NEAR Protocol: the router earns in dollars, the token inflates in percent
NEAR's House of Stake voted to cut maximum issuance from 2.5% to 1.6%. We measure the gap the vote targets: about $10M of NEAR issued every 30 days against $1.84M of Intents revenue captured for buybacks, plus what the founder told Milk Road.
Protocol analysis, November 11, 2026
NEAR Protocol's tokenholder body, the House of Stake, approved a proposal on November 2 to lower the network's maximum annual token issuance from 2.5% to 1.6% over 24 months [1][3]. The tally was 3,287,729 veNEAR in favor and none against, across 36 votes [1]. The proposal, HSP-029, sits in a Security Council veto window that closes November 16, and it binds nobody until validators adopt the protocol upgrade that carries it [1][3].
Two days after the vote closed, co-founder Illia Polosukhin gave the reason on the Milk Road podcast. The demand NEAR is building arrives as dollar revenue that buys tokens, he said, while inflation is a fixed percentage of supply, so the first does not reliably catch the second [4]. In the same interview he sized the token's future by setting a 1.3 billion supply against a billion prospective holders and told the audience, "you do the math" [4]. The math that can be checked today is narrower: how much NEAR's businesses return to the token against how much the protocol prints.
Three businesses on one token
Polosukhin describes NEAR as a vertically integrated stack, with private AI agents as the front end, a blockchain as the settlement layer, and a market connecting the two [4]. In practice the name covers three operations with different economics.
The first is the chain, a sharded proof-of-stake network with blocks around 600 milliseconds and finality near 1.2 seconds [6]. On November 11 it had 351 validators securing 582 million staked NEAR, 44% of a 1.31 billion supply [7]. Users pay gas in NEAR and, since the nearcore 2.14 upgrade, the protocol burns all of it [5][7]. Gas fees came to about $157,000 over the 30 days to November 9 [9]. The chain is the root of trust. It is not where the money is.
The second is NEAR Intents, a cross-chain router. A user states a trade, Zcash for bitcoin, say. Market makers called solvers quote for it off-chain, and settlement is recorded in one contract on NEAR, intents.near, which holds deposits as ledger entries and releases tokens to other chains only on withdrawal [11]. NEAR's own front end, near.com, carried about a quarter of third-quarter volume, and since September 22 it has listed Ondo's tokenized US stocks and ETFs for users outside the United States and Canada [12][24]. Polosukhin said the same rails now carry non-financial orders placed by agents, a printed photo shipped by post among them [4]. DefiLlama counts $37.2 billion of cumulative Intents volume through November 9 and $3.73 billion in the latest 30 days [8].
The third is NEAR AI, which sells model inference run inside hardware enclaves, Intel TDX with NVIDIA GPU enclaves, and publishes attestation evidence with each response [14][15]. Its named users include Brave, Venice and the Government of Bermuda [14]. It operates as Jasnah Inc., with Polosukhin as chief executive, and publishes no revenue figure [4][15].
What the router pays the token
Intents fees have two layers. The protocol takes one pip, 0.0001%, on every transaction. The front end that brought the user adds a distribution fee: NEAR's own API charges 20 basis points by default and one basis point on stablecoin pairs, and splits anything a partner adds on top evenly [10]. Gross fees were $5.62 million in the 30 days to November 9, about 15 basis points of volume [8]. The rate has not compressed. It was 13.7 basis points in the 30 days to August 27 [8].
NEAR does not keep most of it. Solvers and third-party wallets keep their share, and what reaches NEAR's revenue wallets is the pip plus the cut from front ends NEAR runs. DefiLlama measures that captured revenue at $1.84 million for the same 30 days, 33% of gross fees, and $9.72 million since launch [8]. The fee switch went live in February, and since February 23 captured revenue has bought NEAR on the open market [8][13]. The purchases are not burned. The buyback account, buybacks.multisignature.near, held 2.25 million NEAR on November 11, liquid and unstaked [7].
Nansen's third-quarter report, built on NEAR's own data, runs somewhat higher than DefiLlama on most lines [12].
| Measure | Nansen | DefiLlama |
|---|---|---|
| Intents volume, third quarter 2026 | about $10.4 billion | $9.79 billion |
| Cumulative fees, early October | more than $52 million | $52.9 million |
| Captured by NEAR, cumulative, early October | more than $8 million | $7.5 million |
| Lifetime capture share, October 6 | 16.4% | 14.2% |
| 30-day capture share, October 6 | 28.4% | 25.4% |
This article uses DefiLlama's series for the arithmetic because it can be recomputed by date. Both sources show the same direction: NEAR's share of the router's fees has risen from about a sixth over the product's life to about a third in the latest month.
What the AI cloud pays the token
NEAR AI's catalogue is wider than its privacy claim. Its models page lists 63 models, and the homepage counts 11 as private and running in enclaves [15][27]. Forty-six are frontier models from Anthropic, Google, OpenAI, xAI and others, served through a router that hides the customer's account while the model runs at the provider, a tier the models page labels "No hardware privacy guarantee" [27]. The enclave tier is open-weight models priced per token in dollars, and since October 15 one of them, GLM 5.3 Flash, has answered Brave's Leo assistant from NEAR AI's enclaves [15][28].
None of that revenue has a published route to the token. Customers pay Jasnah Inc. in dollars by API key [15]. The Agent Market, where buyers hire agents by the job, pays out in dollars through Stripe or in USDC and takes a marketplace fee whose rate is not stated [30]. Polosukhin told Milk Road the market runs on the same intent rails as the router and that compute providers will later be paid in NEAR [4]. The Agent Market's own description mentions neither Intents nor the token [30].
The link that exists is staking. Since July 30 a holder can stake NEAR with NEAR AI and receive compute credits while the stake stays withdrawable [31]. By NEAR AI's description the credits are the yield, converted. Its example is $5 of agent-hosting credit a month for 500 staked NEAR, about 3.2% of principal a year at $3.75, and it describes the inference allowance as a function of the stake, the NEAR price and the staking APY [25][31]. Polosukhin pitched this as turning an AI bill into a balance-sheet asset [4]. Nansen counted more than 750,000 NEAR staked this way by the end of September, roughly 0.13% of all staked NEAR [7][12]. The pitch leaves out a dependency. The yield that pays for the compute is the issuance HSP-029 cuts.
The gap the vote was written for
At the 2.5% cap, a 1.31 billion supply mints about 89,800 NEAR a day [5][7]. At $3.75, that is roughly $10.1 million every 30 days against $1.84 million of captured Intents revenue, about $5.50 issued for each dollar the router returns [7][8][25]. Realized issuance runs slightly under the cap, 2.45% by the proposal's own measurement of supply growth [2].
HSP-029 moves one parameter. The cap falls linearly each epoch, about 3.75 basis points a month, starting one quarter after ratification, and the 90/10 split of rewards between stakers and the protocol treasury stays [2]. At the 1.6% endpoint, on today's supply and price, issuance would be about $6.5 million per 30 days, still 3.5 times current captured revenue. The cut removes 36% of the print. Revenue has to do the rest, and Polosukhin's own point makes that harder than it looks: buybacks are sized in dollars and issuance in tokens, so a higher NEAR price leaves the same revenue retiring fewer tokens against an unchanged print [4]. He described 1.6% as a two-year holding point while NEAR researches a fixed supply [4].
The cost falls on stakers, and on anyone drawing inference from staking yield. The proposal models gross staking yield dropping from 5.27% to 3.37% at the staking ratio measured in September, a 36% fall, and the treasury's tenth shrinks in step [2]. The 57% of supply that was unstaked at that snapshot gives up nothing [2].
The author's position is disclosed. Sal Ternullo runs SVRN, a Nasdaq-listed company holding about 55 million NEAR, more than 90% of it staked, which it puts at over 8% of all staked supply [2]. By its own estimate the cut costs SVRN roughly 855,000 NEAR a year in yield [2]. It is also a large holder of the asset whose dilution it proposed to reduce. In the forum thread a Bitwise representative backed the change; Bitwise's NEAR ETF, ticker NRR, has traded on NYSE Arca since September 29 and stakes its holdings, and Grayscale filed in January to convert its NEAR trust into an ETF [2][23][26]. Everstake, a validator operator, asked for more analysis first, objecting that the model holds the staking ratio constant, includes no pause if the validator set thins, and cuts treasury inflows by about 36% [2]. The active validator set was 406 at the proposal's September 22 snapshot and 351 on November 11 [2][7].
A vote that passed, after one that did not
NEAR has cut issuance before, and the manner of it is why the remaining steps matter. In 2025 a validator vote to halve the cap from 5% failed. Accounts differ on how: Brave New Coin said it needed 66.67% approval and put support at 45%, and at 91% among validators able to take part, while DL News reported more than 90% of participating validators in favor and a failure on quorum [19][20]. NEAR One shipped the halving anyway in nearcore 2.9.0, to activate once 80% of block-producing stake adopted it [18]. Chorus One declined to upgrade and called the route a "dangerous precedent" [19]. The cut took effect at the end of October 2025 [19]. HSP-029 records that halving as adopted at 68.41% approval, a figure neither document reconciles with NEAR One's 80% [2][18].
This time the tokenholder vote drew no vote against. It was also small. The 3.29 million veNEAR that voted compares with 582 million NEAR staked with validators [1][7]. veNEAR is locked NEAR carrying voting weight, so the units are close without being identical. The decision that counts is still the validators', and stake among them is concentrated: the nine largest held more than a third of it on November 11 [7].
The application screens, the chain does not
Polosukhin drew a line in the interview that the past six weeks tested from both sides. The chain, he said, is permissionless and deterministic. Intents is an application sitting between chains and cannot be, so it runs an off-chain monitoring system called Shield that can refuse funds and pause routes [4]. Nansen reports Shield rejected more than $50 million tied to the Bitget hack in late September [12]. On October 1 Intents itself lost about $3.8 million through a bug in how its contract interacted with the Omni bridge infrastructure. Services were halted, the contract patched, and the funds were reported returned within days [21][22].
The line showed again in November. Polosukhin said the Binance thieves had tried only about $100,000 through Intents [4]. Elliptic then traced about 117,000 NEAR, roughly $430,000, of the same theft through six meme tokens the launderers minted on NEAR itself between November 5 and 7 [17]. That is not a protocol flaw. A permissionless chain did what it does. The screening belongs to the router, which is also the part that earns the fees.
NEAR AI's test came four days after the recording, and its own posts set the terms. In February the company wrote of its enclaves, "The host OS, the GPU operator, and even NEAR AI are cryptographically locked out" [33]. On October 2 its head of product published a guide to checking that claim and conceded its limit: attestation "does not prove that the code has no bugs" [32]. On October 29 NEAR AI released two open-weight models with refusals removed for security testing, publicly available, and on Milk Road Polosukhin recommended them as the private place to do defensive work [4][29].
On November 8 an attacker used an unpatched flaw in SGLang, the open-source inference engine on NEAR AI's servers, to run code on one inference server and reach other infrastructure from it [16]. The company's report lists account details for about 100 users, chat history and settings from 18 agents on a legacy deployment outside the enclaves, and API keys for two cloud customers [16]. It also says requests in processing on that server could have been observed, confidential ones included, and draws the line itself: enclaves protect data from the operator and the host, not from a vulnerability in the software running inside them [16]. The report does not say the hardware was defeated. It says a bug inside the enclave was enough.
What would change the reading
Three dates and one number. The veto window closes November 16 [1]. Ratification then needs validators to adopt an upgrade, and given 2025, the share of stake that adopts it is the real vote [3][18]. Activation follows one quarter later [2]. The number is captured revenue against issuance, 18 cents on the dollar today, recomputable any day from DefiLlama and the chain. A second number would be new: any published figure for NEAR AI's revenue, or a rule sending part of it to the token.
For portfolio construction the structure matters more than the forecast. NEAR's token is asked to secure the chain, pay its gas, absorb the router's buybacks and meter AI compute, so a holder owns one claim on three businesses of which only one reports revenue [4][8]. Sagix's framework treats that as concentration risk whatever the growth rate: three operating bets and one issuance schedule expressed through a single asset.
Our earlier analysis of crypto as a downstream AI story left open which layer keeps the fee when the customer is a price-sensitive agent, and named the router as the place that question gets settled. NEAR's data adds two facts. The router's fee has held near 15 basis points, and a third of it reaches the token. The other two thirds stays with the wallets and solvers that own the user.
Polosukhin's instruction was to do the math. For the 30 days to November 9 it reads roughly $10 million issued, $1.84 million captured for buybacks, and $157,000 burned.
Educational purpose only: This content is provided exclusively for educational and historical research purposes. It should not be construed as investment advice, financial planning guidance, policy recommendations, or official economic analysis. Any contemporary parallels or policy discussions are presented as academic analysis, not recommendations for action. Historical patterns provide context for learning but do not predict future financial system outcomes or investment performance.
AI-assisted research disclosure: This analysis was researched and written with substantial assistance from artificial intelligence technology (Claude, Anthropic). While extensive efforts were made to verify all statistical claims, citations, and institutional analysis against authoritative sources, readers should independently verify any information before relying on it for academic, professional, investment, or policy purposes.
Accuracy and liability limitations: While extensive effort has been made to ensure accuracy through authoritative sources, the authors make no warranties about completeness, accuracy, or currency of information. Market interpretation involves judgment and debate. Data may contain revisions, measurement inconsistencies, or reporting variations across time periods and institutional sources. The authors and publisher assume no responsibility for errors, omissions, or consequences arising from the use of this information, including any errors resulting from AI assistance. Users assume full responsibility for decisions or actions taken based on this content.
Investment risk warning: This analysis does not constitute investment advice or recommendations. Past performance, whether historical or hypothetical, does not guarantee future results. All investments carry risk of loss. Readers should conduct their own research and consult qualified financial advisors before making investment decisions.
No professional relationship: This content does not create any professional, advisory, fiduciary, or client relationship between the reader and The Genesis Address Publishing LLC, its authors, or affiliated entities. Readers seeking financial, investment, legal, regulatory, or policy guidance should consult qualified professionals licensed in their jurisdiction.
Methodological note: This analysis synthesizes on-chain data queried from NEAR mainnet, governance records, protocol documentation, regulatory filings, third-party data aggregators, and established financial media. The numbered citation system allows readers to verify specific claims against original sources rather than relying on secondary interpretations.
References to modern financial systems, cryptocurrency protocols, or DeFi mechanisms are made for educational comparison purposes only. These comparisons do not constitute endorsements, recommendations, or predictions about the performance or suitability of any current financial products or services.
Publication information: Last updated: November 11, 2026 | Series: Protocol Analysis | Publisher: The Genesis Address Publishing LLC
Sources and references
[1] House of Stake. "HSP-029: Establish a 24-Month Issuance Ramp to a 1.6% Target Rate Of NEAR Token Emissions." Proposal and vote record. Submitted October 19, 2026; accessed November 11, 2026. https://gov.houseofstake.org/proposals/29
[2] Ternullo, Sal (SVRN). "HSP-029: Establish a 24-Month Issuance Ramp to a 1.6% Target Rate Of NEAR Token Emissions." NEAR Governance Forum, proposal text and replies. October 7, 2026. https://gov.near.org/t/hsp-029-establish-a-24-month-issuance-ramp-to-a-1-6-target-rate-of-near-token-emissions/42644
[3] Park, Danny. "NEAR governance approves proposal to cut token issuance ceiling to 1.6%." The Block. November 3, 2026. https://www.theblock.co/news/defi/2026-11-03-near-governance-approves-token-issuance-432378
[4] Milk Road. "NEAR's Plan to Power an Economy Run Entirely by AI." Interview with Illia Polosukhin, hosted by John Gillen. Recorded November 4, 2026; published November 6, 2026. Video. https://www.youtube.com/watch?v=xpou72TAJ-s
[5] NEAR Protocol. "Validators." Protocol documentation. Accessed November 11, 2026. https://docs.near.org/protocol/network/validators
[6] NEAR Protocol. "Blocks and Finality." Protocol documentation. Accessed November 11, 2026. https://docs.near.org/integrations/blocks-and-finality
[7] NEAR mainnet. JSON-RPC queries status, validators, block and view_account, at block height 224,069,383. November 11, 2026. https://rpc.mainnet.near.org
[8] DefiLlama. "NEAR Intents." Fees, revenue and volume series with methodology. Accessed November 11, 2026. https://defillama.com/protocol/near-intents
[9] DefiLlama. "Near." Chain fees series. Accessed November 11, 2026. https://defillama.com/chain/near
[10] NEAR Intents. "Fees." Documentation. Accessed November 11, 2026. https://docs.near-intents.org/resources/fees
[11] NEAR Intents. "Verifier Contract." Documentation. Accessed November 11, 2026. https://docs.near-intents.org/integration/verifier-contract/introduction
[12] Nansen. "NEAR Q3 2026 Report." October 20, 2026. https://nansen.ai/post/near-q3-2026-report
[13] NEAR. "Evolving NEAR Tokenomics." February 24, 2026. https://www.near.org/blog/evolving-near-tokenomics
[14] NEAR AI. "NEAR AI Cloud Integrates Intel Trust Authority: Independent Attestation for Confidential Inference." August 12, 2026. https://near.ai/blog/near-ai-cloud-integrates-intel-trust-authority
[15] NEAR AI. "Private AI inference you can verify." Product and pricing page. Accessed November 11, 2026. https://near.ai/
[16] NEAR AI. "Security Incident Report." November 9, 2026. https://near.ai/blog/security-incident-report-sglang-vulnerability
[17] Robinson, Tom. "North Korea is minting meme coins to launder the Binance hack." Elliptic. November 9, 2026. https://www.elliptic.co/insights/north-korea-meme-coins-binance-hack/
[18] NEAR One. "Enhancing NEAR Tokenomics: Version 2.9.0 Halving Upgrade." October 21, 2025. https://blog.nearone.org/announcement/2025/10/21/enhancing-near-tokenomics.html
[19] Luiv, Sven. "NEAR Protocol Cuts Inflation in Half Despite Failed Community Vote." Brave New Coin. October 31, 2025. https://bravenewcoin.com/insights/near-protocol-cuts-inflation-in-half-despite-failed-community-vote
[20] Avan-Nomayo, Osato. "Near blockchain's $2.8bn token economy faces inflation showdown." DL News. October 23, 2025. https://www.dlnews.com/articles/defi/near-blockchain-token-economy-faces-inflation-showdown/
[21] Baird, Kyle. "NEAR Intents halts services after $3.8 million exploit, promises full compensation." The Block. October 1, 2026. https://www.theblock.co/news/ecosystems/2026-10-01-near-intents-halts-services-after-3-8-million-exploit-promises-full-compensation-417404
[22] Gomez, Estefano. "NEAR Intents recovers $3.8 million from exploit and closes its investigation." Crypto Briefing. October 2, 2026. https://cryptobriefing.com/near-intents-recovers-exploit-funds/
[23] Shubnell, Jason. "Bitwise launches first US spot NEAR ETF with staking rewards." The Block. September 29, 2026. https://www.theblock.co/news/markets/2026-09-29-bitwise-near-etf-nrr-launch-staking-417150
[24] The Defiant. "Near Adds Ondo's Tokenized Stocks With Confidential Execution." September 2026. https://thedefiant.io/news/tradfi-and-fintech/near-com-adds-ondo-tokenized-stocks-confidential-execution
[25] CoinGecko. "NEAR Protocol." Price and supply data. Accessed November 11, 2026. https://www.coingecko.com/en/coins/near
[26] Grayscale Near Trust (NEAR). Form S-1 registration statement. U.S. Securities and Exchange Commission. January 20, 2026. https://www.sec.gov/Archives/edgar/data/2025000/000119312526016785/gsnr-20260120.htm
[27] NEAR AI. "Models for private AI inference, with live pricing." Models page. Accessed November 11, 2026. https://near.ai/models
[28] NEAR AI. "NEAR AI Cloud Brings Private Inference to Brave Leo." October 15, 2026. https://near.ai/blog/near-ai-cloud-brings-private-inference-to-brave-leo
[29] NEAR AI. "Cybersecurity Models and NEAR AI Red Launch." October 29, 2026. https://near.ai/blog/cybersecurity-models-and-near-ai-red-launch
[30] NEAR AI. "What Is Agent Market? The Marketplace Where AI Agents Get Hired by the Job." October 16, 2026. https://near.ai/blog/what-is-agent-market
[31] NEAR AI. "Announcing Staking for NEAR AI: Put Your NEAR to Work Powering Confidential AI." July 30, 2026. https://near.ai/blog/staking-for-near-ai
[32] Astretsov, Sergey. "Assume We're Lying: How to Verify Private Inference." NEAR AI. October 2, 2026. https://near.ai/blog/assume-were-lying-how-to-verify-private-ai
[33] NEAR AI. "NEAR AI Launches IronClaw, Confidential GPU Marketplace, and Multimodal Confidential Inference." February 23, 2026. https://near.ai/blog/near-ai-launches-ironclaw-confidential-gpu-marketplace-and-multimodal-confidential-inference
Publisher: The Genesis Address Publishing LLC
