What is NEAR Protocol (NEAR)?

NEAR launched in 2020 as a sharded proof-of-stake chain with an unusual emphasis on usability. Accounts are human-readable — alice.near, not a 42-character hex string. Contracts can pay their users' gas, so an application can onboard someone who has never held a token. Nightshade shards the chain so capacity grows by adding shards rather than by demanding more of each validator.

These were good ideas that did not win the Layer-1 race. By 2024 NEAR had solid technology, a reasonable ecosystem and no clear reason for a developer to choose it over Solana or an Ethereum rollup. What it has done since is more interesting than most chains in that position attempt: it largely stopped competing as a Layer 1.

NEAR Intents

Chain abstraction is the idea that users should say what they want, not which chain it happens on. NEAR Intents lets someone swap assets across chains without bridging, holding multiple wallets or knowing where anything settles — solvers compete to fulfil the request and NEAR coordinates the settlement.

It has processed over $24 billion in cumulative swap volume, which makes it one of the few chain-abstraction products with real usage rather than a whitepaper. Note carefully what that figure is: volume routed, not revenue earned, and the gap between those two is where the token question lives.

The AI turn

NEAR's second bet is AI infrastructure, and specifically confidential inference — running models where the operator cannot see the prompt or the output. On 30 July 2026 NEAR AI introduced staking-based payments for these services: stake tokens, receive credits, spend them on private inference and hosting. More than 500,000 NEAR is now staked for this purpose, unlocking access to over 40 models.

The mechanism is genuinely novel — staking as prepayment for compute rather than as security — and the scale is small. Both facts matter, and coverage of NEAR tends to report only the first.

• Gas for transactions, with 30% of each fee rebated to the contract that was called — a direct developer subsidy built into the protocol.
• Staking to validators, which secures the sharded chain.
• Staking as prepayment for AI compute credits, the mechanism introduced in July 2026.
• Storage staking: contracts lock NEAR proportional to the state they occupy, which removes tokens from circulation as usage grows.
• Governance and account registration.

The fee rebate deserves a mention because almost no other chain does it. Thirty per cent of the gas a user pays goes to the contract they called, so a popular application earns revenue from its own traffic without charging anything extra. It is a real and underused piece of protocol design.

NEAR has no hard cap. It issues 5% annually, of which 70% pays validators and 30% goes to the protocol treasury. All transaction fees are burned apart from the 30% rebated to contracts, so net issuance falls as the chain gets busier.

At current activity the burn does not come close to offsetting 5% issuance, so NEAR is meaningfully inflationary. That is the plain reading and it should not be softened: unless usage rises very substantially, holders are diluted at a rate that staking barely compensates for.

Storage staking as a supply sink

Contracts must lock NEAR in proportion to the on-chain state they use. It is a sensible answer to state bloat — charging for storage rather than only for computation — and it doubles as a demand source that scales with real usage rather than with speculation. It is small today and it is the kind of mechanism that matters if adoption arrives.

Early allocations to the team, foundation and backers vested over the years following the 2020 launch and are largely behind the market now.

Staking NEAR means delegating to a validator, funded from the 70% of issuance allocated to them. Because issuance is fixed at 5% and split by stake, the effective yield depends on how much of the supply is staked — more stakers, lower rate each.

Unstaking takes a few epochs, on the order of two to three days. Slashing exists in the protocol for provable misbehaviour but has not been a material practical risk for delegators.

The AI staking introduced in July 2026 is a different thing wearing the same word, and conflating them would be a mistake. That is not securing the network — it is locking tokens to buy compute credits for confidential inference and model hosting. It pays no yield. It buys a service.

Two pivots in three years

NEAR has repositioned from general-purpose Layer 1, to chain abstraction, to AI infrastructure. Each move was a reasonable response to losing the previous fight, and a project that changes its central story repeatedly is harder to value than one that does not — because you are underwriting a team's judgement rather than a thesis.

Volume is not revenue

$24bn of cumulative Intents volume is real activity. What NEAR earns from it, and how much of that reaches the token, are different and much smaller numbers. Chain abstraction is a thin-margin routing business by nature, and thin margins on somebody else's liquidity is a difficult place to build token value.

The AI market is not short of competitors

Confidential inference is a genuine need and NEAR is a small entrant against well-funded crypto-AI projects and, more importantly, against conventional cloud providers building the same capability with hardware enclaves. 500,000 NEAR staked and 40 models is an early product, and it should be read as one.

Persistent inflation

Five per cent annual issuance against a burn that does not offset it means holders who do not stake are diluted, and stakers are roughly treading water. This only resolves through a large increase in activity.

Sharding complexity

Nightshade is one of the few sharded designs running in production, which is an engineering achievement and a larger surface area than a single-shard chain. Cross-shard interaction is where this class of system historically breaks.