is the most efficient air-cooled machine you can buy, and the natural entry point for a first-time buyer. It runs on single-phase 220V, it is the easiest current-generation miner to host, and it has the deepest support and parts network of any model. When price per machine matters more than squeezing out the last joule, this is the one.

Antminer S23 (air-cooled), the container workhorse. At around 14.5 J/TH and 270 TH/s, the air-cooled S23 slots into standard container deployments without any specialized coolant infrastructure, while still delivering a real efficiency gain over the previous generation. For operators expanding standard air-cooled fleets today, it is the natural primary choice.

Antminer S21 Pro, the value buy. At 234 TH/s and 15 J/TH, the S21 Pro has shifted into a value role now that the S23 has arrived, and secondary-market prices have come down accordingly. For a tighter budget that still wants current-generation, air-cooled performance, it remains cash-flow positive at typical hosting rates, and it is a sensible way to get real hashrate without paying flagship prices.

Whatsminer M63S Hydro, the non-Bitmain option. At roughly 390 TH/s, MicroBT's hydro unit brings high hashrate and a reputation for rugged reliability, giving buyers a credible alternative to Bitmain. Its efficiency trails the S23 family, so it makes the most sense where price, supply, or a preference for MicroBT tips the balance.

If you are mining from home for the interest rather than the margin, that is a different goal, and small, quiet machines like the Bitaxe or the Avalon Nano exist for exactly that, near-silent, low-power, and honest about being lottery tickets rather than income. But for anyone buying to actually earn, the machines above are the field.

Why now, honestly

Now the timing, and this is the part worth slowing down for, because the case is real but it is not a promise. Four things line up in a buyer's favor right now.

First, the hardware is cheap. When Bitcoin's price falls and sentiment turns fearful, demand for miners drops with it, and prices on both new and secondary machines come down. You are buying the same efficient hardware you would have paid a premium for at the top, at a discount, simply because most people only want to buy when it feels good.

Second, the network got easier. As the price fell, weaker and older machines became unprofitable and switched off, and when miners leave, the network's difficulty adjusts downward to compensate. That means every machine still running, including any you buy today, earns a larger share of the same daily Bitcoin than it did before the others left. Fewer competitors, more for the boats still fishing.

Third, efficiency is doing the heavy lifting. Two years after the last halving cut the block reward in half, margins are thin for old hardware and healthy for new, and the gap is entirely about efficiency. A current sub-15 J/TH machine can stay profitable at electricity rates that put a three-year-old miner deep in the red. Buying an efficient machine now is not just adding hashrate, it is buying the thing that actually keeps you profitable through the next stretch.

Fourth, you are positioning before the next halving, not after. The block reward is scheduled to halve again around 2028, and the miners who do best through a halving are the ones who bought efficient hardware early and cheaply, not the ones scrambling once the reward has already dropped. A new-generation machine bought today will stay competitive through far more difficulty increases than an older one, which means more of its life is spent earning.

Here is the honest caveat that belongs with all of that. None of it guarantees a profit. Bitcoin's price can keep falling, difficulty can climb again, and your own electricity rate can turn a good machine into a bad investment. These are reasons the timing is favorable, not a promise that it will pay, and anyone who tells you a miner is guaranteed income is lying to you. Which brings us to the one thing you actually have to get right.

The one number that decides it: your power rate

You have read it twice already because it is that important. The single factor that most determines whether any of these machines makes you money is what you pay for electricity. An S23 Hydro on cheap industrial power is a strong business. The same machine on expensive residential power can lose money every day it runs.

This is why so many buyers never plug these machines in at home at all. If your household electricity is too expensive to profit, and for most people it is, the answer is not a different machine, it is different power. Running your miner in a hosting facility built for it, on industrial-rate electricity with cooling and around-the-clock staff, is often the difference between earning and not, and it is what makes the efficient hydro machines at the top of this list accessible to people who could never run them at home. Before you buy anything, model the machine you want against your real, specific electricity rate, and if that rate does not work, price out hosting before you give up on the idea.

How to choose and buy

If you want to go deeper before committing, the full comparison in the best Bitcoin miners of 2026 and the wider buyer's guide to mining machines break down every model, and the explainer on what J/TH efficiency really means makes the one spec that matters click. To pressure-test the economics, run your numbers and your power rate through a mining profitability calculator. The deeper case for the timing is laid out in why now is the best time to buy a miner and mining through the bear market, and if you are thinking further ahead, how to prepare for the 2028 halving. When you are ready to buy, the machines here are in the Bitcoin miner catalog, and if home power is the problem, hosting is how most people solve it.

The bottom line

The most profitable miners in 2026 are the efficient ones, led by the sub-10 J/TH Antminer S23 Hydro, the 12 J/TH S21 XP Hydro for most serious operators, and the air-cooled S21 XP as the best first machine. And the reason to buy now is not hype, it is that the downturn has put efficient hardware on sale at the same moment the network got easier to mine and the next halving still sits ahead of you. That window closes when confidence returns and prices climb back.

Just remember the rule that governs all of it: buy the efficient machine, run it on the cheapest power you can find, and model your own numbers before you spend a cent. Do that, and the timing genuinely is in your favor. Skip it, and no machine on this list will save you.

Frequently asked questions

What is the most profitable Bitcoin miner in 2026?
For pure efficiency, the Antminer S23 Hydro leads at roughly 9.5 J/TH, the first production miner under 10 joules per terahash, which gives it the best margin per watt for hosted or facility deployment. For most operators, the S21 XP Hydro at 12 J/TH offers the best balance of efficiency, proven design, and resale value. The most profitable machine for you specifically is whichever efficient model you can run on the cheapest power, because your electricity rate matters as much as the hardware.

Is now actually a good time to buy a Bitcoin miner?
The timing is favorable for four concrete reasons: hardware prices fall during downturns, the network's difficulty has eased as weaker miners switched off (so each machine earns more), efficient new hardware carries much healthier margins than old machines, and buying before the 2028 halving positions you better than buying after. None of that guarantees a profit, since it still depends on Bitcoin's price and your power rate, but the conditions favor buyers more than they did at the top of the market.

How much can a Bitcoin miner earn per day?
It depends entirely on the machine's efficiency, the current Bitcoin price and difficulty, and your electricity rate, so any single figure is only an illustration. As a rough example, an efficient air-cooled machine on cheap power (around $0.06 to $0.07 per kWh) might net a few dollars per day at current conditions, while the same machine on expensive residential power can net nothing or lose money. Always model your specific power rate rather than trusting a headline number.

Should I buy an air-cooled or hydro miner?
Hydro miners like the S23 Hydro and S21 XP Hydro are more efficient and better for large hosted or facility deployments, but they need water cooling and three-phase power. Air-cooled machines like the S21 XP and S23 are easier to deploy, run on standard power, and are the right choice for first-time buyers, home setups, and standard container fleets. Match the cooling to where the machine will actually run.

Can I still mine Bitcoin profitably at home?
Sometimes, but only with an efficient machine and a genuinely low electricity rate, typically under about $0.07 per kWh, and ideally lower. At standard residential rates most machines lose money, which is why many home-minded buyers use hosting instead. If you want to mine at home for the interest rather than the income, small quiet machines exist for that, but treat them as a hobby rather than a business.

Sources and image credit

Machine specifications and efficiency figures are current as of mid-2026, drawn from manufacturer specifications (Bitmain, MicroBT, Canaan) and multiple independent hardware analyses; hashrate, J/TH, and pricing vary by batch, retailer, and region. Bitcoin price, network difficulty, and therefore all profitability figures move constantly and are illustrative snapshots, not guarantees. Hero image graded to brand navy; credit to be added on publish.

Buy the efficient machine, run it on the cheapest power you can find, and model your own numbers first. The current lineup is in the Bitcoin miner catalog, and you can pressure-test any machine against your power rate on the mining profitability calculator.

Specs, prices, and all profitability figures are mid-2026 snapshots and move fast; re-verify before acting. Mining profit is never guaranteed.

Can a $2,000 Gaming GPU Replace a $30,000 AI Chip?

MillionMiner — Thu, 16 Jul 2026 10:24:29 +0000

Here is a question that makes enterprise sales reps wince. NVIDIA's H100, the chip the AI boom was built on, costs between twenty-five and forty thousand dollars. NVIDIA's RTX 5090, a gaming card you can buy at a computer store, costs around two thousand. That is a fifteen-fold price gap. So the obvious question is: for actually running AI, how much of that gap is real?

The honest answer is the interesting part. For the work most people and most small teams actually do, the gap in results is small, and on value the cheap card wins outright. For the work large companies do, the H100 earns every dollar. The expensive mistake, the one people make constantly, is buying the wrong side of that line: a thirty-thousand-dollar chip for a job the gaming card crushes on value, or the gaming card for a job it physically cannot do. Here is how to tell which side you are on.

The specs that actually matter

Forget the gaming benchmarks. For AI, five numbers decide everything, and here is where the two chips land.

The RTX 5090 has 32 GB of GDDR7 memory at roughly 1.79 terabytes per second of bandwidth, 21,760 CUDA cores, fifth-generation Tensor Cores with native FP4, and a 575-watt power draw, for about two thousand dollars at list. The H100 has 80 GB of HBM3 at about 3.35 terabytes per second, its Hopper Transformer Engine, NVLink for high-speed multi-GPU, ECC memory, and a 700-watt draw, for twenty-five to forty thousand.

Three of those differences matter more than the rest. The H100 has two and a half times the memory, which decides what models fit at all. It has nearly twice the bandwidth, which decides how fast tokens come out. And it has NVLink and ECC, which decide whether you can scale to many GPUs and trust a multi-day training run. Hold those three in mind, because they draw the entire line between these cards.

Where the 5090 wins, and it is not close

Start with the good news, because it is better than most people expect. For any AI workload that fits inside 32 GB of memory, the RTX 5090 is the best performance per dollar you can buy, full stop.

That covers a lot more than it sounds. A 5090 comfortably runs inference on models up to around 13 billion parameters at full precision, and considerably larger ones when quantized, a 32-billion-parameter model at four-bit sits inside 32 GB with room to work. Its bandwidth is a genuine surprise, reaching a bit over half of an H100's despite costing a fifteenth as much, and because token generation is limited by memory bandwidth, that ratio is what actually shows up in speed. For models under about 32 billion parameters, the cost per token is wildly in the 5090's favor.

It gets more lopsided for fine-tuning. Adapting a model like Llama 8B or Mistral 7B with LoRA or QLoRA is exactly what the 5090 was born for, and the economics are almost comical: a fine-tuning run that costs roughly two to three dollars of electricity on a 5090 you own can cost forty to fifty dollars to rent the equivalent cloud time on data-center hardware. Run experiments daily and the card pays for itself inside a year, before you count the value of your data never leaving the building. For a solo developer, a researcher, or a small team whose work fits in 32 GB, there is no real argument. Buy the gaming card.

Where the H100 earns its price

Now the other side, because it is just as real. The moment your workload crosses a few specific lines, the H100 stops being expensive and starts being necessary.

The first line is memory. If your model does not fit in 32 GB, no amount of value changes the fact that it will not run. A 70-billion-parameter model needs roughly 140 GB at full precision, and even squeezed down to four-bit it lands around 35 to 40 GB, which is already past the 5090's ceiling. Large models, long context windows, and big batches all live in the H100's 80 GB and cannot fit in the 5090's 32.

The second line is reliability at scale. The H100 has ECC memory, which quietly corrects the bit errors that occur during long computations. On a consumer card without it, a single flipped bit during a forty-eight-hour training run can silently corrupt the result, and you find out at the end. For production serving that has to stay up, and for training runs measured in days, that protection is not a luxury.

The third line is scale itself. This is NVLink, and it is the H100's decisive advantage. It connects GPUs at 900 gigabytes per second, roughly seven times faster than the PCIe link the 5090 is limited to. When you train a model across many GPUs, they constantly exchange gradients, and over the slow PCIe path they spend more time waiting than computing. The 5090 has no NVLink at all. For serious multi-GPU training, the H100 is not just better, it is the only practical choice.

The question everyone actually asks: two 5090s or one H100?

Because the price math is so tempting, this comes up constantly. Two RTX 5090s give you 64 GB of combined memory for about four thousand dollars, and they can indeed fine-tune a 70-billion-parameter model at four-bit. So why pay eight times as much for one H100?

Because they are not doing the same job to the same standard. The single H100 fine-tunes that same model at higher precision, with ECC catching errors, and with NVLink syncing the work at full speed, none of which the pair of 5090s can match. The rule that falls out is clean. If you are a researcher iterating fast on models that fit, the 5090 path gives you far more experiments per dollar, and that is the right buy. If you are running a production pipeline where a corrupted checkpoint means restarting a multi-day job, the H100 pays for itself the first time it saves the run. Match the card to the stakes, not just to the memory number.

The catch nobody puts in the spec sheet

There is one more thing that quietly settles a lot of arguments: the 5090 is a consumer card, and NVIDIA means it. It ships with gaming drivers, not certified professional ones. It has no ECC. It is not rated for continuous data-center operation, and running four of them at full load for days is a genuine thermal project. And NVIDIA's licensing restricts using GeForce cards in data centers at all, which is why the intended cloud and server products are the H100, H200, and B200, not the 5090.

So the 5090 is a spectacular local and workstation card, and a poor foundation for a serving farm. If your plan is a handful of cards under your own desk doing real work, it is close to unbeatable. If your plan is racks of GPUs serving thousands of users around the clock, that is what the data-center parts exist for, and trying to build it out of gaming cards will cost you more in headaches than you saved on hardware.

So which one do you actually need?

Two questions settle it. First, does your model fit in 32 GB at the precision you need? Second, do you need production reliability, multi-GPU scale, or ECC? If it fits and you do not, buy the RTX 5090 and do not look back, you would be paying for an H100's capabilities you will never touch. If it does not fit, or you need to serve at scale with guarantees, you need H100-class hardware.

And here is the honest final note, the one that matters most for the expensive card. If you do need an H100, the worst thing you can do is buy one and let it sit half-used in a closet, because at thirty thousand dollars its economics live or die on how hard it runs and what it pays for power. A card like that belongs somewhere it runs at high utilization on cheap electricity, whether that is the cloud or a hosting facility built for it, not idling on a desk. For the gaming card, you own it and run it. For the data-center card, utilization and power are the whole game, which, if you have read anything else on this site, is a lesson that should sound familiar.

How to decide for your actual workload

Spec arguments are fun, but the only comparison that matters is real performance on the model you are running, at the memory you have. The MillionMiner GPU and AI benchmark scores the RTX 5090, the H100, and everything around them on real inference and, just as importantly, shows you what fits in what memory, so you can see the line between them for your own model before you spend anything. If you are choosing a card to run models on your own machine, the guide to the best GPU for running LLMs locally goes deeper on the consumer tier, and the wider guide to choosing a GPU for AI walks through it by budget and use case. When you are ready to buy either one, they are both in the AI hardware catalog. And if your question is really about the data-center tier, the breakdown of the H100, H200, and B200 compared covers what sits above all of this.

The bottom line

A two-thousand-dollar gaming card can absolutely replace a thirty-thousand-dollar AI chip, for the enormous range of work that fits in 32 GB, and for that work it is not close: the 5090 wins on value by a mile. What it cannot replace is 80 GB of memory, NVLink, ECC, and a card built to run flat out for years, and the moment your work needs those, the H100 is worth every cent.

The trap is not picking the "worse" card. The trap is picking the wrong one for your job, overpaying for capabilities you will never use, or under-buying and hitting a wall you cannot quantize your way around. Figure out whether your model fits and whether your stakes are high, and the answer picks itself.

Frequently asked questions

Can the RTX 5090 run a 70B model?
Only with heavy compromise. A 70-billion-parameter model needs around 35 to 40 GB even at four-bit quantization, which is past the 5090's 32 GB. You can push it with very aggressive quantization or by splitting it across two 5090s (64 GB combined), but for comfortable 70B work at good quality, an 80 GB H100 or a higher-memory card is the better tool.

Why is the H100 faster for LLM inference if the 5090 has similar compute?
Because inference speed is limited by memory bandwidth, not raw compute. The H100's HBM3 delivers about 3.35 TB/s against the 5090's 1.79 TB/s, so it reads model weights out of memory faster and generates tokens faster, especially at large batch sizes. For small models and single-user workloads, the gap narrows and the 5090's price advantage dominates.

Is it cheaper to buy an RTX 5090 or rent cloud GPUs?
If your work fits in 32 GB and you use the card most days, buying a 5090 usually wins, often paying for itself within a year, with the bonus that your data never leaves your machine. If you need more than 32 GB, multi-GPU scale, or only occasional bursts of compute, renting is usually better than buying an expensive data-center card that would sit idle.

Does the RTX 5090 support NVLink for multi-GPU AI?
No. The RTX 5090 has no NVLink, so multiple cards communicate only over PCIe, which is far slower. This makes it impractical for the distributed training that large models require. If you need to scale training across GPUs efficiently, you need NVLink-equipped data-center cards like the H100, H200, or B200.

Can I put RTX 5090s in a data center to serve AI at scale?
It is not the right tool, and NVIDIA's licensing discourages it. The 5090 uses consumer drivers, lacks ECC, and is not rated for continuous data-center operation, and GeForce cards are restricted from data-center use under NVIDIA's terms. For serving at scale, the H100, H200, and B200 are the products built and licensed for the job.

Sources and image credit

Specifications and pricing current as of mid-2026, from NVIDIA documentation and multiple independent benchmark and cloud-provider analyses; memory, bandwidth, and cost figures are widely corroborated. Real-world performance varies by model, quantization, batch size, and software stack. Hero image graded to brand navy; credit to be added on publish.

The only comparison that matters is real throughput on your model, at your memory. See the RTX 5090, H100 and the rest scored on real inference (with a VRAM-fit table) at millionminer.com/gpu-ai-benchmarks, and browse either card in the AI hardware catalog.

So You Want to Mine Bitcoin in 2026? Read This First

MillionMiner — Wed, 15 Jul 2026 11:16:57 +0000

An honest, no-hype roadmap through every decision that actually matters, and the two numbers that decide whether any of it is worth your money.

Mining Bitcoin in 2026 is not what it was in 2021. The easy money is gone, the machines are more powerful and more expensive, and the price of electricity now separates the people who make money from the people who quietly lose it. None of that means you should not do it. It means you should do it with your eyes open.

This is the map. Not a sales pitch, and not a promise that you will get rich, because nobody can honestly make you that promise. It is the real sequence of decisions a person faces when they go from "I want to mine Bitcoin" to actually running a machine that earns, and at each fork it points you to the deeper guide that answers that specific question. Read it top to bottom once, and you will know more than most people who have already spent their money.

First, be clear on what you are actually doing

Before anything else, it helps to understand what mining really is, because most people picture it wrong. You are not solving math problems for coins, and you are not "making money from nothing." You are running specialized hardware that competes with everyone else on earth to secure the next block of transactions, and in return you earn newly issued bitcoin plus fees. If that sentence is fuzzy, start with the plain-English explanation of what Bitcoin mining is, and if you want to feel it rather than just read it, you can even watch mining happen and test it yourself before you commit a cent.

Understand that one thing and the rest of this page stops being jargon and starts being a series of business decisions. Because that is what mining is: a small, real business, with revenue, costs, and a machine that either earns more than it eats or it does not.

The only two numbers that decide everything

Here is the part the hype never mentions, and the part that decides your entire outcome. In mining, two numbers matter more than every other number combined: your electricity rate, and your machine's efficiency. Everything else is noise.

Efficiency is how much power a machine burns to do its work, measured in joules per terahash, and it is the single most important spec on any miner. A modern machine sips power for the work it does; an old one drinks it. The line to understand is roughly this: anything around 15 joules per terahash or better is competitive in 2026, and anything above 20 is usually a space heater that loses money. The full explanation of what J/TH efficiency means is worth ten minutes, because it is the number you will live and die by. Pair it with a realistic sense of how much electricity a miner actually uses, and you will already be ahead of most buyers.

Why do these two numbers rule everything? Because mining revenue is roughly fixed by the network, so your profit is almost entirely a story of cost. Two people can run the identical machine and one gets rich while the other goes broke, purely because of what they pay per kilowatt-hour. Cheap power and an efficient machine, and you have a business. Expensive power and a thirsty machine, and you have an expensive hobby. Keep those two numbers in your head for the rest of this page.

Will it actually pay? Run the math before you spend a cent

This is the step almost everyone skips, and it is the one that saves people from expensive mistakes. Before you buy anything, work out whether mining will be profitable for you specifically, at your power rate, with the machine you are considering.

Start with the honest, current answer to whether Bitcoin mining is worth it in 2026, then get concrete with the breakdown of what it actually costs to mine one Bitcoin right now, which lands somewhere between roughly thirty and ninety thousand dollars depending almost entirely on your electricity. To put your own numbers in, run them through a mining profitability calculator, and if you plan to plug a machine into a normal household outlet, read the hard truth about whether mining pays at home electricity rates first. If the math does not work at your power rate, that is not a reason to give up. It is a reason to change where the machine runs, which is a decision we will get to.

Buy a machine, or build a rig?

Once the math looks survivable, the first real fork appears. Do you buy a purpose-built machine, or build something yourself?

For Bitcoin specifically, the answer is almost always to buy a purpose-built ASIC, a machine that does one job with brutal efficiency, rather than assembling a rig from graphics cards. GPU rigs made sense for coins like Ethereum years ago, but for Bitcoin they cannot compete on efficiency, which, as you now know, is the whole game. If you are weighing it up, the full comparison of buying a machine versus building your own and the head-to-head on mining rigs versus ASICs will settle it. For the vast majority of people getting into Bitcoin in 2026, you are buying a finished machine, not building one.

Which miner should you actually buy?

Now the fun part, and the part where people overspend. The instinct is to buy the machine with the biggest hashrate number. That instinct is wrong. The machine you want is the one with the best efficiency you can afford, running where power is cheap, not the one with the loudest headline spec.

The current field of the best Bitcoin miners in 2026 and the wider buyer's guide to mining machines walk through the efficiency tiers, the sound, the power draw, and the price per terahash that actually matter. When you are ready to see real machines and prices, the Bitcoin miner catalog is where to look. Buy for efficiency and for what it costs to run, not for the number on the box.

Run it at home, or host it somewhere else?

This is the biggest fork on the whole map, and for most people it is the decision that makes or breaks the business. You have a machine. Where does it actually live?

You can run it at home. It works, and for one quiet machine on cheap power it can even make sense. But be honest about what that means: retail electricity, which is often too expensive to profit, a machine that genuinely sounds like a vacuum cleaner that never turns off, real heat, and you as the sole technician at two in the morning. Or you can send the machine to a hosting facility, a place built for exactly this, buying power at industrial rates with cooling, security, and staff around the clock, and simply collect what it earns.

This is where the two magic numbers usually force the decision. If your home power rate makes mining unprofitable, hosting is not a luxury, it is the difference between earning and not. Read the honest comparison of home mining versus hosted mining, see how the numbers work in the ASIC hosting cost breakdown, and if hosting is the path, MillionMiner's hosting runs machines in US facilities at industrial power rates while you keep the coins. For most people mining seriously in 2026, the machine does not live in the spare room. It lives somewhere the power is cheap and the crew never sleeps.

If you host, where? And is it even legal?

If you are hosting, or building your own small site, geography suddenly matters, because power price and the law both change dramatically from one place to another. In the United States, some states are genuinely good for mining and some are hostile to it, and the gap is enormous. The rundown of the best states for Bitcoin mining in 2026 covers where the cheap, mining-friendly power actually is, and the plain answer on whether Bitcoin mining is legal in the US clears up the fear before it costs you a bad decision. Short version: it is legal, but where you do it changes everything about whether it works.

A warning about cloud mining

Somewhere in your research you will run into "cloud mining," which promises all the profit with none of the hardware or hassle. Tread very carefully here, because the space is full of outright scams, and the honest ones rarely beat simply owning a machine.

Before you hand anyone money for a cloud contract, read why so much of cloud mining is a scam and the clear-eyed comparison of cloud mining versus real hosting. If the idea of owning no hardware appeals to you, there is a legitimate version of that through transparent cloud plans backed by real machines, but the rule holds: if a return looks guaranteed and effortless, it is almost always neither.

You are not fishing alone: join a pool

Once your machine is running, you will not mine on your own. Because the network only produces one block roughly every ten minutes across the entire planet, a single machine mining solo could wait years to ever land one. So miners pool their power together and share the rewards in proportion to what each contributes, turning a rare, enormous, unlikely payday into a small, steady, daily one. Nearly everyone uses a pool, and choosing the right one matters a little more than people think. The comparison of the best Bitcoin mining pools in 2026 covers the payout models and which pool fits which situation.

When should you actually buy? The timing question

Here is a decision most guides ignore entirely, and it may be the most important one of all: when you buy matters as much as what you buy. Mining hardware is cheapest, and least competitive, exactly when everyone is fearful and the Bitcoin price is low, which feels like the worst possible time and is often the best. When the price crashes, weaker miners switch off, the network gets easier, and the machines still running quietly earn a bigger share.

This is worth genuinely understanding before you buy at the top out of excitement. The case for the best time to buy a miner and the deeper look at mining through a bear market explain why the quiet, ugly part of the cycle is when patient people position. And because the next supply cut is already on the horizon, it is worth reading how to prepare for the 2028 halving, the event that will halve every miner's block reward and reshape who stays profitable.

Actually setting it up

Say you have made your decisions: an efficient machine, bought at a sensible time, running somewhere with cheap power, pointed at a good pool. The last step is the practical one of getting it configured and earning. The walkthrough on how to set up an ASIC miner covers the first boot, the network setup, and pointing it at your pool, and if you have not bought yet, how to buy ASIC miners safely helps you avoid the grey-market traps that catch first-time buyers.

One honest word on who to trust

The mining world has more than its share of hype, fake reviews, and vendors who vanish after the wire clears. So the last piece of advice is simply this: buy from people who tell you the truth, including the parts you do not want to hear, like the fact that your power rate might make mining a bad idea. If you want to know where we stand on that, the honest breakdown of whether MillionMiner is legit lays it out. Whoever you go with, hold them to the same standard.

The whole map, in five lines

If you remember nothing else, remember this. Mining is a small business, not free money. Two numbers decide your fate: your electricity rate and your machine's efficiency. Run the real math before you spend anything. If home power is too expensive, host the machine somewhere it is not. And buy when the market is fearful, not when it is euphoric.

Do those five things and you will already be doing what most people who lose money at mining failed to do, which is treat it seriously. The machines are just tools. The decisions are the business. Now you have the map, so start with the fork that matters most to you, and take it one honest step at a time.

A note on the figures

This roadmap is current as of 2026. Bitcoin's price, network difficulty, hardware efficiency, and electricity markets all move quickly, so treat every specific number, especially the cost to mine one coin and any profitability estimate, as a snapshot to re-check on the day you decide, not a fixed promise. Mining profitability is never guaranteed; it depends mostly on your power cost and machine efficiency, and it can and does go negative for miners who ignore those two numbers.

Read the full roadmap
This is the short version. The complete roadmap links a step-by-step guide for every fork above, buy vs build, which miner, home vs hosted, where to host, pools, timing, and setup, on the original page. Two places to start right now: work out your numbers, because your power rate and your machine's efficiency decide everything, and if home power is too expensive, hosting is usually the fix.

●Model your numbers: mining profitability
●Run it where power is cheap: ASIC hosting

Bitcoin Mining Makes No Sense Until You Picture a Fishing Fleet

MillionMiner — Tue, 14 Jul 2026 10:20:14 +0000

A complete, slightly salty guide to how mining actually works, told through boats, fuel, fish, and one very strict harbor warden. No jargon. Just the sea.

Bitcoin mining is explained badly almost everywhere. You get told about hashes and nonces and cryptographic puzzles, your eyes glaze over, and you walk away thinking it is either magic or a scam, possibly both. It is neither. It is a fishing fleet.

That is not a cute simplification to throw away after the first paragraph. It is the whole thing. Every confusing part of mining, difficulty, halvings, why your electricity bill decides your fate, why nobody can cheat, what happens when the coins run out, maps cleanly onto boats going out to sea. Picture the fleet once and you will never be lost in a mining headline again. Let us go to the water.

The sea, and the fish in it

Start with the sea. There is one ocean, shared by everyone, and in it swims a fixed and known number of fish: twenty-one million, and not one more. Everyone can see the whole ocean at once. There is a harbor logbook, and every boat keeps its own identical copy, so no one can lie about who caught what. When a fish is landed, every logbook updates at the same moment, and they must all agree or the catch does not count.

That shared, verified logbook is the blockchain. The fish are bitcoin. The fact that there will only ever be twenty-one million of them, and that anyone can count exactly how many are left, is the entire reason the fish are worth anything at all. This is a strange ocean: fully transparent, impossible to fake, and running low on fish by design.

The boats

A miner is a fishing boat. That is the whole secret, and everything else is detail.

Your boat goes out and puts nets in the water. A bigger, better boat puts more nets in the water, and more nets means more chances to land the next catch. In mining, "nets in the water" is hashrate, the number of guesses your machine makes every second. You are not smarter than the other boats, and you are not luckier. You simply hold some share of all the nets in the sea, and that share is your share of the fish. Put ten percent of the fleet's nets in the water and, over time, you land about ten percent of the fish. That is all hashrate is. Stop picturing a puzzle. Picture nets.

The one catch, every ten minutes

Here is the first rule that makes the whole system tick. The fishery allows exactly one catch to be landed every ten minutes. Not more, not less, on average, no matter how many boats are out there. When a catch is landed, it goes to whichever boat's nets happened to pull it up, and right now that catch is worth 3.125 fish.

Ten minutes. One catch. Split, over time, by who had the most nets out. Whether there are ten boats on the water or ten million, the fishery lands one catch every ten minutes and no faster. Hold onto that, because the next part is where most people finally understand Bitcoin for the first time in their lives.

The warden who keeps the catch on schedule

If the fishery must produce one catch every ten minutes, but boats are free to come and go as they please, something has to give. That something is the mesh of the nets, and the one in charge of it is the warden.

The warden has exactly one job: keep the catch landing every ten minutes. So the warden counts the nets. When a thousand new boats show up and the sea gets crowded, catches start landing too fast, so the warden tightens every net in the ocean, finer mesh, smaller window, harder to land a fish. When boats leave and the sea empties out, catches come too slowly, so the warden loosens the nets and fishing gets easier again. In Bitcoin this warden is not a person. It is a rule written into the code that re-measures the whole fleet every two weeks and adjusts the difficulty up or down so the ten-minute rhythm never breaks.

Now sit with what that means, because it is the most important and least understood fact in all of mining. The fish are shared out among the boats, but the total number of fish is fixed by the clock, not by effort. So when more boats crowd in, everyone catches less. And when boats leave, everyone who stayed catches more. You do not win by fishing harder. The sea has a fixed catch and a warden making sure of it. Your only real questions are how big your slice of the fleet is, and how cheaply you can keep your boat running while you hold it.

Fuel is the only thing that has ever sunk a boat

Boats run on fuel. Miners run on electricity. And here is a truth the industry took years and many bankruptcies to fully respect: fuel is the only thing that has ever sunk a fishing boat, and electricity is the only thing that has ever killed a miner.

Two boats can put the same nets in the water and land the same fish, and one goes broke while the other gets rich, entirely because of what they pay for fuel and how much fuel they burn per fish. A modern boat sips fuel. An old boat drinks it. When fish are expensive, even the thirsty old boats make money and everyone feels like a genius. When the price of fish falls, the thirsty boats burn more in fuel than their catch is worth, and they have to tie up at the dock. The efficient boats keep fishing straight through it.

This is why, in mining, the two numbers that decide everything are your electricity rate and your machine's efficiency, the fuel per fish. Not the size of the boat. Not the brand on the hull. Fuel. You are not really buying a machine. You are buying a boat, and betting you can fuel it cheaper than the boat next to you.

The co-op, so you don't starve waiting

There is a problem with fishing alone. Because only one catch lands every ten minutes across the entire ocean, a small boat fishing by itself might wait months or years to ever personally pull one up. It might never happen. That is a brutal way to live: feast or famine, mostly famine.

So the small boats formed co-ops. Thousands of them agree to pool all their nets together, and whenever any boat in the co-op lands a catch, the whole thing is shared out among the members in proportion to the nets each one contributed. You stop waiting for a rare, enormous, lonely payday and start taking a small, steady share every single day. In mining these co-ops are called pools, and almost everyone uses one. You give up the fantasy of landing a whole catch by yourself, and in return you get paid like it is a job instead of a lottery.

The lone dinghy and the commercial fleet

Not all boats are equal, and not all docks are equal either.

You can run a single boat off the little dock behind your house. It works, but you pay retail for fuel, you are the mechanic and the deckhand and the night watch all at once, and the boat screams in your backyard while you try to sleep. Or you can run a whole commercial fleet from a deep-water port, buying fuel by the tanker, with crews and mechanics on hand around the clock. The fleet lands the same fish per net as your dinghy, but it fuels each boat far more cheaply and never sleeps through a breakdown.

This is the difference between mining at home and mining at scale, and it is why a lot of people stop running a boat off their back porch and instead dock it at somebody else's port, one built where the fuel is cheap and the crew never goes home. Same boat, same fish, a fraction of the fuel bill and none of the noise. It is the single biggest lever an ordinary miner has, and it is worth knowing your own boat's fuel math cold before you decide where to tie up.

The halving: half the fish, and it never gives them back

Every four years, something happens that would end most industries overnight. The fishery permanently cuts the size of every catch in half.

It is not a bad season. It is forever. The catch that was 50 fish became 25, then 12.5, then 6.25, and in 2024 it became 3.125, and around 2028 it will become 1.5625. Same nets, same effort, same warden, half the fish, and it never comes back. At every halving, the boats that were only surviving on expensive fish and thirsty engines get wiped out, because suddenly their catch is worth half as much and their fuel bill did not budge. The boats that live through each one are the boats that fish where fuel is cheapest or run the most efficient hulls afloat, and usually both. The halving does not take your boat. It takes half your fish, and it dares you to still be profitable in the morning.

Storms and lean seasons

The price of fish is not steady. It booms and it crashes, in long cycles, and the crashes are where the fleet turns brutal.

When fish prices collapse, the thirsty boats cannot cover their fuel, and one by one they tie up at the dock and stop fishing. The fleet shrinks. And you already know exactly what the warden does when boats leave: it loosens the nets, so every boat still on the water starts landing more fish. So the deep, ugly bottom of a downturn is a strange moment. The fish are cheap, half the fleet has gone home, and the boats that kept fishing are quietly pulling up a bigger share of the catch than they have in years. It is uncomfortable, and it is not for everyone, but it is the season when patient fishermen have always bought boats, while everyone else is selling theirs for scrap.

Why nobody just forges the logbook

People always ask the obvious question. If it is all a shared logbook, why can't someone just write themselves a few thousand fish?

Because of how the logbook stays honest. For a lie to stick, every boat's copy has to agree, and the fleet only accepts the version of history backed by the most fishing effort. To rewrite the logbook in your favor, you would need to out-fish everyone else combined, to own more nets than the entire rest of the fleet put together, and keep them all fueled. That costs a staggering fortune, far more than you could ever steal, and the instant anyone noticed, the fish you were trying to steal would be worthless anyway. So nobody does it. Not because fishermen are saints, but because honesty is simply cheaper than cheating. The whole system is held together by the plain fact that going fishing pays better than faking it.

When the last fish is caught

If there are only twenty-one million fish, and the catch halves forever, the fish must eventually run out. Around the year 2140, the last new fish will be landed.

You might think that is the end of the fleet. It is not. The boats do more than catch new fish. They also run the harbor. Every trade, every transfer of fish from one person to another, has to be logged and verified by the fleet, and people pay a small tip to have their transaction written into the next catch. As the new fish dwindle, those tips become the main reason to keep a boat on the water. The fleet stops living off new fish and starts living off the fees for keeping everyone's records straight. The harbor keeps running. The boats just get paid a different way.

Where the metaphor springs a leak

No metaphor is seaworthy forever, and it is only fair to show you where this one takes on water.

There are no actual fish, for a start. The "catch" is really a cryptographic proof that your boat did an enormous amount of genuine work, and that proof is what secures everyone's records. The warden is not a wise old sailor but a cold, automatic rule that no single person controls. And "landing a catch" is not skill or patience: each ten-minute round is a fresh roll of the dice, weighted by how many nets you hold, so a small boat can get lucky and a huge fleet can hit a dry spell, even though over months the shares come out exactly as the net counts predict. Hold the fishing fleet loosely enough to let those details in, and it will still explain more of Bitcoin, faster, than any textbook.

The payoff

That is the entire machine. A shared sea with a fixed number of fish. Boats that are really just nets in the water. A warden holding the catch to one every ten minutes, so crowding thins every boat's share and departures fatten it. Fuel that decides who lives and who ties up. Co-ops so the small boats still eat. A halving that takes half the fish forever. Storms that clear out the thirsty boats and quietly reward the patient. A logbook too expensive to forge. And a far horizon where the boats trade catching new fish for running the harbor.

Go and read any mining headline now. Difficulty at an all-time high. Miners capitulating. Hashrate recovering. The next halving approaching. You will not see jargon anymore. You will see boats, fuel, fish, and the warden. And if you ever decide to put a boat of your own in the water, the only two questions that have ever mattered are the same two that have always mattered at sea: how efficient is your hull, and how cheap is your fuel. That, in the end, is the whole business MillionMiner is built around, efficient boats and a port where the fuel is cheap and the crew never sleeps. Everything else is just weather.

A note on the mechanics

The mappings here are accurate, not just pretty. The network targets one block roughly every ten minutes; the difficulty re-adjusts every 2,016 blocks, about two weeks, to hold that rhythm as total hashrate rises and falls. The block reward is 3.125 BTC after the April 2024 halving and is scheduled to halve again around 2028, on the way to a fixed supply of 21 million coins. Security rests on the fact that rewriting history requires a majority of the network's hashing power, which costs far more than any attack could yield. As new issuance trends to zero over the next century, transaction fees are designed to replace it as the miners' reward. Figures are current as of 2026.

If you ever put a boat in the water, the only two numbers that matter are your hull's efficiency and your fuel. Work out your own on the mining profit calculator, and if home fuel is too dear, hosting is docking at a port built where the fuel is cheap.

H100 vs H200 vs B200: The Real Differences, and How to Choose in 2026

MillionMiner — Fri, 10 Jul 2026 10:41:58 +0000

Three NVIDIA GPUs, three completely different bottlenecks. Most of the money wasted on AI hardware comes from treating them as a simple "faster" ladder. They are not.

Most people read NVIDIA's data center lineup as a straight line: the H100 is fast, the H200 is faster, the B200 is fastest, so buy the newest thing the budget allows. That instinct is wrong, and it is one of the most expensive mistakes in AI infrastructure. These three chips do not clear the same bottleneck harder. They clear three different bottlenecks, and the right choice depends entirely on which one your workload hits first.

Here is what each one is, the numbers that matter, and a simple way to pick the right one instead of overpaying for a spec you will never use.

The H100: the workhorse that built the boom

The H100, on NVIDIA's Hopper architecture, is the chip the current AI wave was trained and served on. It carries 80 GB of HBM3 memory at 3.35 terabytes per second of bandwidth, and its fourth-generation Tensor Cores deliver about 989 teraflops of dense FP16 and roughly double that in FP8. It draws 700 watts, links to other H100s over NVLink at 900 gigabytes per second, and still sets the baseline every newer chip is measured against.

Its compute was never really the problem. Its ceiling is capacity. Eighty gigabytes fits a lot, but a 70-billion-parameter model in full precision needs around 140 GB, which means two H100s and the overhead of splitting a model across them. As models grew, that 80 GB wall, not the math, became the thing people kept hitting.

The H200: not a new chip, a bigger memory system

Here is the fact that reframes the entire lineup: the H200 uses the exact same compute die as the H100. Identical Tensor Cores, identical FP16 and FP8 throughput. On raw compute, they are the same chip.

What NVIDIA changed is the memory. The H200 carries 141 GB of faster HBM3e at 4.8 terabytes per second, which is 76 percent more capacity and 43 percent more bandwidth than the H100, in the same 700-watt envelope and the same socket. In most H100 servers it is close to a drop-in upgrade.

Why build a "new" GPU that only touches memory? Because the workload that dominates AI spending, inference, is limited by memory bandwidth, not compute. Generating each token means reading the model's weights out of memory, doing a little math, and repeating; most of the time is spent moving data, not calculating. The H100 already had enough compute for that job. It did not have enough memory speed. The H200 fixes the part that was the real bottleneck, and the result is roughly 40 percent faster inference on large models for zero extra power, a gain that tracks its 43 percent bandwidth increase almost exactly.

The B200: a real generational jump, and a different class of machine

The B200 is where "newer" finally means fundamentally different. It is built on NVIDIA's new Blackwell architecture, and it is not one die but two, 208 billion transistors joined into a single package by a 10 terabyte-per-second link. It carries 192 GB of HBM3e at 8 terabytes per second, about 2.4 times the H100's bandwidth, and its fifth-generation Tensor Cores add native FP4, a four-bit precision mode that roughly doubles inference throughput again for models that tolerate it.

The numbers are a leap, not a memory bump. Around 2,250 teraflops of dense FP16, 4,500 in FP8, and 9,000 in FP4, alongside NVLink 5 at 1.8 terabytes per second, double the Hopper interconnect. In practice a single B200 serves large models at roughly 2.3 to 2.5 times an H100's token rate, trains about twice as fast, and with FP4 can push inference throughput far higher still. One B200 fits a 70-billion-parameter model that used to need two or three H100s.

But it is a different class of machine, and that is the catch. The B200 draws 1,000 watts, uses a new SXM6 baseboard rather than a card that drops into an existing server, and cannot go into most H100-era, air-cooled infrastructure without new power delivery and cooling. It is also where the lineup stops being about single GPUs at all: the B200 is the building block of the GB200 rack, which packs 72 of them with 36 Grace CPUs into one liquid-cooled system that behaves like a single enormous GPU. Demand has run well ahead of supply, with Blackwell reportedly backlogged into mid-2026.

The one idea that makes the choice obvious

Forget the ladder. The right way to choose is to find the bottleneck your workload hits first, then buy the chip that clears it. Ask three questions, in this order.

First, does your model fit? Capacity is a hard gate. If the model plus its working memory does not fit on the card at your precision, nothing else matters. A 70-billion-parameter model needs roughly 140 GB in full precision, which rules out a single H100 before speed is even a question.

Second, is your workload memory-bandwidth-bound? Almost all inference is. If you are serving a model and it fits, bandwidth decides your token speed, which is exactly why the H200 beats the H100 at the same model size despite identical compute.

Third, do you need more compute or FP4? This is the last question, not the first. Heavy training, very large models, or high-traffic FP4 serving are what justify a B200. If your job is memory-bound, paying for Blackwell's compute buys you a number you will not use.

The most expensive mistake in this whole category is buying for peak compute when the workload was starved for memory the entire time. Match the chip to the bottleneck, not to the top line on the spec sheet.

What the numbers look like

On real inference, serving a Llama-class model in FP8, a single H100 generates on the order of 115 to 135 tokens per second, an H200 around 160 to 185, and a B200 roughly 300 to 360. The H200's lead over the H100 tracks its bandwidth advantage; the B200's lead tracks its own, plus a new architecture and FP4 on top.

Price and availability tell the rest. The H100 has fallen sharply as supply caught up, and is now the cheap, proven option. The H200 sits in the sweet spot for memory-bound inference: more capacity and speed for the same power. The B200 costs several times more per hour and is hard to get, but for the largest models the right measure is cost per result, not cost per hour, and there it can win decisively. For anything that fits comfortably on Hopper, it usually does not.

The part nobody puts on the spec sheet: power

There is a reason large GPU deployments stall, and it is almost never the chips. It is power and cooling. An H100 or H200 at 700 watts is demanding; a B200 at 1,000 watts is more so; and a full GB200 rack pulls well over 100 kilowatts and requires liquid cooling. Securing that much power, at a workable price, in a facility built to move that much heat, is the hard part of AI infrastructure, and the part that quietly decides whether a build is viable at all.

This is the same lesson that has always governed mining, where the machine is the easy purchase and the power and the site are the real constraint. Whether you run Hopper or Blackwell, the silicon is available to anyone with a purchase order. Industrial power and the cooling to match are not, which is exactly the problem MillionMiner's data-center and AI hosting side is built to solve.

How to compare them for your workload

Spec sheets flatter the newest chip. The only comparison that matters is real performance on the job you are running, which is why it helps to look at measured throughput rather than peak teraflops. The MillionMiner GPU and AI benchmark scores every card, including the H100, H200, and B200, on real inference performance rather than the number on the box, alongside memory capacity and compute, so you can see which one clears your bottleneck before spending anything. Browse the cards themselves in the AI hardware catalog, and if you are choosing within a budget or for a specific use case, the guide to choosing a GPU for AI walks through it.

It is the same trap we see on the mining side, where the highest hashrate is not the most profitable machine, and only real, measured output tells the truth. The profit calculator applies that logic to mining hardware the way the benchmark applies it to GPUs: measure the result, not the headline number.

The bottom line

The H100, H200, and B200 are not three rungs on one ladder. The H100 is the proven workhorse with an 80 GB ceiling. The H200 is the same chip with a much bigger, faster memory system, built for the memory-bound reality of inference. The B200 is a real generational jump and a new class of machine, with the power and cooling demands to match.

Choose by bottleneck, not by generation. Find the resource your workload runs out of first, capacity, then bandwidth, then compute, and buy the chip that clears it. The newest GPU is only the best one if the thing it does better is the thing you needed.

Frequently asked questions

What is the real difference between the H100 and the H200?
Only the memory. The H200 uses the identical compute die as the H100, so their FP16 and FP8 throughput are exactly the same. The H200 adds 141 GB of HBM3e at 4.8 TB/s, versus the H100's 80 GB at 3.35 TB/s, for 76 percent more capacity and 43 percent more bandwidth in the same 700-watt envelope. Because inference is limited by memory bandwidth, that makes the H200 roughly 40 percent faster at serving large models, despite no change in compute.

Is the B200 just a faster H200?
No. The B200 is a full architecture change, Blackwell rather than Hopper. It uses a dual-die design with 208 billion transistors, 192 GB of memory at 8 TB/s, native FP4 precision, and NVLink 5 at 1.8 TB/s, and it draws 1,000 watts. It delivers roughly 2 to 2.5 times an H100's performance, and more with FP4, but it needs new baseboards, power, and cooling, so it is not a drop-in upgrade the way the H200 is.

Which GPU should I use for LLM inference?
Start with whether the model fits. If it fits in 80 GB at your precision and batches are healthy, the H100 is the cheapest option. If it fits but decode is memory-bandwidth-bound, which most inference is, the H200 is faster at the same size. If the model exceeds 141 GB, or you need FP4 throughput for high-traffic serving, you need a B200. Capacity first, then bandwidth, then compute.

Why does memory bandwidth matter more than TFLOPS for AI?
Because generating text is memory-bound. To produce each token, the GPU reads the entire model's weights from memory, does a small amount of math, and repeats, so token speed scales with how fast weights move, not with peak compute. This is why the H200 beats the H100 on identical compute, and why buying a GPU for its headline teraflops often means paying for performance the workload cannot use.

Do I need liquid cooling for these GPUs?
The H100 and H200 at 700 watts run in well-designed air-cooled facilities. The B200 at 1,000 watts pushes most air-cooled infrastructure past its limits, and rack-scale Blackwell such as the GB200 NVL72, which draws well over 100 kilowatts per rack, requires liquid cooling. Power availability and cooling capacity, not the GPUs themselves, are usually the real constraint on large deployments.

Sources and image credit

Specifications and pricing current as of mid-2026, from NVIDIA documentation and multiple independent benchmark and cloud-provider analyses; memory, bandwidth, and throughput figures are widely corroborated. Real-world token rates are representative of Llama-class FP8 serving and vary by model, batch size, and software. Hero image graded to brand navy; credit to be added on publish.

The only comparison that matters is real throughput, not peak teraflops. See the H100, H200, and B200 scored on real inference at millionminer.com/gpu-ai-benchmarks, and browse the cards in the AI hardware catalog.

Specs are mid-2026 accurate and move fast; refresh before publishing on major NVIDIA releases.

You Don't Actually Understand Bitcoin: 7 Things Almost Everyone Gets Wrong

MillionMiner — Thu, 09 Jul 2026 09:56:44 +0000

Not because you aren't smart. Because the popular version of Bitcoin is wrong in specific, expensive ways. Here are the seven that trip up almost everyone, and the reality behind each.

Ask ten people what Bitcoin is and you will get ten confident answers. Most of them are a little bit wrong, and a few of those small errors are the kind that cost people real money. This is not trivia. The gap between what most people believe about Bitcoin and how it works is exactly where lost coins, drained exchange accounts, and bad decisions live. Here are seven things almost everyone gets wrong, ranked loosely from harmless to expensive.

"My Bitcoin is in my wallet"

The belief: your coins sit inside your wallet app, like cash in a purse.

The reality: they do not. Bitcoin exists only as entries on one shared public ledger, the blockchain. Your wallet holds no coins at all. What it holds is your private keys, the secret codes that prove certain entries on that ledger are yours to spend. Move "your" bitcoin and nothing physically travels anywhere; the ledger simply updates to say those entries now belong to someone else.

This is why backing up a seed phrase is not paranoia. Lose the keys and the coins stay on the ledger forever, visible to everyone, spendable by no one. Your wallet is a keychain, not a vault. People who think the coins live in the app panic when they switch phones, and stay far too relaxed about who can see their keys.

"There are 21 million bitcoin, so that is the supply"

The belief: the cap is 21 million, so 21 million are out there to go around.

The reality: the 21 million cap is real and hard-coded, and about 19.9 million have already been mined. But an estimated 3 to 4 million are gone forever, locked behind dead hard drives, forgotten passwords, and keys that went to the grave with their owners. One man in Wales has spent years trying to get permission to excavate a landfill for a drive holding thousands of coins. That is not an outlier so much as a vivid example of a constant leak.

So the usable supply is meaningfully under 21 million, and it shrinks a little every time someone fumbles a key. The scarcity is tighter than the headline number suggests, which is also why self-custody discipline is the whole game.

"Bitcoin is anonymous"

The belief: it is untraceable, which is why criminals love it.

The reality: this one gets almost everyone. Bitcoin is one of the most transparent systems ever built. Every transaction that has ever happened is public, permanent, and copyable by anyone with an internet connection. It is pseudonymous, not anonymous: addresses are not names, but the moment one address is tied to a real person, its entire history is exposed, both backward and forward. Analytics firms de-anonymize wallets as a business, and law enforcement uses the permanent public record to trace funds precisely because it never disappears. Physical cash is far more private than Bitcoin.

Why it matters: people make privacy assumptions that are simply not true, and act on them.

"When the last bitcoin is mined, it is game over"

The belief: mining stops around the year 2140, and the network dies with it.

The reality: the issuance of new coins ends, but mining does not. Miners keep validating and securing every block; they just get paid in transaction fees instead of newly minted coins. The reward shifting from new coins to fees is written into the design from the first day, and the changeover happens gradually over more than a century. The network keeps running exactly as it does now. Nothing switches off.

"Someone could just change the rules, or switch it off"

The belief: a founder, a company, or a government could raise the supply or shut Bitcoin down.

The reality: there is no CEO, no company, no headquarters, and no off switch. The rules live in software run independently by tens of thousands of computers around the world, and a rule only changes if that network agrees to it. That is why the 21 million cap has never moved despite constant incentive to raise it. Satoshi Nakamoto, the pseudonymous creator, has been silent for more than a decade and has never touched an estimated one million coins.

As for switching it off: China, home to most of the world's mining at the time, banned it outright in 2021. The network's computing power dropped by roughly half, then fully recovered within months as miners simply moved elsewhere. A government can make Bitcoin harder to buy and sell where you live, by regulating the on and off ramps, but "ban the network" and "make it inconvenient locally" are very different things.

"The coins on my exchange are mine"

The belief: your balance on an exchange is the same as holding bitcoin.

The reality: it is not. When your coins sit on an exchange, you own a claim, an IOU, a promise that the company is holding coins on your behalf. If that company is hacked, mismanaged, or insolvent, the promise can evaporate and the coins with it. This is not hypothetical. Customers of FTX, Celsius, and Mt. Gox learned it in the most expensive way possible, collectively losing billions.

The five-word version of this entire lesson is "not your keys, not your coins." It is the single most costly misconception on this list, and the easiest one to fix: move anything you are not actively trading into a wallet whose keys you control.

"It crashes 50 percent, so it must be broken"

The belief: an asset that regularly falls by half is failing.

The reality: Bitcoin is about sixteen years old and still small next to gold's trillions and thousands of years of history. It trades in roughly four-year cycles of boom and bust, amplified by margin and sentiment. Deep drawdowns are not a malfunction; they are what a young, thinly held, emerging asset does on the way to being understood. It has "died" in the headlines hundreds of times and kept going each time.

Volatility is the price of being early. Mistaking normal cyclicality for failure is how people end up buying at the top and selling at the bottom, which is the exact opposite of a plan. The quiet, ugly part of the cycle is, not coincidentally, when patient buyers tend to position, but that is another article.

So how many did you actually know?

If you got all seven, you are rarer than you think, and probably insufferable at dinner. If you missed a few, welcome to the large and well-meaning club of people who own or talk about Bitcoin without quite understanding it. There is no shame in it. The seven points above are most of what separates a confident headline-repeater from someone who understands the asset: where the coins really live, who really controls them, and what you really own.

And they all point at one thing. Nearly every myth on this list dissolves the moment you understand where new bitcoin actually comes from, which is mining. Mining is how coins enter circulation, how the 21 million schedule is enforced, how the network stays secure without a company running it, and how, if you choose, you can produce and hold coins yourself instead of trusting someone else to keep them. You can even model what a machine would earn at your own electricity rate, or read whether Bitcoin mining is worth it before spending a cent. Understand mining, and most of this list stops being confusing. That is the part MillionMiner spends its days on.

Frequently asked questions

Is Bitcoin really not anonymous?
Correct. It is pseudonymous. Every transaction is recorded on a public ledger that anyone can read, and that record is permanent. Your address is not your name, but once an address is linked to you, its full history is visible in both directions. Analytics firms and law enforcement de-anonymize wallets routinely, which is why cash is considered far more private than Bitcoin.

What actually happens if I lose my private keys?
The coins stay on the blockchain forever, exactly where they were, but nobody can move them. There is no password reset and no support line, because there is no company holding them for you. This is the trade-off of true ownership, and it is why a securely stored backup of your seed phrase matters more than almost anything else in crypto.

Can a government shut Bitcoin down?
No single government can switch off the network, because it runs on tens of thousands of independent computers worldwide. China banned mining in 2021 and the network fully recovered within months. What governments can do is regulate the exchanges and services you use to buy and sell, which affects how easy Bitcoin is to access where you live, without touching the network itself.

Is my Bitcoin safe sitting on an exchange?
An exchange balance is a claim on the company, not coins you hold directly. It is convenient for trading, but if the exchange fails, is hacked, or misuses funds, that claim can be lost, as FTX, Celsius, and Mt. Gox customers discovered. For anything you are not actively trading, moving it to a wallet whose keys you control removes that risk.

What is the difference between owning Bitcoin and mining it?
Buying Bitcoin means acquiring existing coins from someone else. Mining means running specialized hardware that helps secure the network and, in return, earns newly issued coins directly. Mining is the process that creates new supply and enforces the 21 million cap, and it lets you produce and self-custody coins rather than buying them on an exchange. Whether it makes financial sense depends mostly on your hardware efficiency and electricity rate.

Sources and image credit

Facts current as of July 2026. Supply and network figures from public blockchain data; exchange-collapse references are matters of public record (FTX, Celsius, Mt. Gox). Hero image graded to brand navy; credit to be added on publish.

Most of this list traces back to one thing: mining, where new coins actually come from. You can model what a machine would earn at your own power rate on the mining profit calculator, and if home electricity is the blocker, hosting runs it at industrial rates while you keep the coins.

Facts current as of July 2026.

After the Bear Market: Why Now Is the Best Time to Buy a Bitcoin Miner

MillionMiner — Wed, 08 Jul 2026 09:20:42 +0000

Bitcoin is down by half, sentiment is at extreme fear, and miners are switching off in record numbers. Which is precisely why the people positioning for the next cycle are buying hardware today, instead of waiting for the rally that will make it obvious.

The most useful thing you can know about Bitcoin mining is that its clock runs backward from everyone else's. When mining feels exciting, when Bitcoin is on the news and your neighbor is asking how to buy a miner, it is the worst time to start. When it feels dead, when the headlines have moved on and operators are unplugging machines at a loss, it is the best. We are in the second kind of moment right now, and it will not last.

Look at where things stand in the summer of 2026. Bitcoin peaked near $126,000 in October 2025 and has since fallen to around $60,000, a drawdown of roughly half. The Fear and Greed Index sits at extreme fear, a level last seen during the collapse of Terra in 2022. And on the production side, something more telling is happening: the miners themselves are capitulating. Galaxy Research confirmed in June that Bitcoin miners had entered a capitulation phase, with network difficulty down more than 20 percent from its November peak, the largest such decline since China banned mining in 2021. Public mining companies have sold Bitcoin at a record pace, more than 32,000 coins in the first quarter alone, exceeding their combined sales for all of the previous year.

To most people, that reads as a reason to stay far away. To anyone who understands the cycle, it is the signal.

What happens after a bear market

Bitcoin has moved in cycles that have been consistent for over a decade, and the engine is the halving. Roughly every four years, the reward paid to miners for each block is cut in half, tightening the supply of new coins. The pattern that has followed each halving is the same in shape: a supply shock, a bull market that peaks twelve to eighteen months later, a hard correction, a long quiet accumulation, and then the next halving resets the clock. Tops formed in late 2013, late 2017, late 2021, and October 2025. Bottoms formed in early 2015, late 2018, and late 2022. Each roughly four years apart.

If that rhythm holds, and no one can promise it will, the phase that follows a bear market is the one that has historically produced the largest gains. The accumulation period, when prices are low and boring and most people have lost interest, is described by long-time market watchers as the point of maximum upside, because it is the moment when coins change hands from those giving up to those willing to wait. The next halving is due in 2028. The people who tend to do well are the ones already in position when it arrives, not the ones scrambling once the rally is undeniable.

That is the strategic backdrop. But there is a specific, mechanical reason the bear is the miner's moment, and it comes down to three things happening at the same time, right now.

Reason one: the hardware has never been this cheap for what it does

Mining machines are priced by profitability. When mining is lucrative, ASIC prices climb; when margins compress, they fall, often hard. We are deep in the second condition. Hashprice, the daily revenue a machine earns per unit of computing power, has fallen to around $29 per petahash per day, down roughly two thirds from the October 2025 peak and near the lowest level since the last halving. When revenue per machine collapses like that, two things follow. Manufacturers and distressed operators discount hardware to move it, and the cost of buying hashrate drops. The best machines now cost around $10 per terahash, against roughly $20 in 2020. You are buying more capacity per dollar than at almost any point in the last four years.

There is a caveat that matters, and skipping it would be dishonest. Cheap hardware is only an advantage if it is efficient hardware. The same squeeze that discounts machines is also making older, power-hungry models worthless, because they cannot cover their electricity at these revenue levels. The line runs at roughly 15 joules per terahash. Current machines like the Antminer S21 XP at about 13.5 joules, and the newest S23 and SEALMINER units pushing under 10, still hold a margin at sensible power rates. Anything much above 20 joules is a space heater. Buying in the bear means buying the efficient tier, not the discounted gear being flushed off the network.

Reason two: fewer competitors means more Bitcoin per machine, today

Here is the part almost no one thinks about. Bitcoin pays a fixed amount to all miners combined and splits it by share of total computing power. When competitors switch off, your slice of every block grows, immediately.

That is not a theory right now, it is happening. As high-cost miners have capitulated, the network's difficulty has adjusted downward again and again. When difficulty dropped more than 10 percent in a single adjustment in June, every machine still running started earning roughly 9 to 11 percent more Bitcoin per day, for doing nothing different. The network has shed close to a quarter of its computing power from the October peak. For the miners who stayed, and for anyone arriving now, the puzzle got easier and the reward per machine went up. You would be buying into a network that is temporarily less crowded than it has been in a year.

The word temporarily is the whole reason this is a window and not a permanent state. The moment Bitcoin's price recovers, those switched-off machines come back and difficulty climbs again. Earlier this year an 11 percent difficulty drop was reversed within two weeks by a record 15 percent jump the instant conditions improved, because surviving operators are well-funded and quick to scale back up. The advantage of a thinner network belongs to whoever is already plugged in when it happens. It is not something you can buy after the fact.

Reason three: you accumulate coins cheaply, and you are already running when it turns

Strip mining down to its economic core and it is a way to acquire Bitcoin. In a bear market it becomes a way to acquire Bitcoin cheaply, while keeping a productive asset in your hands.

Two forces combine. First, your cost basis is low. Every coin a machine earns during the accumulation phase is produced while the price is depressed, which is exactly when disciplined buyers want to be acquiring, except that a miner produces coins rather than buying them, often below the market price when the hardware is efficient and the power is cheap. Second, and this is the piece that separates miners from everyone else, you are already in production when the cycle turns. This is the trap that catches latecomers. When the next bull market becomes obvious, everyone tries to start mining at once. ASIC demand explodes, lead times stretch from days to months, prices spike, and difficulty climbs as all that new hardware floods in. By the time a machine ordered in the euphoria finally ships and switches on, the easy part of the move is over. The miner who bought in the quiet has been hashing the whole way up, at a rate locked in when hardware was cheap and competition was thin.

The bull market rewards the miners who are already mining. It punishes the ones trying to become miners.

The uncomfortable part, and why it is the point

**
None of this is a secret. Every investor has heard buy low, sell high, and be greedy when others are fearful. Almost no one does it, and the reason is not ignorance. It is that the best time to act always feels like the worst. Buying a mining machine today means committing capital while the headlines say Bitcoin is finished, while the chart is ugly, while the people around you think you have lost the plot. That discomfort is not a flaw in the strategy. It is the price of admission, and it is exactly why the opportunity exists at all. If it felt comfortable, the hardware would not be cheap and the network would not be thin, because everyone would already be doing it.

Mining also reframes the bet. Instead of trying to time the exact bottom of a chart, which no one does reliably, you buy a productive asset that accumulates Bitcoin through the boring part of the cycle and is already working when the interesting part arrives. It turns a guess about price into a position built on discipline.

Being honest about the risk

A thesis this clean deserves its counterweight, because buying a miner in a bear market is a considered bet, not a sure thing, and anyone who tells you otherwise is selling something.

Bitcoin's price could fall further before it recovers. Analysts are split: some see a bottom forming in late 2026 in the fifty thousands, others warn of a deeper flush toward the forties, and a few argue the familiar four-year cycle has weakened as the asset matures. Difficulty relief is temporary and reverses fast, as we have seen. Mining is not passive income; it depends on hardware efficiency, uptime, and above all your electricity rate, and at residential power prices even the best machine loses money at today's revenue. The whole argument rests on one assumption, that Bitcoin has a future and that its cycles, in some form, continue. If you do not believe that, none of the rest follows, and you should not buy a miner at all.

What makes the bear defensible is not certainty. It is that you would be buying low on every input that matters at once: cheap hardware, thin competition, and a low coin price to accumulate against. You will not get all three aligned again until the next time it feels this bad.

What buying now requires

If the logic holds for you, the execution comes down to three decisions, and they are the same three that decide whether any miner survives a bear.

Pick efficient hardware. Sub-15 joules per terahash is the survival line at current revenue, so this is not the moment to chase a bargain on an old machine. The current miners ranked by real profit is the place to start, and the full ASIC catalog shows what is shipping today.

Solve the power, because it is the whole game. At a home electricity rate the math does not work at these revenue levels, plainly. The way most people make a bear-market miner profitable is to run it where the power is industrial. Hosting puts your machine in a facility at roughly 7 to 8 cents per kilowatt-hour, keeps it cooled and online, and pays every coin to you, which is often the difference between a home machine that loses money and a hosted one that does not.

Model it before you commit, including against the case this article is making. Put your exact hardware and power rate into the mining profit calculator, run it at both your home rate and a hosted rate, and look honestly at the payback. If the numbers work at a rate you can get, the bear market is handing you an entry the bull market never will. That is the standard MillionMiner builds its tools and its hosting around.

The bottom line

The crowd will start mining Bitcoin when it is obvious, sometime after the next rally is well underway, and they will pay the most for hardware that earns the least, into the most crowded network of the cycle. That is not a prediction. It is what has happened every time.

Right now is the opposite of that. Hardware is cheap, the network is thin, coins are on sale, and almost no one wants in. It is uncomfortable, and that discomfort is the entire reason the opportunity is here. The best time to buy a Bitcoin miner has never once felt like a good time. It feels exactly like this.

If the logic holds, two things make a bear-market miner work: efficient hardware, run on cheap power. Model your own rate on the mining profit calculator, and if a home electricity rate is the blocker, hosting runs the machine at industrial rates while you keep every coin.

Market figures are a July 2026 snapshot; refresh them before publishing if the market has moved.

Two GPU Spec Sheets Were Lying to Him. Here's How to Read Them Right.

MillionMiner — Tue, 07 Jul 2026 12:07:09 +0000

A higher TFLOPS number looks faster. A pricier card has more memory. Neither tells you which one answers your users quicker. The biggest number on a GPU spec sheet almost never predicts real AI performance, because AI is not one workload. It is two, and they break on different things.

Ravi had two GPU spec sheets open, and both were misleading him. One card led on TFLOPS, so it looked faster. The other carried more memory and more bandwidth and cost more. He was building an app that serves a language model to users, and he could not tell from the numbers which card would answer a prompt quicker. He is not alone, and the answer is counterintuitive: the headline compute number is the wrong thing to optimize for, and a card with fewer TFLOPS often wins.

Here is how to compare GPUs for AI, the four numbers that decide it, the one piece of math that explains why the slower-looking card wins, and how to settle any matchup in about a minute instead of an afternoon of spec-sheet guessing.

AI is two jobs, not one

An AI model is a huge pile of numbers, called weights, arranged in layers. Running it means multiplying your input by those weights, over and over, which is why a GPU with thousands of parallel cores beats a CPU with a few fast ones. Modern GPUs add Tensor Cores built specifically for that matrix math, and when you see an AI performance figure, it is almost always the Tensor Cores doing the work.

The whole comparison hinges on one fork. There are two AI jobs, and they stress the GPU differently. Training teaches a model by running data through it and adjusting the weights, for days or weeks, and it is math-heavy from start to finish. Inference is using the finished model to answer a request, which happens millions of times a day every time someone prompts a chatbot. At scale, inference is where most GPU time and money go. Because they are different jobs, they are limited by different parts of the card. Miss that, and every comparison you make is wrong.

The four numbers that decide it

Ignore the marketing and a GPU's fitness for AI comes down to four measurable things. A good benchmark reports all four, because any one alone can mislead.

Compute, in TFLOPS, is the raw rate the Tensor Cores multiply. Higher compute finishes training faster. This is the number spec sheets lead with, and the one most often over-weighted. Memory capacity, in gigabytes of VRAM, is whether the model fits on the card at all. A model that does not fit either fails or spills to system memory, where performance collapses, so capacity is a hard gate. Memory bandwidth, in terabytes per second, is how fast weights travel from memory into the cores, and it is the number that quietly decides inference speed, the one buyers most often overlook. Precision support, FP16, FP8, FP4, is whether the card can represent each weight in fewer bits to shrink the model and roughly double throughput, but only if the hardware supports the format. Blackwell's native FP4 is an example older cards cannot use.

Why the biggest TFLOPS number lies

This is the part that trips up almost everyone. Picture a factory. The cores are workers who assemble products incredibly fast. The weights are raw materials in a warehouse, delivered by truck, the memory bus. During training there is so much math per delivery that the workers stay busy, so more workers, more TFLOPS, means more output. During inference, specifically generating text token by token, the math per delivery is tiny. The GPU reads the entire model's weights just to produce one token, does a little arithmetic, and waits for the next truckload. The workers sit idle. The factory is delivery-bound, not worker-bound.

That gives a rule precise enough to reason with: for token generation, tokens per second is roughly the memory bandwidth divided by the size of the model in memory. Add compute and nothing changes. Add bandwidth and it speeds up.

This explains the industry's favorite example. The H200 uses the same compute die as the H100, so their TFLOPS are identical, yet the H200 serves large models roughly 40 percent faster. The only thing that changed is memory bandwidth, from 3.35 to 4.8 terabytes per second. Rank them by compute and they look equal. In production they are not close. And it compounds: the KV cache, the running memory of a conversation, grows with every token and every user, so longer prompts and more traffic make a workload more memory-bound, not less.

Measure real work, don't trust the datasheet

If raw specs mislead, the fix is to measure real work. The industry standard is MLPerf, a suite that runs the same models on every accelerator so results are comparable, with inference tests refreshed through 2026.

A practical tool makes that usable. The MillionMiner GPU and AI benchmark scores every card on an AI Inference Index normalized to the RTX 3090 at 100, alongside VRAM, compute, and training throughput. The normalized index is the key move: instead of asking you to weigh terabytes against teraflops in your head, it expresses real inference performance as one relative number, so a card scoring 400 does about four times the inference work of the 3090 baseline. Capacity and compute sit beside it for the cases where they, not throughput, decide. It is the spec sheet corrected for how GPUs behave.

Read the benchmark through your workload

The right GPU is entirely a function of the job. Serving a language model to users is memory-bound: lead with the inference index and VRAM, confirm the model plus KV cache fits, then take the highest throughput you can afford. Training or fine-tuning is compute-bound: lead with TFLOPS and training throughput, and the interconnect if the job spans multiple GPUs, since NVLink moves data between cards at around 900 gigabytes per second while PCIe becomes the bottleneck the moment you split a model.

Running a large model on one card makes capacity the gate: a 70B model in full precision needs roughly 140 GB, which rules out most cards before speed even matters. Prototyping or budget work is where a consumer card like the RTX 5090 is strong value, until the index shows you have outgrown it.

Comparing two cards then takes about a minute. Name your workload first, training or inference, model size, single card or not, because that decides which column you read. Open the benchmark tool, pick any two GPUs, and put them side by side.

*Read the column that matches the job and ignore the rest: *
inference index for serving, TFLOPS and training throughput for training, VRAM for fit. Sanity-check that the model fits and weigh performance against price, power, and cooling. Then act, on the card's product page or, if you would rather not own and cool the hardware, on hosting. The wider questions of picking within a budget and how the flagships stack up are covered in the guide to choosing a GPU for AI and the H100 vs H200 vs B200 comparison.

The same trap catches Bitcoin miners

If this pattern feels familiar, it should. The mistake Ravi almost made with TFLOPS is the exact mistake buyers make with mining hardware and hashrate. The highest terahash number is not the most profitable miner, because profit is output minus the power it burns, and efficiency, not raw hashrate, decides the winner. The fix is identical: stop reading the headline spec and measure the real result. Our mining profit calculator ranks machines by actual daily profit at your electricity rate the same way the benchmark ranks GPUs by real inference, not by the number the box wants you to read. One lesson, two kinds of hardware: the biggest number on the sheet is marketing, and the metric that pays is the one you have to measure.

The bottom line

Ravi was serving a model, so he stopped reading the TFLOPS line and sorted by the inference index and VRAM. The card with the lower headline compute won on both, fit his model, and answered users faster for less money. The spec sheet would have sent him the wrong way.

Comparing GPUs for AI is not about the biggest number. It is about knowing which number your workload depends on, then measuring real performance instead of trusting a datasheet. Name the job, read the right column, and the choice is obvious. When you are ready to see the figures for your shortlist, run them through the benchmark tool and browse the cards in the AI hardware catalog. If power and cooling are the real constraint, that is a question of the site, not the silicon, and one MillionMiner is built to solve.
See the real inference numbers for any two of 78 GPUs, normalized to the RTX 3090 at 100: millionminer.com/gpu-ai-benchmarks.

Why Hashrate and TFLOPS Are the Wrong Metrics for Mining and AI Hardware

MillionMiner — Mon, 06 Jul 2026 09:30:05 +0000

The two headline specs everyone reads are gross theoretical ceilings. Neither predicts what a machine earns or how much real work it does. Here is the math, with live numbers, and the two figures of merit that replace them.
Two numbers dominate hardware buying decisions and both are close to useless on their own. For a Bitcoin miner it is hashrate, in terahash per second. For a GPU it is FP16 throughput, in teraflops. Each is a gross, theoretical ceiling, and each hides the variable that decides the outcome. Optimize for either headline and you will routinely buy the wrong machine.

Mining: profit is an identity, and hashrate is one term in it
Daily mining profit is not a mystery. It is a simple identity:

profit_per_day = income_per_day - power_cost_per_day
income_per_day = hashrate x network_revenue_per_hash x coin_price
power_cost_per_day = (watts / 1000) x 24 x electricity_rate

Hashrate appears once, inside income. Power is a separate term, subtracted. The figure of merit that ties them together is efficiency:

efficiency (J/TH) = watts / hashrate

A worked example from a live catalog, at an electricity rate of 0.07 USD/kWh:

Machine Hashrate Power Efficiency Daily profit
Whatsminer M79S 1350 TH/s 20000 W 14.81 J/TH ~ +$15.1
Antminer S23 Hyd 3U 1160 TH/s 11020 W 9.50 J/TH ~ +$23.3

The M79S has about 16 percent more hashrate and earns about 35 percent less profit. The reason is entirely in the efficiency column: 20,000 watts against 11,020 for a smaller hashrate lead. Rank the catalog by hashrate and the M79S is near the top; rank it by daily profit and it drops behind machines producing far less hash. Efficiency, not hashrate, tracks the outcome.
It sharpens across generations:

Machine Hashrate Power Efficiency Daily profit
Antminer S23 318 TH/s 3498 W 11.00 J/TH ~ +$5.6
Antminer S19 XP Hyd 512 TH/s 10600 W 20.70 J/TH ~ +$0.7

The 318 TH/s machine out-earns the 512 TH/s machine by roughly eight to one, because it does more work per joule. Push back one more generation to S19-class hardware near 21 to 22 J/TH and, at the same rate, daily profit goes negative while hashrate still reads in the hundreds of terahash.

There is a second term that swings the sign of the whole equation harder than any hardware choice: electricity_rate. Hold the machine fixed and vary only the rate, and profit crosses from negative to positive somewhere between a residential tariff (0.13 to 0.22 USD/kWh across much of the US) and an industrial one (0.04 to 0.08 USD/kWh). The same box loses money at home and earns at a hosting facility. Any honest ROI model treats the rate as a first-class input, not an afterthought, which is why a live profit calculator that lets you set your own rate is the only kind worth using. The mining numbers above come from one such live tool that ranks 285 machines this way at whatever rate you enter.

AI: TFLOPS is a peak, real inference is a measurement
FP16 teraflops is the theoretical peak floating-point rate of the chip. Three things make it a poor predictor of real performance:

Vendors frequently quote FP16 with 2:4 structured sparsity, which doubles the printed number relative to the dense math most workloads run.
Inference is usually memory-bandwidth-bound, not compute-bound. The card spends its time moving weights and activations, so peak FLOP throughput is not the binding constraint.
The software stack, kernels, drivers, and framework support, determines how much of the theoretical hardware you can reach at all.

The fix is to measure, not to read the datasheet. Run real workloads, language-model token generation, diffusion image generation, and vision, then normalize every card to a common baseline. With the RTX 3090 fixed at an index of 100:

GPU FP16 TFLOPS Inference index (RTX 3090 = 100)
RTX 3090 35.6 100
RTX 4090 165.2 133
RTX 5090 419.1 207
A100 40GB 312.0 152

The RTX 4090 has about 4.6x the paper teraflops of the 3090 and delivers about 1.33x the measured inference. The 5090 has nearly 12x the paper number and delivers about 2x. The theoretical ratio and the measured ratio are different quantities, and only the measured one shows up in tokens per second.

Training is a separate axis again. The A100 posts strong training throughput, roughly 1,396 images per second on a standard benchmark against the 3090's 905, yet sits around 152 on the inference index. A card's rank on one workload does not carry to another, so a single scalar, teraflops or hashrate, cannot summarize hardware worth.

This is exactly what a workload-indexed benchmark is for. MillionMiner's GPU and AI benchmark comparison indexes 78 cards on measured LLM, image, and vision performance against the RTX 3090 at 100, with VRAM and training throughput alongside and a head-to-head that reports the measured gap rather than the datasheet one.

The Machines Ship in Weeks. The Power Takes Years.

MillionMiner — Fri, 03 Jul 2026 10:34:58 +0000

A Gulf family office had the capital and the hardware. A fleet of the latest miners and a rack of GPUs for a second revenue line, priced and ready. What it could not buy at any speed was a place to plug them in. Every US site came back with the same answer from the utility: yes, we can serve that load, in about five years. The machines were ready in three weeks.
Omar solved the easy part first. His family office had the money, and his team had already priced the miners and the GPUs. What they did not have, and could not buy with any amount of capital, was energized power. Every US site they liked came back with the same reply from the local utility: we can serve that load in about five years. The machines were three weeks out. The power was half a decade away. That gap is the entire business, and the contract that closes it is the power purchase agreement.

If you are looking at buying a mining farm or an AI data center, the power purchase agreement is the document that decides whether the deal works. Not the hashrate, not the GPU model. The rate and terms under which electricity flows to the site are where the money is made or lost. Here is what a PPA actually is, why energized power has become the scarcest asset in the entire build-out, and how serious buyers get it.

What a power purchase agreement actually is

A power purchase agreement is a contract between a buyer, the offtaker, and a seller of electricity, which can be a utility, an independent power producer, or a generator built on the site itself. In plain terms, it is how you agree to buy power at a set price for a set period instead of paying whatever the market charges on a given day. The difference between that and a floating utility bill is certainty, and certainty is what makes an operation something a lender or an equity partner will underwrite for a decade rather than a bet on next winter's energy prices.

Five terms carry the weight. The price, a rate in dollars per kilowatt-hour or megawatt-hour, fixed or indexed, which flows straight into your margin. The term, usually 5 to 20 years, since a long horizon is what makes the operation bankable. The volume and load profile, meaning how many megawatts and how steadily you draw them, because a flat, high-utilization load earns a better rate than a spiky one. Interruptibility, whether you agree to power down when the grid is stressed, which lowers your rate and, for a flexible load like Bitcoin mining, can earn demand-response payments. And cost allocation, who pays for transmission upgrades and interconnection, a single line item that has sunk more data center deals than any disagreement over price.

Why the power is the moat, not the machines

A few years ago the hard part of mining was getting the hardware. Today the hardware is a commodity you can order tomorrow, and the power is the constraint. The numbers behind that shift are stark.

In Texas alone, the interconnection queue holds more than 438,000 megawatts of proposed load, and data centers account for roughly 90 percent of it. Nationally, US power demand is growing about 23 percent a year, and interconnection delays now stretch past five years. The equipment is no better: lead times for the high-voltage transformers and switchgear a large site needs run three to five years, and often longer. You cannot pay to skip that line. Money moves the machines. It does not move the queue.

The result is an inversion that caught most of the industry off guard. The Bitcoin miners who spent 2021 grinding through utility queues, negotiating interconnection, and energizing megawatts now hold exactly what the AI build-out needs most. Pre-energized sites like Riot's 700 megawatt Rockdale facility and Core Scientific's 1.2 gigawatt footprint are worth far more than any greenfield parcel, and the market has begun to reprice the companies that own them. The miners learned early that you cannot bully your way past a utility queue. The hyperscalers are learning it now.

The practical takeaway for a buyer is blunt. A parcel of land with a signed PPA and a live interconnection is the deal. The ASICs or GPUs that fill it are the easy, late-stage part.

The four ways to secure the power

Every path to a working site ends in a price per kilowatt-hour you can rely on. They differ in how long they take, what they cost, and how much execution risk you absorb, and that is the order to weigh them in.

The first is a direct utility PPA and grid interconnection. You contract with a utility or power producer and connect to the grid. It is the classic route and it produces clean, grid-reliable supply, but it lives and dies by the interconnection queue, and a new large load can wait three to five years or more with that timeline risk on you. The FERC generator interconnection process and the national queue data from Berkeley Lab show just how long the line has become.

The second is behind-the-meter, or bring-your-own-power.

Instead of waiting for the grid you generate on site, most often with natural gas, sometimes solar, and for the largest players nuclear. This is the fastest and often the cheapest power, which is why the industry now treats bring-your-own-power as its own category. The catch is execution risk: you are running a power plant and a data center at once, with gas siting, air permits, and construction all in the critical path. Gas is expected to dominate new data center power over the next five years precisely because it is buildable, but the permitting is the gate.

The third is surplus interconnection. You attach your load to an existing interconnection agreement, a solar plant with spare capacity for instance, and skip the queue. It is faster and can be cheaper, but your rights sit subordinate to the host's, so if the host loses its interconnection, you lose yours. A real tool, used carefully.

The fourth is to buy a site that is already energized. The fastest route is not to build power at all but to acquire a site where the PPA and the interconnection are already in hand and the megawatts are live. You inherit a locked rate and a working connection on day one, and you skip the queue, the transformer wait, and the permitting risk. For most buyers this is the only route that turns a multi-year infrastructure project into a transaction that closes this quarter.

How the rate decides everything
Because power is the dominant cost, 70 to 80 percent of a mining operation's operating expense and the largest line in an AI data center, the rate you lock is not one input among many. It is close to the entire model, and a swing of a few cents compounds across every hour of every year.

Take a modest 6 megawatt site. The difference between a behind-the-meter rate near 3 cents per kilowatt-hour and a grid rate near 7 cents is four cents. Run it out: 4 cents times 6,000 kilowatts times 8,760 hours a year is more than 2 million dollars in annual margin, from the power contract alone, before a single machine is chosen. Over a ten-year PPA that one term is worth more than 20 million dollars. This is why buyers who understand the space negotiate the rate first and shop for hardware last. It is also why the market is now closing a valuation gap: operators with cheap, locked power were priced like miners while data center operators trade at multiples several times higher, and as the power owners deliver on AI leases, that spread compresses. The rate on the page is doing the heavy lifting.

Mining and AI want different power deals

The same site can host Bitcoin miners or AI servers, but the two loads want different contracts, and that shapes the PPA you should sign. Bitcoin mining is flexible: a miner can power down in seconds without harm, which lets you sign an interruptible or curtailable PPA at a lower rate and, in markets like Texas, sell power back during peak demand for real credits. Flexibility is a feature you get paid for. AI compute is the opposite. A training or inference cluster cannot switch off mid-job, so it needs continuous, firm supply, a more expensive but necessary product. The strongest sites keep both doors open, absorbing cheap, interruptible power with a flexible mining load today while holding the interconnection and buildout headroom to host firmer AI demand as it arrives. The pivot from mining to AI is, underneath, a story about who controls the power.

The traps that sink powered-land deals

Once you are shopping for an energized site, due diligence is where deals are won or lost, and the same failures recur. A will-serve letter is not powered land: a utility saying it can serve a load someday is not a contract for capacity by a date certain, and what matters is a firm interconnection with a real energization date, or, for on-site generation, air permits and fuel supply already in hand. Co-location carries live regulatory risk: federal regulators have already pushed back on behind-the-meter arrangements, rejecting an amendment that would have let a data center draw directly from a nuclear plant and then opening a broader review of how co-located loads are treated, so a deal built on the old assumptions can unwind.
Much of what gets marketed as a project is only paper, with no load study, no signed interconnection, and no real line of sight to energization, so ask for the study, the agreement, and the date, and if they are missing, the megawatts are hypothetical. And a headline rate below what infrastructure actually costs tends to hide fees, curtailment surprises, or a counterparty that cannot hold the price, because a durable rate beats a cheap number that resets.

How to get a PPA, or buy a site that already has one

There are three honest paths, and they suit very different buyers. You can negotiate a PPA directly, which is viable if you have scale, strong credit, and patience, since you will spend years in queues and carry the interconnection risk yourself; this is the domain of utility-scale developers and the largest operators. You can build your own power, where the bring-your-own-power route is fast and cheap on paper but loads a power plant's worth of permitting and construction risk on top of the data center. Or you can acquire a site that is already energized, which for most investors is the practical answer: you inherit the rate and the connection and deploy machines in weeks instead of years.

That third path is what MillionMiner's turnkey mining farms and AI data centers are built around, each listed site coming with the power already secured, so the timeline that kills most projects is behind you before you start. If you would rather run machines without owning the site at all, US hosting rents you space in a facility that already holds the power, and the guide to what a mining farm is covers the layer above the contract.

The bottom line

Omar stopped shopping for utilities and bought a site that was already energized. His miners were hashing six weeks later, on power locked at a rate his spreadsheet could finally trust, while the operators still chasing interconnection were looking at 2029. The lesson is the one this whole market is learning: in mining and in AI, the machines are the commodity and the power contract is the moat.

A power purchase agreement is not paperwork you handle at the end. It is the asset. Understand the rate, the term, and above all whether the interconnection is real, and you understand the deal. When you are ready to look at sites where the power is already solved, start with the turnkey farms and check where the economics are strongest in the best states for US mining guide.
If you would rather look at sites where the power is already secured and live, these turnkey mining-farm and data-center listings are built around exactly that: millionminer.com/mining-farms.

Hero photograph by Vyacheslav Argenberg, via Wikimedia Commons, CC BY 4.0. Cropped and graded to brand navy.

DEV Community: MillionMiner

The Bitcoin Mining Death Spiral Everyone Fears: Does It Actually Exist?

Bitcoin Mining and AI Are Becoming the Same Business, and 2026 Is the Year to Get In

The Most Profitable Bitcoin Miners Right Now, and Why 2026 Is the Time to Buy

Can a $2,000 Gaming GPU Replace a $30,000 AI Chip?

So You Want to Mine Bitcoin in 2026? Read This First

Bitcoin Mining Makes No Sense Until You Picture a Fishing Fleet

H100 vs H200 vs B200: The Real Differences, and How to Choose in 2026

You Don't Actually Understand Bitcoin: 7 Things Almost Everyone Gets Wrong

After the Bear Market: Why Now Is the Best Time to Buy a Bitcoin Miner

What happens after a bear market

Reason one: the hardware has never been this cheap for what it does

Reason two: fewer competitors means more Bitcoin per machine, today

Reason three: you accumulate coins cheaply, and you are already running when it turns

The uncomfortable part, and why it is the point

Being honest about the risk

What buying now requires

The bottom line

Two GPU Spec Sheets Were Lying to Him. Here's How to Read Them Right.

AI is two jobs, not one

The four numbers that decide it

Why the biggest TFLOPS number lies

Measure real work, don't trust the datasheet

Read the benchmark through your workload

The same trap catches Bitcoin miners

The bottom line

Why Hashrate and TFLOPS Are the Wrong Metrics for Mining and AI Hardware

The Machines Ship in Weeks. The Power Takes Years.

What a power purchase agreement actually is

Why the power is the moat, not the machines

The four ways to secure the power

*The second is behind-the-meter, or bring-your-own-power. *

Mining and AI want different power deals

The traps that sink powered-land deals

How to get a PPA, or buy a site that already has one

The bottom line

The second is behind-the-meter, or bring-your-own-power.