The Hook
On July 31, 2025, NEAR Protocol announced what looked like a small feature: stake NEAR and receive monthly computing credits for AI models. The statement stressed that the user's principal is not consumed. Forty-three models are already accessible. The tone was matter-of-fact, almost celebratory. But the single most important sentence was the one nobody wrote: someone has to pay the model provider. I watched the silence break the noise of 2021, when a thousand tokens promised to be protocols and only a handful survived contact with real costs. This announcement has the same shape: a generous promise with an invisible balance sheet.
The Context
NEAR is not a stranger to infrastructure long games. It is a sharded proof-of-stake L1, led by engineers with a deep history in Rust and scalability research, and it has spent years building NEAR AI into a model aggregation layer. The July feature takes an old PoS mechanic and points it at a new consumer: users lock NEAR, receive a monthly credit allowance based on the staked amount, and spend that allowance across 43 models. The principal remains intact, and the user effectively gets access to AI services by parking capital rather than paying cash. This feels like a natural evolution of stake-for-security, but it is really a hybrid. NEAR is not running the models. It is almost certainly routing calls to Anthropic, OpenAI, Google, or similar providers through an API layer. The chain is responsible for the credit ledger, not the computation. That distinction is the whole analysis.
There is also a network context that makes this feature more consequential than a simple product update. NEAR operates with an inflationary supply schedule, so every staking reward is created by diluting all holders. If AI credits are quietly paid from staking rewards, then the feature is not merely a new way to use NEAR; it is a mechanism for transferring value from the entire NEAR community to a subset of AI users. That is a governance question disguised as a payment feature.
The Core
The first thing I ask when reviewing any staking feature is who bears the cost. Here, the user doesn't. The NEAR is locked, not burned. The model provider still needs cash, and cash is not generated by the act of locking. This is the hidden variable in the announcement. If credits are funded by staking rewards, then the NEAR inflation curve is subsidizing AI users. If NEAR AI or the Foundation pays the model providers directly, then this is a customer acquisition subsidy, not a sustainable fee channel. If there is an anticipated upgrade path where users eventually buy extra credits, then the announced feature is just a free trial with a deposit requirement.
The user's NEAR remains unconsumed, which means someone else is consuming real capital. A fee channel with no fees is not a fee channel; it's a subsidy wearing a payment rail.
Based on my audit experience reviewing staking contracts, the most dangerous line in an announcement is the one that promises to preserve principal. It shifts attention away from the cash flow gap. It creates the illusion that access is free. In NEAR's case, the credit system looks like a zero-liquidation CDP: the user deposits collateral, does not pay interest, and receives service credits instead. That is not a payment rail. It is a loyalty lock with an accounting quirk.
Also unresolved: whether the staked NEAR is self-staked or delegated. If delegated, the user inherits slashing risk. If merely locked, the staking yield assumption disappears and the credit formula becomes a pure creation of the foundation. Neither path is disclosed. The announcement withholds the exact conversion ratio, the contract audit status, and the procedure for cost settlement with model providers. Those are not minor omissions; they are the economic core.
Staking as a payment method only creates a durable token sink if the credits have a defined floor and the cost model has a defined source. Without those, this feature is a marketing artifact.
Regulators will read this the same way. If the user receives only service credits, the arrangement resembles a prepaid deposit and the securities case is weaker. If the user also earns staking APR, the service credit narrative collides with the expectation of profit. The announcement did not say which world we are in. That ambiguity is itself a compliance red flag.
There is also a hidden product layer. The credit window is monthly, which means the user must remain staked across an entire cycle to keep access. That is not a payment method; it is an attention lock. And if staked NEAR flows through liquid staking derivatives like stNEAR, the same tokens can simultaneously sit in DeFi and count toward AI credits. That might lift NEAR DeFi TVL, but it also creates a measurement problem: no one can tell whether the AI feature is driving demand or merely re-labeling existing locked supply.
I have seen this pattern before. In 2022, many projects confused token utility with cash flow. They asked users to deposit capital, promised access or rewards, and quietly assumed that real-world costs would be solved later. The ones that survived were the ones that disclosed the cost source before asking for trust. NEAR has not done that yet.
The Contrarian Read
Here is the contrarian read: the target audience is not AI developers. It is idle NEAR holders. The feature creates a psychological sink for dormant capital. By converting inactive NEAR into a quasi-subscription, NEAR can encourage lockups without launching a new token or promising yield. It is a lockup campaign wearing an AI coat.
The narrative shifted from 'decentralized inference' to a decentralized checkout counter, and that transition matters. The ETF didn't create a new asset class; it standardized the old one. NEAR AI doesn't decentralize AI; it standardizes a crypto wrapper around centralized APIs. The upstream dependency is the real structural weakness. Anthropic, OpenAI and Google can adjust pricing, change terms, or block resellers. If the top models demand better terms, the credit system either becomes more expensive or more subsidized. NEAR's position is not the maker of the road; it is a toll booth on someone else's highway. Short term, this can work. Long term, the model provider holds the most important pricing power.
There is another counter-intuitive risk: if NEAR's price drops while the user is staked, the effective cost of AI access rises even though no cash leaves the user's wallet. In fiat terms, the user is paying interest through opportunity cost and price risk. This is what makes the feature feel free and expensive at the same time.
For Web3 builders, the deeper trap is building a business on a credit system whose exchange rate can be changed by a foundation. Today the credits are generous. Tomorrow they may be diluted, recalibrated, or tied to a premium tier. The user is not a customer; the user is a counterparty to an unpublished contract.
Ethical Resonance
A product that cannot name its beneficiary cannot claim to be a public good. The ethical test here is simple: who becomes better off if this feature succeeds? If it is NEAR holders through higher lockup and a stronger AI narrative, then it is a legitimate marketing strategy. If it is a small group of AI users receiving subsidized inference at the expense of the broader staking community, then it is a hidden transfer. NEAR has not disclosed which one this is. In a market that is still healing from the trust failures of 2022, that silence matters more than the code.
The Takeaway
Watch for three numbers in the next quarter: monthly AI calls, incremental staked NEAR attributable to the feature, and a disclosed cost model. If none materialize, treat the announcement as a positioning exercise. The question I keep asking is not whether NEAR can pay for AI, but whether a blockchain belongs in the payment path at all. History doesn't reward the toll booth; it rewards the road. NEAR has built a shiny booth near the junction. The road still belongs to the model giants.