On the eve of Anthropic's IPO, AI tokens worth paying attention to.

By: 0xFacai, Rhythm

The token VVV of Venice AI reached a new high today, once breaking through $25. The reason for this surge was that a mathematical research controversy made many people aware of the importance of "privacy inference."

In a public statement, New York University mathematician Tristan Buckmaster wrote that he and his collaborators had input the entire draft of the project into Codex. After learning that OpenAI's internal team had also made progress, he inquired whether their model had accessed these conversations or trained on them. The response he received was that the model did not consult user data, but his follow-up questions about training did not get answered.

OpenAI's response denied that researchers or agents accessed specific user data to solve problems, while acknowledging that, although unlikely, they could not exclude the possibility that de-identified data from private conversations with Codex had helped improve the model.

The community began to question whether these labs could see all of your work when the stakes are high enough and beat you to the results.

Trader based16z has expressed his trading opinion. He disclosed that he is long on VVV through spot, perpetual contracts, and over-the-counter call options with a strike price of $25. In his view, this incident widely increased awareness of the importance of privacy inference, and Venice is the most suitable liquidity asset to carry this narrative. He also compared VVV’s circulating market cap to ZEC when it was at $50. This reflects his judgment of the revaluation of privacy value.

As AI transitions from answering common sense questions to participating in papers, code, product development, and trading strategies, the content that users enter in the input box becomes more valuable.

Who can allow users to retain control over the content and execution process while using AI? Three Web3 projects have given their answers.

VVV: Turning Privacy into a Business

Venice AI, founded by Erik Voorhees, offers a chat application for consumers and an API for developers to call models. It aggregates different models, making privacy and fewer restrictions its product differentiators.

Venice AI's privacy is divided into three levels. Venice's "Anonymous Mode" hides the user's identity, while the upstream model can still see the request content. The "Zero Retention Mode" relies on service providers to fulfill their promises. The "TEE Mode" moves inference into a protected hardware environment, while the "End-to-End Encryption Mode" encrypts from the user's device until it decrypts upon entering the protected environment.

This business is already taking shape. Participating investor Banyan revealed that Venice's annual revenue rose from $14 million in January to over $100 million in August.

VVV is the token issued by Venice on Base, where every $100 purchase of Venice's API credits results in $5 being used to buy and burn VVV. On the supply side, the new emissions of VVV tokens are also decreasing, with the annual emission dropping from 3 million tokens on September 1 to 2.5 million tokens, and it is planned to further decrease to 2 million tokens by October 1.

The other source of demand for VVV is DIEM. Holders can lock their staked VVV to mint DIEM, and by staking 1 DIEM, they can receive an updated $1 API credit daily. This allows developers and agents to hold an asset that continuously generates usage credits, preparing for future model calls.

On September 14, the target supply of DIEM will complete a phased expansion from 38,000 to 40,000, making space for new minting. This move also suggests the Venice team's optimistic attitude toward user expansion.

The logic for being bullish on VVV is very clear: under the privacy threats in AI, Venice has found its market position, with more people paying to use Venice, leading to more buybacks; more people needing continuous inference credits lead to a demand for locking assets.

NEAR: Verifiable Privacy Inference

Along the clues from Venice AI, we also discover a familiar figure: NEAR. This public chain is expanding the use of the NEAR token through confidential computing and agent services.

In March of this year, Venice announced integration with NEAR AI, allowing users to choose verifiable privacy inference services provided by NEAR AI. Consumers initiate requests on Venice, while the underlying privacy computing capabilities are provided by NEAR AI and other service providers.

The core capability of NEAR AI Cloud is to run models in a TEE, which is a trusted execution environment isolated by hardware. As designed, the plaintext during computation is restricted to a protected area, preventing infrastructure operators from reading it directly; users can verify hardware attestations to confirm that requests have entered the corresponding environment. This offers cloud computing options for teams wishing to protect their research and commercial data.

Open weight models can be deployed in this environment. When calling closed-source models like Claude, GPT, and Gemini through gateways, requests are still sent to upstream service providers, and NEAR's confidential computing cannot cover the other server. The value of NEAR along this route comes from the ability to protect the computing process and organize model services.

This capability has already established a connection with the NEAR token. The staking pay introduced on July 30 allows token holders to convert their NEAR staking rewards into inference credits, with the credits varying with the amount staked, coin price, and yield. Users retain ownership of the underlying tokens and can unstake and exit. For teams that consistently call models, this adds another use case for holding NEAR.

Another opportunity within NEAR lies in payments. To complete tasks, AI agents need to call models, purchase data, pay for services, and mobilize assets across different chains. The intent execution method provided by NEAR Intents allows users to submit desired outcomes, which are then competed for completion and execution by solvers. This infrastructure can integrate complex cross-chain operations into the AI agent's task processes.

According to DeFillama statistics, NEAR Intents generated approximately $9.32 million in total fees in the second quarter of this year, with about $1.5 million of that retained by the protocol; the retained revenue is used for market buybacks of NEAR.

Therefore, NEAR's bullish narrative is supported by two business areas: providing computation for sensitive tasks and execution and payment for cross-chain tasks. The former can reach users alongside applications like Venice, while the latter has the opportunity to grow with AI agents.

TAO: Open Supply AI Models

Bittensor has always been the most attention-grabbing AI project in Crypto. TAO is the native token of the Bittensor network.

Bittensor organizes different tasks into independent subnets, with miners providing inference, storage, prediction, and other services, and validators evaluating service quality, with rewards distributed according to rules. A team can organize competition around a specific demand, while the entire network accommodates multiple such markets.

As AI demand expands, applications need more interchangeable suppliers, and small teams need to find customers, computing resources, and funding. Bittensor attempts to organize this supply through a public incentive market, allowing different teams to compete in specific tasks.

Some subnets have already begun to receive external revenue. For example, Chutes, which provides model inference services, generated approximately $1.37 million in revenue in the second quarter of this year, sourced from subscriptions, pay-per-use, and instance services.

How does TAO accommodate this growth? Each subnet has its own alpha token, paired with TAO to form a trading pool. Staking TAO to a subnet effectively exchanges it for the corresponding alpha token; thus, TAO serves as the foundational asset for inter-subnet capital allocation. If more competitive services appear in the network, there will be opportunities to expand demand in these markets.

On the supply side, TAO retains a familiar scarcity design from the crypto market. The total cap of TAO is set at 21 million tokens, with the first halving set to complete in December 2025; currently, 0.5 tokens are issued per block, amounting to about 3,600 tokens daily.

The explosive rise of VVV provides us with a window of observation. As models become stronger, the sensitive information people entrust to them increases, and "encrypted/privacy AI" has naturally found its product-market fit (PMF).