Five protocols from five trillion-dollar companies, all shipping within weeks of each other. The race to own the trust layer for non-human buyers has started.

Five protocols. Five different bets on how AI agents should discover products, verify identity, and settle payments. Google's Universal Commerce Protocol. Visa's Trusted Agent Protocol. Stripe and OpenAI's Agentic Commerce Protocol. Alipay's Trust Protocol. Coinbase's x402. Each one backed by a company that already processes trillions in annual volume. All shipping within weeks of each other.

The competitive framing misses the point. These protocols are not mutually exclusive. Google's UCP is explicitly compatible with Visa's infrastructure. Stripe's ACP works across AI agents, not just ChatGPT. Coinbase's x402 handles settlement that the others deliberately avoid. They are building different floors of the same building, and they know it.

The real battle is not over which protocol wins. It is over which company owns the trust layer: the identity verification for non-human buyers. And the gap between what shipped this month and what merchants are actually ready for is wider than anyone wants to admit.

Alipay launched AI Pay in 2025 as a payment rail purpose-built for AI agents. Not a chatbot bolted onto a checkout page. A full payment infrastructure designed so that software, not people, could initiate, authorise, and settle purchases.

By February 2026, it was processing 120 million agent-driven transactions per week, making it the first AI payment service to hit that milestone. The use cases are already mundane in the best possible way: ordering coffee from Luckin Coffee through an AI assistant, managing grocery runs via Taobao Instant Commerce, even paying for things through Rokid smart glasses using nothing but a voice command.

In January, Alipay formalised what was already happening by launching the Agentic Commerce Trust Protocol with partners including Alibaba's Qwen App and Taobao Instant Commerce. The protocol sets standards for how agents identify themselves, verify user intent, and maintain data privacy during a transaction. Qwen App became the first platform to adopt it, letting users chat with AI to place food and beverage orders without ever opening a separate app.

The developer story is equally telling. Alipay and Rokid built their integration on the Model Context Protocol (MCP), the same standard now being built into Chrome itself, allowing developers to wire AI payment capabilities into smart devices through natural language rather than complex coding.

China is not debating whether agents will shop. They are debugging the receipts.

While Alipay was counting transactions, Google was rewriting the checkout page.

In January 2026, Google unveiled the Universal Commerce Protocol (UCP) at the National Retail Federation conference. By February, it was live: US shoppers could purchase items from Etsy and Wayfair directly inside AI Mode in Google Search and the Gemini app. No redirect. No cart page. No checkout flow. Just a conversation that ends with a confirmation.

UCP is not a payment method. It is a commerce language, and as we explored in our breakdown of the three-layer agentic commerce stack, it sits at the protocol layer between consumer intent and payment settlement. Merchants declare what capabilities they support, including bespoke functionality. Agents discover those capabilities, negotiate what they can handle, and complete the transaction. The protocol was co-developed with Shopify, Target, Walmart, Visa, Mastercard, Stripe, Adyen, and more than 20 other global partners. It is compatible with Agent2Agent (A2A), the Agent Payments Protocol (AP2), and MCP.

The expansion roadmap is aggressive. Shopify, Target, and Walmart integrations are next. Hundreds of additional retailers have contacted Google about joining. The protocol will extend to India, Indonesia, and Latin America.

Google VP Vidhya Srinivasan put it bluntly: "Agentic commerce is no longer just a concept, it's reality."

Senator Elizabeth Warren responded within hours, raising concerns about data collection practices, product recommendation bias, and whether partners receive preferential treatment. The commerce layer is being rebuilt at speed. The regulatory response is already trailing.

Google gave agents a storefront. Alipay gave them a payment rail. Coinbase gave them a wallet.

On February 12, Coinbase launched Agentic Wallets: crypto wallet infrastructure built specifically for AI agents to hold funds, send payments, trade tokens, and earn yield autonomously. Not agent-assisted. Agent-operated.

The wallets run on the x402 protocol, an open payment standard that repurposes the HTTP 402 "Payment Required" status code for instant blockchain settlements. The protocol, co-developed with Cloudflare, has already processed more than 50 million transactions across automated use cases including API access fees, computing payments, and resource purchases between agents.

Security is enforced through what Coinbase calls Smart Security Guardrails: spending limits, session duration restrictions, and enclave isolation that keeps private keys within Coinbase's secure infrastructure. The keys are never exposed to a prompt or a large language model.

Stripe followed days later with its own machine payments tool leveraging x402 on the Base network, allowing developers to bill AI agents directly in USDC stablecoins.

This is the detail that matters for anyone who moves money: Coinbase did not build a payment button for agents. It built a financial identity for them. An agent with an Agentic Wallet can accumulate value, make spending decisions within programmed guardrails, and settle transactions without a human approving each one. That is a fundamentally different category than "agent-assisted checkout."

Five major protocols. Five different bets on how agents should transact. Each one backed by a company that already processes trillions.

Google's UCP handles the full commerce journey: discovery, negotiation, checkout, and fulfilment. It is the most ambitious play, attempting to become the universal language for agentic shopping. Visa's Trusted Agent Protocol (TAP), developed with Cloudflare, focuses narrowly on the trust layer: cryptographic identity verification that lets merchants confirm an agent is legitimate before a transaction begins. Stripe and OpenAI co-developed the Agentic Commerce Protocol (ACP), an open standard under Apache 2.0 that lets merchants enable agentic payments with as little as one line of code. Alipay's Agentic Commerce Trust Protocol governs the Chinese market. And Coinbase's x402 handles the crypto-native settlement layer.

The competitive framing misses the point. These protocols are not mutually exclusive. Google's UCP is explicitly compatible with Visa's infrastructure. Stripe's ACP works across AI agents, not just ChatGPT. Visa's TAP does not process payments. It verifies who is making them.

The numbers explain the urgency. Visa reported a 4,700 percent surge in AI-driven traffic to US retail sites. Nearly half of American shoppers (47 percent) now use AI tools for at least one shopping task. Consulting firm Edgar Dunn & Co. projected that AI-driven commerce could reach $1.7 trillion by 2030, up from $136 billion this year. PayPal CEO Alex Chriss called agentic commerce the "biggest transformation since the advent of e-commerce," predicting 25 percent of online sales will come from AI agents by 2030.

Visa is working with more than 100 partners worldwide. Google has 20-plus. Stripe lists Microsoft, Anthropic, Perplexity, and Vercel among its early integrators. Everyone is cooperating publicly and competing privately.

The infrastructure is shipping. The guardrails are not.

One in three web requests to retail sites now comes from a bot, and AI traffic consumes up to 70 percent of costly dynamic resources like search and personalisation engines. Most merchant checkout flows were built for humans: CAPTCHA challenges, dropdown menus, JavaScript-heavy confirmation pages. None of that works when the buyer is software.

The security surface is expanding faster than defences can cover it. Proofpoint acquired Acuvity this month specifically for agentic AI security. Non-human identities already outnumber human employees 96-to-1 in financial services, according to Total Retail.

Merchants now face a narrow path, and as we covered in the marketplace extinction event, some are already choosing sides. Block all bot traffic, and you lose agent-driven demand to competitors whose checkouts are machine-readable. Open the gates without guardrails, and you invite fraud at a scale that human-built defences were never designed to handle.

The protocols being shipped this week are, at their core, an attempt to build that middle path: verified identity, scoped permissions, and auditable intent for every non-human transaction. Whether they can scale trust as fast as they are scaling transactions will determine whether this week's milestone becomes the foundation of a new commerce era, or a cautionary tale about moving too fast.

If your checkout cannot serve a non-human buyer today, how long before that buyer finds a merchant whose checkout can?