Solana pitches permissionless compute and machine payments for AI

The October 2 blog post frames Solana as a potential base layer for AI operations, making the case that speed, low fees and programmability are the properties software agents actually need from a settlement rail. The post covers the full stack: permissionless access to compute, training, identity, memory and machine-native payments.

The architecture is where the argument gets specific. Four layers: Compute, Model, Coordination and Application. Each, according to Solana, can be opened through on-chain markets, programs and identities rather than staying buried inside opaque cloud contracts or platform-controlled data stores. The post’s reliance on the “can be” construction throughout means the document is a design thesis rather than a deployment announcement.

Start at the Compute Layer. GPU-hours and inference capacity could be tokenized and traded on open markets, with Solana claiming the network is fast enough to run full order books on-chain so real-time pricing stays visible on the ledger. Trusted Execution Environments enter here too, enabling private, verifiable inference, with the blockchain as a public bulletin board for keys and attestations. The post does not specify an implementation schedule or current usage figures in the material reviewed.

The Model Layer covers permissionless training and what the post calls sovereign agents. Solana points to Nous Research’s Psyche network as an example of distributed model development. It also presents the ledger as an open dataset of economic behavior — raw material, in theory, for training financial models aimed at market making, execution or risk management.

The Coordination Layer is where the payment mechanism is most precisely specified. Solana’s post highlights x402, a machine payment protocol that revives the HTTP 402 Payment Required status code. The described capability: automated, sub-second settlement with sub-cent fees for per-request payments made by AI agents using stablecoins on Solana. Escrow is framed as conditional money — funds locked until deliverables are verified or released against milestones — giving agents a rules-based payment path; the post describes this as automated programmatic escrow, though it does not explicitly state that no human involvement is possible in every configuration.

Identity gets similarly concrete treatment. Solana wallets would function as machine-native identities: verifiable, portable, not revocable by platforms. Decentralized Autonomous Zones, abbreviated DAZ, would give agents persistent on-chain identity, memory and assets. Tokenized memory, per the post, could store an agent’s accumulated state as an asset owned by the agent or its principal — portable, composable, not stuck on some platform’s servers.

The blog supplies the design logic. The post does not specify an adoption timeline or a transaction-volume target in the material reviewed.

A related payment launch gives the pitch a live market reference. PivotNews.ai reported that Verona launched verUSD on September 28 across Ethereum, SSOL$119.75▲1.67%, Polygon, Avalanche, Optimism, Arbitrum and Celo for machine-to-machine and AI-agent transactions, claiming more than $100 million in institutional launch commitments. The $100 million represents committed participation.

CoinGecko lists SOL at $119.81, up 0.6% over the previous 24 hours.

The Solana blog describes a proposal. The document leaves the adoption question entirely unanswered.