Google and Solana launch Pay.sh

Software payments are getting rebuilt for machines. That is the real story here. Solana Foundation said on May 5 that it launched Pay.sh with Google Cloud, a gateway that lets AI agents discover APIs, pay per request with stablecoins on Solana, and get access without the usual account setup, API keys, or subscriptions. ### What is Pay.sh, exactly? Pay.sh is basically a payments-and-access layer that sits between an AI agent and a paid API. Instead of a human opening an account, adding a card, agreeing to a plan, and storing credentials, the agent uses a Solana wallet, gets a live price, pays, and the request goes through. Solana says Cloud Run, and Vertex AI. ### Why is that a new thing? Because most “agentic” software still hits a very human bottleneck at the moment money enters the loop. Models can search, summarize, route tasks, and call tools — but paid tools usually assume a person is managing billing relationships and credentials. Pay.sh is trying to swap that model out for pay-per-call access where the payment itself becomes the credential. That is the conceptual jump. ### Which Google services are actually involved? The official list in Solana’s launch post names Gemini, BigQuery, Bigtable, Cloud Run, and Vertex AI Model Garden, with more to come. Solana also says the marketplace includes 50-plus community API facilitators spanning ecommerce, market data, communications, and onchain marketplace for agents. ### Why Solana and stablecoins? The whole design depends on cheap, fast settlement. Solana is pitching its network as good at exactly that, and its payments docs say the chain is built for predictable low fees and fast confirmations. Google Cloud has already been leaning into Solana tooling too — it hosts Solana resources, BigQuery data access, and even a PYUSD faucet for developers on Solana devnet. ### How does access stay controlled? This is the part product teams will care about. Solana says Pay.sh uses x402 for payment-triggered authorization, while the gateway still enforces verified endpoints, rate limits, quotas, and access controls. In plain English, the conversational layer and the spending layer can be split apart. The agent can ask for something, but execution still runs through policy gates. ### Does this replace subscriptions? Not fully — but it attacks the places where subscriptions make the least sense. If an agent needs one BigQuery job, one inference call, or one niche data lookup, monthly plans are clunky. Pay-per-request is closer to how software agents actually behave: bursty, composable, and task-specific: an autonomous workflow does not become an autonomous expense leak. ### Why does Google Cloud care? Because if AI agents become real software customers, cloud platforms need a native way to bill them. Google Cloud already supports Solana-related datasets and developer tooling, so Pay.sh looks like a practical extension of that relationship rather than a random crypto experiment. It gives Google a way to test machine-to-machine commerce on infrastructure it already runs. ### What changes now? The immediate change is small but important: an agent can move from “I can call tools” to “I can buy tool access on demand.” That sounds narrow, but turns out it is one of the missing pieces for autonomous workflows that cross company boundaries. If this sticks, the interesting shift will not be chatbots shift to execution rights instead of user accounts and monthly plans.