$75M in One Week: The Agent Payment Infrastructure War Has Begun
Five startups raised $75 million in a single week to build payment infrastructure for AI agents. Add the $33 million raised by two others in the preceding four weeks, and you get $108 million flowing into agent-native payments since early July — from a16z, Dragonfly, Forerunner, Alibaba, and Tribe Capital.
This is no longer a niche. This is an infrastructure land grab. And the most interesting part isn't who raised what — it's what they're all missing.
The Week That Changed Everything
Here's every deal, in chronological order:
| Company | Round | Lead Investor | Date | Thesis |
|---|---|---|---|---|
| AIsa | $6.5M Seed | Alibaba, Tribe Capital | Jul 7 | "Transaction network for AI agents" — managed marketplace with fiat + stablecoin rails |
| Natural | $30M Series A | Forerunner | Jul 30 | "Stripe for AI agents" — payment orchestration layer |
| Pilot Protocol | $4.5M Seed | Undisclosed | Jul 31 | "Internet for agents" — 250K agents, 2B requests/day |
| Sapiom | $35M Series A | Dragonfly | Aug 4 | Agent payment rails + runtime ops; 270M+ txns, 100K agent runs/day |
| Catena Labs | $30M | a16z | Aug 4 | Filing for national trust bank charter — for AI agents |
| MAGNE.AI | $2.64M | Strategic | Aug 5 | x402-compatible "Agent Pay" + edge-AI hardware |
| Total (7 weeks) | $108.6M |
And that's just the venture dollars. In the same period:
- Cloudflare launched Wallets (Aug 4) — stablecoin wallets with
cloudflare.payhandles spent limits for agents. This completes their x402 two-sided market: Monetization Gateway (sell side, Jul 1) + Wallets (buy side, Aug 4). - Mastercard closed its $1.8B BVNK acquisition — stablecoin settlement infrastructure at global scale.
- Arkham Intelligence and Glassnode both added x402 support within a single week — enterprise data providers adopting agent-native payments.
- Ripple joined the x402 Foundation and XRP Ledger crossed 1.4 million agent-initiated transactions.
The $28K/Day Reality Check
Here's what makes this fascinating: all this infrastructure is being built for a market that barely exists yet.
The x402 protocol has processed 200 million transactions. But independent analysis (Forkast, Aug 2026) found that 95% of those are protocol signaling — agents testing the plumbing. Real daily commerce across the entire ecosystem: approximately $28,000.
The average x402 transaction is $0.52. Human wallet confirmation costs often exceed the payment itself. Visa and Artemis, in their July 2026 report, adjusted cumulative x402 volume to $15 million — a fraction of the $50 billion headline number that circulates.
This isn't bearish. It's the most bullish signal possible. The infrastructure is being built before the commerce exists — which means the builders (Cloudflare, a16z, Dragonfly, Stripe, Visa) believe the commerce is inevitable. They're not waiting for demand to prove itself. They're building the supply.
What Everybody Is Building — And What Nobody Is Building
Look at the funding table again. Every single company is building payment infrastructure:
- How agents hold money (Cloudflare Wallets, Catena Labs trust charter)
- How agents move money (Sapiom, Natural, MAGNE.AI)
- How agents comply with regulations (Catena Labs, Bridge/Stripe integration)
Zero of them are building the thing that comes before payment:
How does an agent discover which of 300+ services to call? How does it compare them? Trial them without committing funds? Trust that the service will actually deliver?
The highway is being built by trillion-dollar companies. The vehicles (agents) are multiplying exponentially. But nobody is building the map — the discovery layer where agents find, compare, trial, and trust services before they spend a single cent.
This is the open frontier. And it's the position minia2a occupies.
The Two Models: Managed vs. Permissionless
The funding wave reveals a clear fork in the road:
| Managed Marketplace | Permissionless Discovery |
|---|---|
| AIsa, Coinbase AgentKit | minia2a — open index, any agent, any service |
| Curated listings, approval required | Self-listing — curl one command, service is live |
| Platform-controlled pricing | Provider-set pricing with 5% platform fee |
| Fiat on-ramps, KYC | USDC on Base, no KYC, 500 free trial credits |
| Builds a walled garden | Builds a public utility |
Managed marketplaces make sense for enterprises that need compliance guarantees and curated quality. But the agent economy won't be one marketplace — it'll be thousands of services, from individual developers selling CAPTCHA-solving endpoints to Glassnode selling on-chain data. A permissionless discovery layer is the only architecture that scales to that diversity.
What the Numbers Say
As of August 7, 2026, minia2a has served:
- 300 services — up from 180 in late July, growing ~10/day
- 9,667 free trials used by 321 unique agents
- 380,589 total API requests — real agents making real calls
- 49 agents with registered wallets — the bridge from trial to commerce
The top 5 services by trial usage tell a clear story about what agents actually need:
- CAPTCHA solving (1,280 trials) — agents hitting web forms they can't pass
- Knowledge retrieval / RAG (1,285 trials) — agents needing external context
- Service discovery (859 trials) — agents looking for other agents to call
- Gas price data (708 trials) — agents optimizing on-chain decisions
- Time/date utilities (420 trials) — agents that need to know when it is
These aren't speculative use cases. These are the actual API calls agents are making, right now, in production.
What Comes Next
The $108 million question isn't "will agents pay each other?" — that's already happening. The question is how they'll discover what's worth paying for.
We're building that answer as an open, permissionless layer. Not another managed marketplace. Not another payment rail. The discovery layer that sits between the wallets and the services — where any agent can find, trial, and trust any endpoint, without asking permission.
The infrastructure war is being won by trillion-dollar companies. The discovery war is still wide open.