NEAR Just Turned Staking Yield Into Compute

NEAR AI is positioned as a confidential, verifiable inference platform: encryption, trusted execution environments, and attestations can identify the model, prompt, output, hardware, and provenance, while on-chain verification is required when AI outputs affect smart-contract execution and money may be at stake.

The episode frames AI infrastructure intermediaries as a major security and governance risk. Routers may rewrite prompts, censor users, alter outputs, or return malicious tool outputs, while the guest argues that a system-prompt change in a closed model could steer outputs seen by billions of users and potentially influence political preferences.

NEAR's proposed economic model ties inference access to foregone staking yield rather than a separate recurring bill, with capacity proportional to NEAR staked and provider compensation distributed through staking. The guest connects this structure to autonomous businesses that need 24/7 intelligence, financial access, and execution without exhausting their treasuries.

NEAR is initially using its own GPUs while developing a confidential marketplace where third-party providers can verify hardware and supply compute without seeing user data. That marketplace is intended to make GPU hours a liquid, tradable asset despite differences in capability and latency and the fact that compute generally cannot be stored.

NEAR Intents is presented as a mechanism for sourcing underspecified compute requirements that do not fit a conventional order book: buyers can request approximate capacity such as 1,000 GPUs with specified capabilities, while solvers locate hardware, negotiate pricing, and settle confidentially and verifiably. The host says approximately 20% to 50% of NEAR emissions had been captured and burned, attributing that activity to NEAR Intents and the blockchain.