Coinbase, Stripe, Google, and Visa Race to Power AI Agent Payments

AI Agent Payments Enter a Four-Way Infrastructure Contest

Coinbase, Stripe, Google, and Visa are now competing to define the default payment layer for AI agents. Each company is building a different architecture for a future where autonomous software can spend money for humans, businesses, and machines.

The race has moved into a more serious phase because the core problem is no longer theoretical. AI agents need a way to request paid services, prove authorization, complete transactions, and leave behind a reliable record of what happened. The companies involved are approaching that challenge from different starting points: crypto-native payments, merchant processing, authorization protocols, and card network tokenization.

The larger question is straightforward but difficult: will AI commerce run on stablecoin rails, traditional payment networks, or a hybrid model that blends both?

Competing AI Payment Protocols From Coinbase, Stripe, Google, and Visa

Coinbase x402 Uses HTTP 402 for Stablecoin Payments

Coinbase’s x402 protocol brings back the long-unused HTTP 402 “Payment Required” status code and applies it to AI-driven payments. Launched in mid-2025, x402 is designed to enable instant stablecoin payments through standard web requests.

The protocol effectively turns a stablecoin wallet into a universal API key. When an AI agent asks for access to a paid service, the service responds with a 402 challenge. The agent then answers with a signed USDC payment on the Base blockchain. Once the payment is verified, access is granted immediately.

This model removes the need for traditional accounts or subscriptions. Instead of asking an agent to log in, hold an API key, or move through a conventional billing process, x402 lets payment itself become the access mechanism.

Stripe and Paradigm Tempo Introduce Machine Payments Protocol

Stripe and Paradigm’s Tempo went live on March 18, 2026, with the Machine Payments Protocol, or MPP. The open standard was co-authored by both companies and is built to support machine-driven payments through a single HTTP round trip.

MPP is payment-agnostic. It can handle credit cards, cryptocurrencies, and Lightning Network transactions without locking AI agents into one payment method. The protocol embeds payment directly into the transport layer and uses a deterministic state machine.

That matters because it removes the need for pre-existing API keys. Instead of requiring prior account setup before an agent can pay for a service, the payment flow can happen directly inside the request process.

Google AP2 Focuses on Authorization and Delegation

Google introduced its Agent Payments Protocol, or AP2, in September 2025. The protocol enrolled more than 60 institutions, including Mastercard, American Express, PayPal, Salesforce, and Coinbase.

AP2 is built around cryptographic “mandates.” These are tamper-proof digital contracts that record user intent and cart approvals. The goal is to let users delegate buying authority to AI agents while keeping an auditable trail of what the agent was allowed to do.

This positions Google’s protocol around authorization, user intent, and accountability. Rather than focusing only on settlement or payment rails, AP2 addresses the trust layer needed when software acts on behalf of people.

Visa Intelligent Commerce Connect Extends Tokenization to Agents

Visa unveiled Intelligent Commerce Connect on April 8, 2026. The platform is designed as a network-agnostic and protocol-agnostic on-ramp for agentic commerce.

The system extends Visa’s existing tokenization infrastructure so it can issue agent credentials. A broader rollout is planned for June 2026.

Visa’s approach builds on its established card network capabilities while adapting them for AI agent activity. Rather than pushing one specific protocol or rail, Intelligent Commerce Connect is positioned as a bridge into agentic commerce across networks and payment systems.

The Emerging AI Payments Stack

Industry analysts have identified six layers in the emerging AI payments stack. Coinbase and Stripe each cover five of those layers, giving both companies broad positions in the race.

Coinbase controls settlement through Base, its blockchain infrastructure, while also covering wallets and the x402 protocol layer. That gives Coinbase a vertically connected crypto-native path for agent payments, especially where stablecoins and direct blockchain settlement are central.

Stripe brings a different kind of reach. It already has merchant relationships and traditional payment processing infrastructure. With Tempo, it adds stablecoin rails and a machine-focused payment protocol to that existing foundation.

Google’s AP2 sits mainly in the authorization and delegation layer. Its mandate-based structure is focused on proving user intent, preserving approvals, and creating an auditable record of agent-driven commerce.

Visa extends its card network with AI-ready tokenized credentials. Its role is tied to making existing payment infrastructure usable for autonomous agents while keeping the system network- and protocol-agnostic.

Crypto Rails, Card Networks, or a Hybrid Future

The battle over AI agent payments reflects a deeper split in how machine-to-machine commerce might work.

One path is crypto-native. Coinbase’s x402 shows how stablecoins can be used to make instant payments over standard web requests. In that model, wallets and blockchain settlement become core infrastructure for agents that need to pay for services without conventional account setup.

Another path is built around traditional payment networks. Visa’s Intelligent Commerce Connect adapts established tokenization infrastructure for agent credentials, giving AI agents a way to interact with commerce systems that already depend on card networks.

A third path is hybrid. Stripe’s integration of x402 on Base in February showed that these approaches do not have to be mutually exclusive. Google’s AP2 also supports cards, bank transfers, and stablecoins, reinforcing the idea that agent commerce may not settle around a single payment type.

Mastercard is also moving into the category with its own Agent Pay program using agentic tokens. That adds more pressure to a market already moving quickly, with major payment companies trying to shape the standard before a clear winner emerges.

Why AI Agent Payments Need New Infrastructure

AI agents create payment requirements that differ from traditional online checkout. They may need to pay for services directly, request access to APIs, purchase items based on delegated user intent, or complete machine-to-machine transactions without a human manually approving every step.

The systems being developed by Coinbase, Stripe, Google, and Visa each address a different piece of that problem:

• Coinbase emphasizes instant stablecoin payment and access through x402.
• Stripe focuses on payment-agnostic machine payments through MPP.
• Google centers authorization, delegation, and auditable user intent through AP2.
• Visa extends tokenized credentials into agentic commerce through Intelligent Commerce Connect.

The result is not one shared roadmap, but a competitive stack forming around the same core need: giving autonomous software a trusted way to spend money.