Venice AI token VVV extended its rally on Tuesday, briefly rising above $25 to a fresh high. The move followed a dispute in academic research that pushed the idea of “privacy inference” into the middle of crypto market discussion.
New York University mathematician Tristan Buckmaster said in a public statement that he and his collaborators had entered the full draft of a project into Codex. After learning that an internal OpenAI team had also made progress on the same problem, he asked whether the model had been exposed to those conversations or trained on them. He said he was told the model had not reviewed user data, but he did not receive an answer to his follow-up question about training.
OpenAI said in response that neither researchers nor agents accessed specific user data to solve the problem. At the same time, it acknowledged that while unlikely, it could not fully rule out the possibility that de-identified private conversations between the research group and Codex had contributed to model improvement. The episode triggered a broader question across the community: when the stakes are high enough, can AI labs see users’ work and ship results first?
Trader based16z has already put on a position around that view. He disclosed a long VVV trade expressed through spot, perpetual futures, and over-the-counter call options with a $25 strike. In his view, the incident made the importance of privacy inference clear to a much wider audience, and Venice is the most suitable liquid asset for that narrative. He also compared VVV’s circulating market cap with ZEC when ZEC traded at $50, framing the trade as a repricing of privacy value.
As AI moves from answering general questions into research papers, code, product development, and trading strategies, the content users type into prompts carries more value. The question is no longer abstract. Which platforms let people use AI while retaining control over both their work and the execution process? The original article points to three Web3 answers: Venice, NEAR, and TAO.
VVV: Venice turns privacy into a business model
Founded by Erik Voorhees, Venice AI offers a chat application for consumers and model APIs for developers. It aggregates multiple models and uses privacy and lighter restrictions as product differentiators.
Venice describes several layers of privacy. Its anonymous mode hides user identity, though upstream models can still see the request itself. A zero-retention mode depends on the provider honoring a commitment not to keep data. TEE mode places inference inside protected hardware. End-to-end encryption mode starts encryption on the user device and keeps data encrypted until it reaches the protected environment, where it is decrypted.
The business has already reached some scale. Investor Banyan said Venice’s annualized revenue rose from $14 million in January to more than $100 million in August.
VVV is Venice’s token on Base. Under the project’s design, for every $100 of Venice API credits purchased, $5 is used to buy back and burn VVV. On the supply side, emissions are declining. Annual issuance fell from 3 million tokens to 2.5 million on Sept. 1, and is scheduled to fall again to 2 million on Oct. 1.
Another source of demand comes from DIEM. Holders can lock staked VVV to mint DIEM. Staking 1 DIEM gives the holder a daily $1 API credit that updates over time. In practice, that gives developers and agents an asset that continuously generates usage credits and can be accumulated ahead of future model calls.
On Sept. 14, DIEM’s target supply is set to expand in stages from 38,000 to 40,000, creating room for additional minting. The original article said the change also hints at the Venice team’s optimism about user growth.
The bull case outlined for VVV is straightforward. If privacy threats in AI push more users toward Venice, paid usage can translate into more buybacks. If more users want recurring inference credits, lockup demand can rise as well.
NEAR: verifiable private inference and execution rails
Following the Venice thread leads to a familiar name in crypto AI: NEAR. The blockchain is trying to broaden NEAR token utility through confidential computing and agent services.
In March, Venice and NEAR AI announced an integration. That allows users to choose verifiable private inference provided by NEAR AI when making requests through Venice. The consumer-facing application sits at Venice, while privacy-preserving compute can be handled by NEAR AI and similar service providers underneath.
The core of NEAR AI Cloud is running models inside a trusted execution environment, or TEE, isolated by hardware. Under that setup, plaintext used in computation stays inside the protected zone, which infrastructure operators cannot directly read. Users can also verify attestation from the hardware to confirm their requests entered the intended environment. That gives teams handling research data or commercial data a cloud option with stronger privacy guarantees.
Open-weight models can be deployed in this environment. But requests routed through a gateway to closed-source models such as Claude, GPT, or Gemini still end up with the upstream provider, and NEAR’s confidential computing does not extend to those external servers. In the article’s framing, NEAR’s value on this path comes from protecting the computation process and coordinating model services.
That capability is already being linked to the token. A staking payments feature launched on July 30 allows holders to convert NEAR staking yield into inference credits, with the credit amount changing according to stake size, token price, and yield. Users keep ownership of the underlying NEAR and can unstake to exit. For teams that call models over long periods, this adds another reason to hold the token.
NEAR also has an opening on the payments side. To complete tasks, AI agents may need more than model calls: they may have to buy data, pay for services, and move assets across chains. NEAR Intents offers an intent-based execution model in which users submit the desired outcome and solvers compete to complete the swap and execution. That infrastructure can connect complex cross-chain operations to AI agent workflows.
According to DeFiLlama, NEAR Intents generated about $9.32 million in total fees in the second quarter of this year, with roughly $1.5 million retained by the protocol. That retained revenue is used for market buybacks of NEAR.
The article reduces NEAR’s crypto AI case to two business lines: compute for sensitive tasks, and execution plus payments for cross-chain tasks. The first can reach users through applications such as Venice. The second may grow alongside AI agents.
TAO: organizing open AI supply through subnets
Bittensor remains one of the most closely watched AI projects in crypto, and TAO is the network’s native token.
Bittensor organizes different tasks into separate subnets. Miners provide services such as inference, storage, and prediction. Validators score the quality of those services, and the network distributes rewards under protocol rules. A team can form a market around a specific need, while the broader network hosts many such markets at once.
As AI demand expands, applications need more interchangeable suppliers, while smaller teams need customers, compute resources, and capital. Bittensor is trying to organize that supply through open incentive markets so teams can compete on concrete tasks.
Some subnets have already begun producing outside revenue. The article cites Chutes, a model inference service provider, as an example. It generated about $1.37 million in revenue in the second quarter from subscriptions, usage-based billing, and instance services.
How does TAO absorb that growth? Each subnet has its own alpha token, paired with TAO in liquidity pools. When users stake TAO into a subnet, the position is effectively converted into that subnet’s alpha token. In that sense, TAO acts as the base asset for capital allocation across subnets. If more competitive services emerge on the network, demand to participate in those subnet markets could grow with them.
On the supply side, TAO keeps a scarcity design familiar to crypto markets. Its total supply is capped at 21 million tokens. The network completed its first halving in December 2025. Current issuance stands at 0.5 TAO per block, or about 3,600 tokens a day.
From VVV’s rally to a wider privacy-AI trade
VVV’s sharp move has given the market a fresh observation point. As models become more capable, the sensitive information people hand to them also becomes more valuable. In that setting, “crypto/privacy AI” is being treated by more market participants as a category that has found product-market fit.
Venice is selling private inference directly to users. NEAR is supplying confidential compute and execution rails. Bittensor is structuring open AI supply through subnet markets. VVV’s rally did not create those models, but it pushed all three lines of the story back into view at the same time.