\The stablecoin thesis got commercial validation 48 hours after State of the Stack 2026 went public. The protocol underneath the new rail is x402.

We published State of the Stack: Agentic Commerce 2026 on May 5. On page nineteen, the most visible public production experiment for x402 had settled $0.11 across five transactions. Two days later, the Solana Foundation and Google Cloud opened a commercial rail running on the same protocol. The launch is called Pay.sh. AI agents can now pay for more than 75 enterprise APIs in stablecoins, including Gemini, BigQuery, BigTable, Cloud Run, and Vertex AI. The number underneath the protocol is no longer five transactions.

The pattern is the one we have been tracking. Stablecoin rails for agents form before the traditional rails catch up. The first commercial deployment happens on a protocol most payments executives have never heard of, distributed through a hyperscaler that does not run a payment network of its own.

Pay.sh is an API proxy built on Google Cloud Platform. It sits in front of Google Cloud services and a growing list of community APIs, and handles payment settlement on behalf of the calling agent. Solana wallets act as agent identity. The wallet signs the request, the proxy verifies funds, the call goes through, and the API provider gets paid in USDC reconciled over Solana's network.

Three things matter about how this is wired.

First, the wallet is the credential. There is no Google account to create. There is no API key to rotate. The agent's signing key is the identity, and payment is the authentication. The Solana Foundation describes this as "payment is the credential." It collapses two layers of agent infrastructure, identity and billing, into one transaction.

Second, the coverage is not narrow. Pay.sh launched with over 75 APIs available, including Google's own foundation models and data services, plus more than 50 community-built APIs. eCommerce APIs (Rye, BigCommerce, Purch), data and intelligence services (Exa, Dune Analytics, Nansen), agent communications layer (AgentMail, StablePhone, StableEmail), and Solana infrastructure providers (Helius, Alchemy, Quicknode, The Graph) are all callable through the same payment proxy.

Third, there are no subscriptions. Pay-per-request, no minimum spend, no manual billing reconciliation. For agents this is the point. Subscriptions assume a human signing up. Agent-native pricing does not.

x402 is the spine here. We have written about it before, both as the open internet payment protocol Coinbase released and as the layer Linux Foundation absorbed into its standards portfolio. State of the Stack tracked it as the most credible public production experiment for agentic payments. The numbers were small. The mechanism was working.

Pay.sh changes the size of the deployment without changing the protocol. The proxy still issues 402 Payment Required responses against unauthenticated agent requests. The agent still presents a signed payment alongside the retry. Settlement still happens automatically against a stablecoin balance. What is different is who is operating the gateway and what sits behind it.

When the gateway is run by an enterprise like Google Cloud, the volume that flows through it is no longer a community sandbox. It is the same surface area as Google's own paid API ecosystem, opened to agent-native callers without a credit card or a billing setup step. Solana's announcement also names MPP, the machine-native payment protocol Stripe has been building, as a foundational standard alongside x402. The agent payment protocol layer is consolidating around two open specifications, both already in production with hyperscaler-class operators.

That is the story underneath the launch. Open agent payment protocols are no longer theoretical. They are how Google Cloud is choosing to monetise enterprise API access for agents.

Google Cloud is not a payment network. It does not issue cards. It does not run interchange. What it does have is a captive enterprise customer base that already pays Google for compute, storage, and inference. Pay.sh does not compete with Google's existing billing relationships. It opens a parallel rail for agent traffic that would otherwise be impossible to bill cleanly.

The same mechanic is now visible across the major distribution platforms. Stripe and Google shipped their own agentic checkout integration earlier this year. Visa and Stripe are working on Tempo, a stablecoin settlement layer. The networks are increasingly comfortable with stablecoin rails for the cases where existing rails do not fit.

What changes with Pay.sh is who owns the wallet on the agent side. Google did not build a payments product. The Solana Foundation did, and Google distributes it through the Cloud Console. The customer relationship for compute and inference still sits with Google. The customer relationship for the payment, identity, and signing does not. That is a meaningful split in the agent infrastructure stack, and it is the first time we have seen a hyperscaler concede the wallet to a chain-native foundation rather than build its own.

State of the Stack named six pieces of infrastructure agentic commerce needs before it works at scale: commitment governance, identity, dispute resolution, settlement, observability, and the audit layer. Pay.sh closes one of those, settlement, in a clean and concrete way. It also makes a partial dent in identity, by treating the wallet as the agent's credential.

The other four are still open.

There is no dispute mechanism for an agent transaction settled in stablecoins on Pay.sh. If the API returns a response the calling principal disputes, the chargeback path does not exist. The Solana network is final by design. The State of the Stack agentic dispute layer is still missing.

There is no commitment governance. Pay.sh enforces rate limits and access controls at the gateway. It does not enforce the deeper question of whether an agent should be making the call in the first place, against what mandate, on whose authorisation, with what audit trail visible to the principal. That layer has to exist somewhere. Pay.sh does not pretend to be it.

There is no observability beyond the ledger. Solana writes the transaction. The proxy logs the response. What the agent did with the call, what state it was in when it called, what the outcome was for the principal, that is not the rail's concern. It belongs to whatever platform sits above the rail, and most of those do not exist yet.

Three things will tell us whether Pay.sh is the agent payments default or a useful sandbox.

The first is volume. State of the Stack tracked $0.11 across five transactions on the last credible public x402 experiment. Pay.sh is a different scale of distribution. We expect the next visible production figure to be at least three orders of magnitude higher within ninety days. If it is not, the launch is a marketing event.

The second is the network response. Visa, Mastercard, and the legacy processors have invested in agentic payments primarily through their existing card rails plus selective stablecoin pilots. Pay.sh sits outside both. If Visa or Mastercard chooses to operate a competing gateway, the rail consolidates. If they do not, agents on hyperscalers route around them.

The third is the regulatory layer. AI agents transacting in USDC across a global gateway will draw attention. Stablecoin issuance rules, money transmission licensing, and consumer protection frameworks were not designed for autonomous principals. The GENIUS Act stablecoin AML compliance regime in the US is one of several coming live in 2026. How Pay.sh handles those obligations, especially identity and sanctions screening for agent wallets, will set the template.

Network rails took half a century to professionalise. Stablecoin rails for agents arrived inside a quarter. Which one is the platform of record when this scales?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.