DEV Community

Cover image for Is Anthropic Quietly Building What Bitcoin Always Promised?
Mashraf Aiman
Mashraf Aiman

Posted on

Is Anthropic Quietly Building What Bitcoin Always Promised?

#web3 #cryptocurrency #bitcoin #ai

Is Anthropic Quietly Building What Bitcoin Always Promised? A Developer's Deep Dive

"Anthropic isn't trying to kill Bitcoin. That would require them to care about it. They're too busy building something that accidentally does the same job better."
— Mohit Aggarwal, Medium (March 2026)

A viral piece by Jose Crespo, PhD, titled "Anthropic is Killing Bitcoin" has been circulating since February 2026 — collecting thousands of reactions, rebuttals, and LinkedIn reposts. The thesis is bold: Anthropic has accidentally built an AI-native currency through its token-based API pricing, one that outperforms Bitcoin by six orders of magnitude in the tasks that actually matter for a machine economy.

Is this hype? Is it right? Is it somewhere in the middle?

As developers, we're uniquely positioned to cut through the noise. Let's go data-first.

The Argument, Distilled

The Crespo article makes three core claims:

  1. Capital is already deserting Bitcoin infrastructure in favor of AI compute contracts — and the former builders of crypto are signing the checks.
  2. Anthropic's API credit system functions as a de facto machine-to-machine currency — faster, cheaper, and more scalable than anything in the crypto stack.
  3. The "total machine economy" — a world where AI agents transact autonomously — needs a different kind of money, and the crypto stack isn't it.

Let's examine each with real data.

Claim 1: Capital Is Fleeing Crypto for AI Infrastructure

This is the most empirically defensible part of the thesis, and the numbers are striking.

The Mining Exodus

According to CoinShares' 2026 Bitcoin Mining Outlook, mining revenue is projected to drop from ~85% of total revenue in early 2025 to under 20% by end of 2026 for companies that have successfully pivoted to AI contracts. The same companies that built crypto's physical backbone are now signing decade-long AI infrastructure deals.

By October 2025, bitcoin miners had announced $65 billion in contracts with major technology companies and cloud providers — a figure that CoinShares describes not as a hedge, but as a structural exodus. Companies that have made the switch can generate 80–90% operating margins from AI deals, versus the notoriously thin margins of Bitcoin mining.

Some notable examples from publicly traded miners:

  • Core Scientific (CORZ) — Signed 12-year contracts with CoreWeave worth an estimated $6.7 billion, providing 200+ megawatts of GPU hosting capacity.
  • Bitfarms — Announced a full exit from Bitcoin mining by 2027 to operate exclusively as an AI data center operator.
  • Bitdeer Technologies — Reduced its BTC holdings to zero to fund AI data center expansion.
  • Riot Platforms — Sold $200 million worth of Bitcoin in the final two months of 2025, treating it as a funding tool for AI capex.

A separate analysis published in February 2026 documented a 400% increase in sector-wide data center capital expenditure between March 2025 and February 2026, driven entirely by AI infrastructure buildout. As one Q1 2026 earnings call put it bluntly: Bitcoin mining "doesn't make a lot of sense" at current hashprices compared to the returns available in AI infrastructure.

The VC Capital Ratio

Crunchbase's end-of-year 2025 report confirmed another number Crespo cites: AI captured $211 billion in venture funding in 2025, up 85% year over year from $114 billion in 2024. That makes AI the dominant sector for venture capital — representing roughly 50% of all global VC deployment.

Crypto, by comparison, raised an estimated $17–18 billion in venture funding across 2025 — meaningful, but less than 10% of what AI attracted. The ratio Crespo references (~12:1 in favor of AI per invested dollar) is directionally accurate.

The physical layer of Bitcoin is quietly becoming the physical layer of AI. That is not speculation — it's a signed contract.

Claim 2: API Tokens as AI-Native Currency

This is where the argument gets more interesting — and more contested.

The claim: Anthropic's API credit system already powers billions of machine-to-machine microtransactions daily — functioning as a currency that is faster, cheaper, and infinitely more scalable than Bitcoin.

Let's break down the mechanics from a developer perspective.

How Token Pricing Actually Works

As of early 2026, Anthropic's Claude API is priced as follows (in USD per million tokens):

Model Input Output
Claude Haiku 4.5 $1.00 $5.00
Claude Sonnet 4.5 $3.00 $15.00
Claude Opus 4.5 $5.00 $25.00

With prompt caching enabled, cache reads drop to 0.1× base input cost — making repeated context retrieval extremely cheap. Batch processing APIs reduce costs by up to 90% for non-time-sensitive workloads.

A typical small agentic API call — say, 1,000 input tokens and 500 output tokens — costs roughly $0.0000035 to $0.000013 depending on model. That's sub-cent micropayments at scale, processed instantly, with no gas fees, no mempool, no 10-minute confirmation windows.

For comparison: Bitcoin's average transaction fee has ranged from $1.50 to $15+ in recent years, with confirmation times measured in minutes. For machine-to-machine micropayments at millisecond latency and sub-cent denominations, the architectural mismatch is real.

The "Currency" Critique

But here's what the rebuttals get right — and they are worth reading. The Medium piece from Dark Energy Articles (published March 2026) makes a sharp counterpoint: Anthropic's API credits are not a currency in any meaningful sense.

They have no:

  • Transferability — you can't send your Claude credits to another user or agent.
  • Interoperability — they only work within Anthropic's closed ecosystem.
  • Decentralization — Anthropic can revoke, reprice, or discontinue them at will.
  • Store of value — unused credits depreciate as pricing changes.

Calling API credits a "currency" conflates a closed-loop billing system with an open monetary network. Bitcoin, whatever its speed and cost limitations, is permissionless, censorship-resistant, and operates without any single company's approval.

What Anthropic actually has is a highly optimized internal unit of account for AI compute — not a currency.

The X402 Protocol: The Real Story Nobody's Covering

Here's what both the viral piece and most of its rebuttals missed: the x402 protocol, a decentralized, blockchain-based micropayment standard built specifically for AI agent-to-agent transactions.

Developed in 2025 and adopted quickly by Google Cloud, AWS, and Anthropic, x402 enables real-time, low-cost micropayments for API access, data, and compute — designed explicitly for the machine-centric economy both sides of this debate are arguing about.

This is important: the actual convergence isn't "Anthropic replaces Bitcoin." It's AI infrastructure adopting blockchain-native payment rails for the agentic layer. The real fight is over which settlement layer becomes the standard for autonomous AI commerce — and that race is genuinely open.

Claim 3: The Total Machine Economy Needs Different Money

This is the most forward-looking part of the thesis, and also the most debatable.

The core insight is correct: agentic AI systems that autonomously pay for compute, data, and services need a different payment primitive than anything currently dominant. Credit card rails are too slow and too human-permission-dependent. Bitcoin is too expensive and too slow for sub-cent micropayments. Ethereum gas fees make microtransactions economically absurd.

What the machine economy actually needs:

  • Sub-millisecond settlement for real-time agent coordination
  • Sub-cent denominations for compute metering
  • Programmable, conditional payments (pay-for-outcome, not pay-for-input)
  • Cross-provider interoperability so agents aren't locked into one cloud

None of the current major players — not Anthropic's API credits, not Bitcoin, not ETH — perfectly satisfies all four. The space is genuinely unsolved.

What Claude Opus 4.6 Actually Did to Markets

One piece of context Crespo references: in early February 2026, Anthropic unveiled Claude Opus 4.6, and the market reaction was telling. Software stocks dropped despite strong earnings. Bitcoin fell as investors fled to stable assets. IBM dropped 11% weeks later when Anthropic announced Claude Code could automate COBOL modernization.

These aren't just "AI hype" moves. They're the market pricing in something real: AI is eating the application layer, and companies whose moats were "specialized software knowledge" are facing an existential repricing. Anthropic's annualized revenue reached $3 billion by summer 2025, up from $1 billion in late 2024 — a growth rate that validates the infrastructure thesis.

The Developer Verdict: What's Real, What's Hype

After pulling the data, here's the honest breakdown:

What Holds Up

  • The infrastructure migration is real and documented. $65B in signed contracts, 400% capex increase, miners zero-ing out BTC holdings to fund GPU buildouts. This is happening.
  • The capital ratio is real. For every dollar going into crypto infrastructure, roughly $12 went into AI in 2025. The physical layer is being reallocated.
  • API microtransactions are architecturally superior for AI-to-AI payments. The latency and cost comparison is not close if you're building agentic systems.
  • The machine economy is coming. Autonomous agents that need to pay for resources is a solvable engineering problem, and it's being worked on now.

What's Overstated

  • Anthropic's API credits are not a currency. They're a billing system. Closed, non-transferable, controlled by a single private company. Calling them a "monetary system" is a category error.
  • Bitcoin isn't dead — it's evolving. CoinShares' own 2026 outlook projects BTC at $110,000–$150,000 depending on macro conditions. Stablecoin volumes now rival Visa and Mastercard combined. Bitcoin's role is shifting toward institutional settlement and store of value, not micropayments — and that may be exactly right.
  • The "six orders of magnitude" framing is misleading. Comparing Bitcoin transaction throughput to Anthropic API call volume is comparing apples to fusion reactors. They're solving different problems.

What's Genuinely Uncertain

  • Which payment layer wins for the agentic web. X402, Solana micropayments, Layer-2 solutions, or something not yet built — the race is real and genuinely open.
  • Whether Anthropic's infrastructure lead is durable. OpenAI, Google DeepMind, and others are building equivalent infrastructure. The "accidental monetary system" argument depends on Anthropic maintaining architectural centrality, which isn't guaranteed.

What This Means If You're Building

If you're shipping AI-native products in 2026, here's the practical takeaway:

1. Design your architecture for compute-as-currency from day one.
Your agents will need to pay for resources. Build payment abstractions into your agent framework now, not after. Assume the settlement layer will change.

2. Watch x402 closely.
It's the most credible candidate for becoming the standard micropayment protocol for AI agents. AWS, Google Cloud, and Anthropic are already in. If this becomes the HTTP of AI payments, early movers will have significant advantage.

3. Don't bet on one provider's credit system as your payment primitive.
Anthropic can (and does) change pricing, rate limits, and model availability. If your business logic is coupled to a specific provider's billing unit, you're one pricing announcement away from a restructure.

4. The infrastructure layer is where the real money is flowing.
$65B in signed contracts, $211B in VC, 400% capex growth. If you're building infrastructure tooling, observability, or orchestration for AI workloads — that's where capital is pointing.

Sources & Further Reading

Found this useful? Drop a comment with what payment layer you're building on for your agents — genuinely curious where developers are landing on this.

Top comments (0)