Originally published at mrnasdog.com/research/icp/inflation by MrNasdog.
The Internet Computer is one of the few chains that genuinely destroys its own coin — ICP is burned, not transferred, when it is converted into cycles to pay for compute — but over the 90 days to Sep 4 2026 it destroyed 111.43K ICP against 2.94M ICP minted for node providers and neuron maturity. The MrNasdog Pressure Framework reads ICP at +0.51% net supply growth over the window and projects +0.46% forward, against a supply monitor at +0.85% — a gap of 0.34 percentage points, inside tolerance. The real constraint is not the emission: it is the 96.00M ICP-equivalent of unminted neuron maturity, 17% of supply, standing as a mint claim with no schedule attached.
The verdict, in one paragraph
For the 90-day window ending Sep 4 2026, the MrNasdog Pressure Framework reads ICP at +0.51% net supply growth and projects +0.46% for the next 90 days. Total sell pressure was 2.94M ICP of fresh minting; total buy pressure was 111.43K ICP burned into cycles. Our supply monitor reads the same window at +0.85%, leaving a gap of 0.34 percentage points — inside the half-point tolerance, so no monitor-gap chip ships on the ICP overview. Almost all of that gap sits in the starting denominator rather than in the flow: the monitor's market-derived supply for Jun 6 2026 reads about 2.04M ICP below what the Internet Computer ledger itself recorded that day, and measured against the ledger's own starting supply the framework figure is +0.51%. The cite-able label for ICP is an uncapped layer-1 with a real usage-linked burn that is currently one twenty-sixth the size of its own issuance.
Sell pressure: where new ICP comes from
New ICP comes from exactly two places, and because they run on different clocks the ledger keeps them in two rows. Sell #5, the node-operator reward mint, is 1.91M ICP and it is the larger of the two: the operators who run the physical machines behind the Internet Computer are paid once a month, in a single event, at an amount denominated in XDR and settled in ICP. Three of those events fell inside this window — 696.68K ICP on Jun 14 2026, 668.53K on Jul 15 2026 and 541.88K on Aug 14 2026, totalling 1.91M ICP, about two thirds of all new supply. Sell #1, protocol inflation, is the other leg at 1.03M ICP: NNS voting rewards are credited as maturity, an internal accounting quantity that is not a coin, and new ICP exists only at the moment a holder disburses it. Together they make 2.94M ICP, counted one minting block at a time — 15,689 of them — and then checked against the Internet Computer's own supply record, which they reproduce to the smallest unit the coin has.
One thing changed inside the window, and it matters for the forward number. On Jul 6 2026 the NNS executed a proposal setting a floor of 2 XDR per ICP in the node reward calculation, so that when ICP trades cheaply the network mints fewer coins to cover the same XDR-denominated bill. The floor binds: the July and August reward events both ran at the floor rather than at the market rate, and at August's rate the floor cut that month's node-provider minting by roughly 22% against the old rule. Because one of the window's three payments happened before the vote and two happened after, the trailing 90-day figure is a blend of a retired mechanism and a live one. The forward column for that row is therefore rebuilt on the latest realised post-vote payment repeated three times, giving 1.63M ICP against the window's 1.91M. Nothing else moves with it: the proposal's own text scopes the floor to node reward calculations, so the maturity leg and the burn keep their measured rate rather than being rebased for a change that does not act on them.
Sell #2, vesting unlocks, is zero, and the reason is structural rather than a quiet quarter. On the Internet Computer, every coin sitting inside a staking neuron is already inside the tradable float that the wider market counts: the market's circulating figure of 556.20M ICP and the chain's own total supply of 556.20M ICP are the same number, 115 coins apart. So a launch-era neuron finishing its dissolve converts locked ICP into liquid ICP without creating a single new coin. The genesis allocation — 469.21M ICP at the May 10 2021 launch, with the DFINITY Foundation on roughly 23.9% and seed participants on roughly 24.7% — was placed on four-to-eight-year dissolve schedules that are now five years spent. Booking those releases as sell pressure would count the same coins twice against the denominator they already sit in.
Sell #3, Foundation and unscheduled unlocks, is zero on value, with no observed release inside the window, but it carries the largest overhang on this page and gets its own section below. Sell #4, long-term locked or bankruptcy, is zero: no bankruptcy estate or court-ordered distribution holds ICP, so there is no trustee release to track. The one genuinely locked pool is voluntary — 289.16M ICP in staking neurons, 242.92M of it not yet unwinding — and it too sits inside the counted float.
Buy pressure: where new ICP goes
Buy #2, protocol fee burn, is the Internet Computer's one genuine buy mechanism, and it is real destruction rather than a transfer to an unspendable address. When a developer converts ICP into cycles — the fuel canisters spend on compute, storage and bandwidth — the ICP is removed from the ledger outright. There is no burn address to inspect, and that is the trap on this coin: every burn block on the ICP ledger has no destination account at all, so a build that looks for a rising dead-address balance finds nothing and reports a burn of zero. The proof is arithmetic instead. Had those coins merely moved somewhere, total supply would have risen by the full mint of 2.94M ICP; it rose 2.83M. The 111.43K difference is the burn, and it lands to the last unit — 111.30K of it cycles conversion, 0.13K ordinary ledger transfer fees. That is a working, usage-linked burn, and it offsets under 4% of what the network created.
Buy #1, programmatic buyback, is zero. No contract repurchases ICP on the open market today. A tokenomics paper published in February 2026 proposes a future cloud-engine revenue split under which the network would take a fifth of revenue to buy and burn ICP, but no revenue flows through it and there is nothing on-chain to read, so it is a watch item rather than a figure. Buy #3, Foundation buy, is zero: no public evidence of a Foundation purchase inside the window. Buy #4, new long-term lock, is zero, and here the measured direction settles it rather than an argument about capacity. Staking a neuron on the Internet Computer is a genuine multi-year commitment with a dissolve delay — but the total staked fell across the window, from 292.22M ICP on Jun 6 2026 to 289.16M on Sep 4 2026, a net 3.07M released. Neurons are dissolving faster than they are being created. And because staked ICP already sits inside the counted float, even a growing stake would offset nothing on this page.
Foundation and overhang
The dominant overhang on ICP is not a wallet — it is a claim. NNS voting rewards accrue as maturity rather than as coins, and the Internet Computer's governance canister reported 96.00M ICP-equivalent of maturity outstanding on Sep 4 2026, up from 95.31M on Jun 6 2026. That is 17% of the entire supply standing as a mint claim with no schedule, no cliff and no deadline: it becomes real ICP whenever holders decide to disburse it, and only then. This is also why ICP's headline inflation figures and its realised inflation disagree so sharply. DFINITY's own tokenomics paper quotes gross issuance of 9.72% falling to 5.42% and states plainly that those figures assume full maturity disbursement, with realised inflation lower in practice. Measured on the ledger, realised minting runs at about 2.1% a year. The framework books the maturity claim as an overhang rather than a flow, because capacity is not cadence — but a change in holder behaviour, not a change in protocol, is what would move this page most.
A nearer-term slice of the same claim is already committed: 22.6K ICP-equivalent of maturity sits in the disbursement queue and clears within days. Behind the aggregate sit two identified group-controlled holdings. The DFINITY Foundation was allocated roughly 23.9% of the 469.21M ICP launch supply and holds it across a set of neurons rather than one published treasury wallet — the Foundation's public governance neuron stakes only 10 ICP and exists to vote, not to custody — so the balance is tracked through disclosure rather than a single address read, and re-walked on every rebuild. The community fund, the Internet Computer's DAO-controlled pool, holds 15.28M ICP staked plus 5.73M ICP-equivalent of its own accrued maturity. If the balance of any of these overhangs falls between refreshes, the outflow enters Sell #3 at the next refresh.
How ICP compares to other uncapped layer-1 chains
Against hard-capped chains, ICP starts from a structurally weaker position and partly earns it back. A halving-model chain with a fixed cap has an issuance schedule nobody can vote to change, and its inflation falls on a calendar. The Internet Computer has no cap at all, and its issuance is set by NNS governance — the Jul 6 2026 rate floor is proof of both the risk and the remedy, since a vote cut node-provider minting by about a fifth in a single month. What ICP has that most capped chains do not is a burn that is tied to actual usage: cycles are bought with ICP and the ICP is destroyed, so every unit of compute the network sells removes supply permanently. A capped chain with no burn can only ever slow down; ICP can, in principle, go into reverse.
Against other uncapped continuous-emission layer-1s, the comparison turns on where the rewards go. Most proof-of-stake chains mint a staking reward and deliver it as a spendable coin immediately, so the full emission hits the float the day it is issued. The Internet Computer routes voting rewards through maturity first, which means a large share of nominal issuance never becomes a coin at all until someone asks for it. The realised gap is stark — a headline near 9.72% against roughly 2.1% actually minted. That is genuinely better than a chain that mints the whole number, but it is a deferral rather than a cancellation: the 96.00M ICP-equivalent claim is still there, and it grows.
Against fee-burning chains, ICP's burn is the right shape but the wrong size, at least so far. A base-fee burn on a busy settlement layer can exceed issuance and flip a chain deflationary. The Internet Computer's cycle burn removed 111.43K ICP against 2.94M minted — a ratio of about 1 to 26. DFINITY's own analysis puts the crossover explicitly in demand terms: reaching its inflation target requires the cycle burn rate to rise roughly fifteenfold from where it sits today. That is a usage question, not a tokenomics question, and it is the single variable that would move ICP from mildly inflationary to genuinely deflationary.
What to watch in the next 90 days
First, Sep 14 2026, Oct 14 2026 and Nov 14 2026: three monthly node-provider reward events land inside the next window, and each will be the largest single mint of its month. Second, the ICP/XDR rate at each of those events — the 2 XDR floor only holds issuance down while ICP trades below it, so a strong price rally would quietly restore the higher minting rate. Third, the maturity balance, now 96.00M ICP-equivalent: a sustained rise in disbursement would show up directly in Sell #1 without any protocol change at all. Fourth, the cycle burn rate, which needs to climb by roughly an order of magnitude before the burn meaningfully offsets issuance. Fifth, any NNS proposal implementing the remaining tokenomics measures — the voting reward pool cap and the cloud-engine buy-and-burn split are both proposed and neither is live, and either one landing would change this page.
Summary
The Internet Computer is an uncapped layer-1 that mints ICP for node providers and staking rewards and destroys it when developers convert it into compute. Over the 90 days to Sep 4 2026 it created 2.94M ICP and destroyed 111.43K, for +0.51% net supply growth, and +0.46% is projected forward on the post-vote run rate. The burn is real and usage-linked, which is more than most chains can claim, but at roughly one twenty-sixth of issuance it is not yet a counterweight. The key risk is the 96.00M ICP-equivalent of unminted neuron maturity — 17% of supply that becomes coins on holder demand rather than on a schedule — and the key constraint is that there is no cap at all: what holds ICP issuance down is a governance vote, which is exactly the kind of thing another governance vote can undo.
MrNasdog Pressure Framework analysis of ICP, Metric 1 — Inflation. Data + explanation only. Not financial advice. Updated Sep 5 2026.
Top comments (0)