In the span of five days this month, three major players launched agent payment infrastructure. Cloudflare shipped Wallets on August 4. Circle opened its Discovery API on August 6. And now Solana, together with Google Cloud, has launched Pay.sh — a unified registry where AI agents discover and pay for cloud services in stablecoins.
This is no longer a technology demo. It's the beginning of a registry war — and the winner determines how agents discover what's worth paying for.
Pay.sh is a discovery and payment registry built on Solana, with Google Cloud as the anchor tenant. The key components:
This matters because Google Cloud has the largest existing API catalog of any pay.sh participant. Gemini alone processes millions of API calls daily. If even 1% of those calls route through agent wallets, it dwarfs every other agent payment marketplace combined.
In one week, the agent payment ecosystem crystallized into three competing approaches to the same problem: how does an agent find a service it can pay for?
| Model | Player | Approach | Services |
|---|---|---|---|
| Curated Registry | Circle Discovery API | Compliance-screened catalog, 15+ chains, 14 query params | 900+ |
| Infrastructure-Layer | Cloudflare Wallets | Budgeted wallets + merchant allowlists, settlement on Base/Solana | Any site behind CF |
| Platform-Native | Solana Pay.sh | Google Cloud anchor tenant + community facilitators, Solana settlement | 50+ facilitators |
These aren't mutually exclusive — an agent could discover through Circle, pay through Cloudflare, and settle on Solana. But each model competes to be the default entry point for agent developers. And that default matters enormously.
Previous agent payment infrastructure came from crypto-native companies (Coinbase, Circle) or infrastructure companies (Cloudflare). Google Cloud is different:
The strategic logic is clear: Google wants AI agents to consume cloud APIs the same way humans consume web pages — with a wallet, not a contract. And Solana wants to be the settlement layer for that economy.
Last week, Forkast News reported that x402 has processed over 200 million transactions — but the real daily commercial volume is estimated at only $28,000. That means roughly 95% of x402 transactions are protocol signaling: machines testing the plumbing, not actual commerce.
The activation gap is real.
328 services listed on minia2a.uk. 12,217 free trials used. 60 registered wallets. Only 14 paid transactions. Over 34,000 issued credits sit idle.
This isn't a minia2a problem — it's an industry-wide pattern. Building the payment rails is the easy part. Getting agents to actually spend money is the hard part.
This is why the registry war matters. Payment rails (x402, USDC settlement) are commoditizing. The value is shifting to discovery + trust: which services are real, which are reliable, and which are worth an agent's limited budget.
Discovery (Circle, Pay.sh, minia2a) → Budget Control (Cloudflare Wallets, .agent-budget) → Settlement (x402, Solana, Base). Each layer has multiple competitors, and the winners at each layer may not be the same company.
When the world's third-largest cloud provider builds agent-native payment infrastructure, it signals that this isn't a crypto niche — it's the future of API consumption. Expect AWS and Azure to follow within 6-12 months.
Circle and Pay.sh are curated — they screen listings for compliance and quality. That's valuable for enterprise buyers. But the long tail of agent APIs (the indie developer who built a useful niche tool) needs open, permissionless discovery. If curated registries become the only game in town, 90% of API builders get locked out.
None of these registries offer a built-in trial mechanism. An agent discovers a service, sees the price, and... decides whether to pay, blind. This is the gap minia2a is built to fill: try before you pay, no wallet required. 12,217 free trials and counting prove agents want to test before they commit.
The agent payment infrastructure is real. The protocol works. The settlement layer is fast and cheap. But the industry is discovering that building rails is the easy part. The hard part — the part nobody has solved yet — is getting agents from "I found a service" to "I paid for it and got value."
That's not a payment problem. It's a discovery and trust problem. And the registry war is about who solves it first.