NEAR Protocol started life as an AI startup, became a blockchain, and is now pitching itself as the network for user-owned AI. In November 2026 its community also voted to cut new token issuance again.
This guide explains what NEAR AI actually builds, how chain abstraction and intents fit in, what the NEAR token does, and the risks you should know about.
- What NEAR is: a proof-of-stake, sharded blockchain founded by Illia Polosukhin (a co-author of the Transformer paper) and Alexander Skidanov.
- The AI pivot: NEAR AI builds private, verifiable AI, including NEAR AI Cloud (inference inside secure hardware) and IronClaw (an agent runtime).
- Chain abstraction: Chain Signatures and NEAR Intents let one NEAR account act on Bitcoin, Ethereum and other chains.
- Token role: NEAR pays for gas and storage, secures the network through staking, and is used in governance.
- 2026 news: proposal HSP-029, approved on November 2, 2026, would cut max yearly issuance from 2.5% to 1.6% over 24 months.
What NEAR Protocol Is
NEAR is a layer-1 blockchain, which means it runs its own network rather than sitting on top of another chain. Smart contracts on NEAR are usually written in Rust or JavaScript.
- Founders: Illia Polosukhin, a former Google Research engineering manager, and Alexander Skidanov, formerly at MemSQL.
- Origins: the project began as Near.ai, an AI startup working on program synthesis, before it switched to blockchain.
- Mainnet: launched in April 2020.
- Consensus: proof-of-stake, where validators lock up NEAR to produce blocks.
- Scaling: a sharding design called Nightshade, which splits work across parallel shards.
Polosukhin is one of the eight authors of the 2017 paper “Attention Is All You Need.” That paper introduced the Transformer, the design behind ChatGPT, Claude and Gemini, and it is a big part of why NEAR’s AI story gets attention.
NEAR’s AI Pivot and “User-Owned AI”
From 2024 onward, NEAR began describing itself as the blockchain for AI. Its slogan is user-owned AI.
The idea is simple to state. Today, a few large companies own the best AI models and see the data you send them. NEAR argues that AI should instead be:
- Private: your prompts and data stay hidden, even from the company running the servers.
- Verifiable: you can check which model ran and that nobody tampered with it.
- Owned by users: people control their own data, AI assistants and digital assets.
The blockchain’s job in this picture is to hold identity, payments and agreements. The private computing happens in secure hardware, with the blockchain recording and settling the results.
Note: “User-owned AI” is NEAR’s own marketing term. It describes a goal and a set of products, not a finished system that replaces big AI labs.
What NEAR AI Actually Builds
NEAR AI is the AI research and product arm led by Polosukhin. Its own site lists these main products.
| Product | What it does | Launched |
|---|---|---|
| NEAR AI Cloud | Runs open AI models inside hardware enclaves (TEEs), so data stays encrypted during use, with proof you can check | Late 2025 |
| IronClaw | Open-source Rust runtime for AI agents that runs inside TEEs and keeps secrets away from the model | February 2026 (NEARCON) |
| Confidential GPU Marketplace | Lets GPU owners rent idle capacity for encrypted AI jobs, with hardware-signed proof | February 2026 |
| Multimodal confidential inference | Text, image and voice models in one private environment | February 2026 |
| NEAR AI Agent Market | A marketplace for AI agents | Q1 2026 (NEARCON) |
What a TEE is, in plain English
A trusted execution environment is a locked area inside a processor. Code and data inside it are encrypted, and even the server owner cannot read them.
NEAR AI uses Intel TDX and NVIDIA confidential computing for this. The chip can also produce an “attestation,” a signed report that proves which code is running.
Who uses it
NEAR AI says Brave’s Leo assistant now runs private inference on NEAR AI Cloud. Nansen’s Q4 2025 report also listed Brave Nightly and OpenMind as early production users.
Chain Abstraction, Chain Signatures and Intents
Chain abstraction is NEAR’s plan to hide blockchain complexity from users. You should not need a different wallet, token and app for every chain.
Chain Signatures
Chain Signatures let a NEAR account, or a NEAR smart contract, sign transactions on other blockchains. It works through multi-party computation (MPC).
- Shared signing: many NEAR nodes each create a share of a signature. No single node can sign alone.
- Derived addresses: a NEAR account plus a path like
ethereum-1creates a matching address on another chain. - Supported chains: NEAR’s docs show Bitcoin, Ethereum, Base, Solana, XRP, Sui, Aptos, Cosmos and more.
NEAR Intents
An intent is a request that says what you want, not how to do it. For example: “swap my BTC for the best price in USDC.”
- Submit. A user or an AI agent states the outcome they want.
- Compete. “Solvers,” which include market makers and AI agents, compete to fill the request.
- Execute. The best offer runs automatically, even across several chains.
Nansen reported that NEAR Intents passed $7 billion in cumulative volume by the end of 2025, across more than 13 million swaps and 25+ blockchains.
How AI Agents Use NEAR
This is where the AI and blockchain parts connect. An AI agent that books travel or trades assets needs three things that a normal chatbot lacks.
- A wallet: one NEAR account can hold and move assets on many chains through Chain Signatures.
- A simple way to act: the agent submits an intent, and solvers handle the hard routing.
- Safe secrets: IronClaw keeps keys in an encrypted vault, so a tricked model cannot leak them.
One example from Nansen’s report is TravAI, a travel platform where AI agents handle multi-step bookings that settle through NEAR Intents.
If you want the bigger picture of AI agents and crypto, my AI crypto tokens explained guide compares NEAR with other AI-focused networks.
What the NEAR Token Does
NEAR’s docs list three core uses for the token, and governance adds a fourth.
| Use | How it works |
|---|---|
| Transaction fees (gas) | Every transaction uses gas, paid in NEAR by the sender |
| Storage staking | Accounts lock NEAR to cover the data they store; deleting data frees it |
| Validator staking | Holders delegate NEAR to validators and share in staking rewards |
| Governance | Locked NEAR (veNEAR) gives voting power in the House of Stake |
Supply and issuance
- Circulating supply: about 1.31 billion NEAR in early November 2026, according to market data cited in the issuance news.
- Issuance split: new tokens go 90% to stakers and 10% to the treasury.
- October 2025: the “Halving Upgrade” cut maximum yearly issuance from 5% to 2.5%.
- November 2026: a new vote aims to lower it again to 1.6% (details below).
Warning: Lower issuance changes staking yields too. Reports modelled gross staking yield falling from about 5.27% to about 3.37% if the 1.6% target is reached and the staking ratio stays the same.
NEAR News in 2026
Here are the main verified events from this year, in date order.
- January 2026: Grayscale filed to list its NEAR product on NYSE Arca, following a Bitwise ETF filing in December 2025.
- February 23, 2026: at NEARCON, NEAR AI launched IronClaw, the Confidential GPU Marketplace and multimodal confidential inference.
- April 16, 2026: at BUIDL Asia in Seoul, Polosukhin described NEAR’s privacy stack as a mix of TEEs and MPC.
- November 2, 2026: the House of Stake approved HSP-029 with about 99.97% support, according to BTC-ECHO.
- November 6, 2026: NEAR announced Delta League, an open AI model competition judged on a hidden test set.
What HSP-029 actually changes
The proposal would lower maximum yearly issuance from 2.5% to 1.6% over 24 months. Modelled savings are about 11.8 million fewer NEAR issued in two years and 66.1 million over six years.
It is not final yet. It still needs a Security Council review, a grace period of about 90 days, and validators to adopt a protocol upgrade.
Risks to Understand
Any honest look at NEAR has to cover the downsides. These are the main ones.
- Price volatility: NEAR’s price can swing sharply in days, and it can fall even after good news.
- Competition: many other networks, such as Bittensor, Render and Akash, also chase AI demand.
- Adoption gap: AI products like NEAR AI Cloud do not always need the NEAR token, so product success may not lift token demand.
- Centralization concerns: a 2023 Messari study estimated about 35% of staked NEAR ran on AWS.
- Security history: the Skyward Finance launchpad lost about $3 million to an exploit in 2022, and a hijacked Discord account pushed a fake airdrop in 2023.
- Governance uncertainty: votes can pass and still change before they take effect, as HSP-029 shows.
- Regulation: rules for crypto and for ETF products differ by country and keep changing.
For two of the AI rivals, see my guides to Bittensor (TAO) and decentralized GPU networks like Render, Akash and io.net.
How to Research NEAR Safely
Most crypto losses come from scams, not from bad analysis. Use this checklist before you trust any claim about NEAR.
- Start with official sources. Read near.org, near.ai and docs.near.org directly. Type the address yourself rather than clicking links in messages.
- Check on-chain data. Use a block explorer such as NearBlocks to confirm transactions, contracts and supply figures you see quoted.
- Read the governance forum. Proposals like HSP-029 are public. Read the full text, not a summary on social media.
- Look at the code. The nearcore repository on GitHub shows how active development is and what each release changes.
- Compare several reports. Treat one article, including this one, as a starting point. Check numbers against a second independent source.
- Ignore “airdrop” and “support” DMs. Real teams never ask for your seed phrase or ask you to “verify” a wallet.
- Write down the risks. List what could go wrong before you think about upside. If you cannot name three risks, keep researching.
Tip: An AI assistant can summarize a long governance proposal for you, but always check its summary against the original text. Models can get numbers wrong.
Common Problems and Fixes When Researching NEAR
- Two different “NEAR AI” sites: near.ai is NEAR’s AI company. Check the domain carefully, because lookalike sites are common.
- Old numbers in articles: inflation was 5% before October 2025 and 2.5% after. Always check the date on any tokenomics claim.
- Confusing intents with bridges: an intent states a goal, and solvers fill it. A bridge moves a token from one chain to another.
- Explorer shows an unfamiliar account name: NEAR uses readable names like
alice.nearand long hex “implicit” accounts. Both are normal. - Conflicting throughput claims: some articles quote very high transactions-per-second figures. Look for the original benchmark and its conditions before you trust a number.
- A “news” post promises returns: that is marketing or a scam. Real project updates describe changes, not profits.
Frequently Asked Questions
What is NEAR Protocol used for?
NEAR is a proof-of-stake blockchain for apps and smart contracts. Its current focus is chain abstraction and private, verifiable AI through NEAR AI.
What does “user-owned AI” mean?
It is NEAR’s term for AI that keeps user data private, can be verified, and is controlled by users rather than a single company. It is a goal and a product direction, not a finished system.
What is NEAR AI Cloud?
It is a service that runs open AI models inside secure hardware enclaves. Data stays encrypted during use, and users get a signed attestation of what ran.
What are NEAR Intents?
Intents let a user or AI agent say what outcome they want, such as a swap at the best price. Solvers compete to fill it, even across several blockchains.
Did NEAR cut its inflation?
Yes. Maximum yearly issuance fell from 5% to 2.5% in October 2025. Proposal HSP-029, approved on November 2, 2026, would lower it to 1.6% over 24 months once implemented.
Who founded NEAR?
Illia Polosukhin and Alexander Skidanov. Polosukhin co-authored “Attention Is All You Need,” the 2017 paper that introduced the Transformer.
Is NEAR a good investment?
This guide does not give investment advice or price predictions. Read the risks section, use official sources, and speak to a licensed financial adviser.
Sources
- NEAR AI: Company and products
- NEAR AI: IronClaw, Confidential GPU Marketplace and multimodal inference
- NEAR Docs: What is chain abstraction
- NEAR Docs: Chain Signatures
- NEAR Docs: Tokens
- Nansen: NEAR Quarterly Report Q4 2025
- Ad-hoc News: NEAR issuance vote HSP-029
- Invezz: NEAR halving upgrade cuts inflation to 2.5%
- Autheo: NEAR privacy and AI agent infrastructure (BUIDL Asia 2026)
- TokenPost: NEAR announces Delta League
- Wikipedia: NEAR (blockchain platform)
