Solana gets pay‑as‑you‑go inference
Salad launched an AI Gateway on Solana that offers lean, pay‑as‑you‑go inference access paid with $RENDER tokens, aiming to lower commitment friction for prototyping and scaling agents. The marketplace‑style access model targets teams that want metered inference without long contracts. (x.com)
AI inference is the step where a model answers a prompt, and Salad is now selling that access on Solana through a gateway that takes RENDER tokens. (docs.salad.com, x.com) Salad’s own documentation says the AI Gateway exposes open-source models through a single OpenAI-compatible endpoint at `ai.salad.cloud/v1`, with “no container groups, no replicas, no cold starts,” and lists the product as being in closed beta as of April 2, 2026. (docs.salad.com, docs.salad.com) The Solana angle is the payment rail. The launch post said usage is paid in RENDER, the Solana-based token used across the Render ecosystem after its migration from Ethereum’s RNDR token. (x.com, know.rendernetwork.com) Render’s knowledge base says the community-approved move created RENDER on Solana, with a 1:1 upgrade path from RNDR and a stated focus on the Solana token going forward. (know.rendernetwork.com) Salad’s pitch is that developers can buy model calls the way they buy electricity: by usage instead of by reserving machines. Its docs contrast the gateway with running your own containerized inference stack, which normally means choosing GPUs, sizing video memory, managing replicas, and handling traffic. (docs.salad.com, docs.salad.com) That fits the way Solana has been courting AI builders. Solana’s AI pages now frame the chain as infrastructure for agents, data, compute marketplaces, and instant payments, with tools aimed at developers building autonomous apps onchain. (solana.com, solana.com) Underneath the gateway is Salad’s broader business: a distributed cloud built from idle consumer machines, mostly gaming PCs, that the company says can be assigned workloads when their owners are not using them. Salad’s docs describe the model as a marketplace for spare compute, and its quickstart materials say workloads can scale across a global network of idle GPUs. (docs.salad.com, docs.salad.com) That marketplace structure is not unique to Salad. Solana’s own AI page highlights other compute and inference networks, including Inference.net, which Solana says processes more than 1 trillion inference tokens across more than 5,000 active GPUs. (solana.com) The practical tradeoff is clear in Salad’s documentation. Teams that want full control can still deploy dedicated containers, while teams that want a simpler meter can use the gateway and skip most of the infrastructure work. (docs.salad.com, docs.salad.com) For Solana developers building agents, the launch adds one more piece to the stack: a way to pay for model responses with a Solana token instead of signing a longer cloud contract first. (x.com, solana.com)
