Stripe and Coinbase Compete to Control the AI Agent Payment Governance Layer

Article by: Prathik Desai

Compile: Block Unicorn

In March this year, OpenAI shut down a feature that allowed AI agents to shop on behalf of users. In the five months since its launch, fewer than 30 Shopify merchants had used the feature. The payment infrastructure itself was not at fault; the issue lay in the lack of rules to ensure a seamless shopping experience. Questions such as what goods agents could purchase, who collects sales tax, how to detect fraud, and who handles returns were never properly addressed.

Providing wallets or building payment infrastructure for agents is relatively straightforward. However, enabling individuals or businesses to use agents for spending in a trustworthy and regulated manner is far more challenging. Only programmability and rules can ensure a trustworthy environment. This absence of governance creates opportunities for the agent economy.

Last year, AI agents processed 176 million transactions totaling $730 million. While this figure may seem insignificant today, McKinsey predicts that by 2030, AI agents will facilitate $3 trillion to $5 trillion in global consumer commerce transactions.

Companies building this economic system are competing to control governance layers, including spending oversight, identity verification, and policy enforcement, which determine which agents have the authority to manage budgets.

Today, we’ll analyze who is building the banking layer for bots and what benefits those dominating this layer can gain.

The economic efficiency of processing agent payments is very low. Over the past 12 months, the average payment amount for AI agents has been just 31 cents.

Consider this: for a payment of 31 cents, how much profit is left for the multiple layers of intermediaries processing the transaction behind the scenes? Stripe’s standard pricing is 2.9% plus a fixed fee of 30 cents, leaving the merchant with less than one-tenth of a cent. Visa’s interchange fees then consume another third. In contrast, a Layer-2 stablecoin payment system processes the same transaction for just $0.0001.

These economic factors provide a basis for the use of cryptocurrencies at the settlement layer.

The payment infrastructure at the settlement layer is largely in place. Coinbase’s x402 protocol handled the majority of the 176 million transactions last year, and approximately 3,900 merchants currently accept proxy payments. Stripe and Tempo jointly developed a competing protocol—the Machine Payment Protocol (MPP)—launched in March, which integrates over 100 services. Google, Visa, and Mastercard also introduced their own proxy payment products during the same period. This means that within just 12 months, five competing payment architectures have emerged.

However, the problem with proxy payments is that no one can become wealthy by processing payments of just 31 cents. Therefore, value is concentrated in the funds in circulation and in the enforcement of rules governing how agents make payments.

Last week, we explained how businesses can capture value by owning the wallet layer that stores stablecoin balances for AI agents. But floating balances are just one of many valuable layers worth capturing. Another value layer is the rules that govern how those floating balances are used.

These rules include expenditure controls, agent identity verification, policy enforcement, audit trails, and responsibility allocation in the event of transaction failures. This layer is fully open.

In April this year, American Express launched the Agent Purchase Protection program, an insurance product designed to cover losses resulting from erroneous purchases made by AI agents. This effectively acknowledges the current state of governance in AI agent systems. Within an industry projected to reach $3 trillion to $5 trillion in size within five years, addressing the lack of governance holds significant value.

This is why current government officials are now competing to gain control of governance.

But which layer should this be built on? It could be a bank, a developer API, or even a wallet.

Every transaction by an agent must pass through the wallet. Therefore, the wallet is the optimal entry point for implementing spending limits, authentication, and manual approval. Once you control the wallet, you control the authority. The payment infrastructure company Stripe recognized this early on.

In June 2025, Stripe acquired Privy, a company that builds embedded wallets for consumer crypto applications. Through this acquisition, Stripe gained access to 75 million wallets distributed across more than 1,000 development teams. These wallets now sit at a critical juncture for fund flows, requiring all policies, spending limits, and manual approvals to be enforced prior to any fund movement.

Stripe has also built an entire proxy payment technology stack. It acquired Bridge to handle stablecoin coordination and fiat conversions. Additionally, it partnered with Paradigm to incubate Tempo, a Layer 1 blockchain focused on payments. Stripe and Tempo jointly developed the Machine Payment Protocol (MPP), an open standard that defines how agents request, authorize, and settle payments.

Stripe’s agent-ready financial solution now supports querying balances, paying bills, holding funds, creating virtual cards, and transferring money. Agents can autonomously execute routine payments, but any actions outside their policy scope are escalated for human review. Fund balances are supported by non-custodial Privy wallets across more than 150 markets.

Even though Amazon had to allow its developers to grant AI agents spending capabilities, it chose two wallet companies—Privy and Coinbase—rather than more established financial institutions like banks or credit card networks, opting instead for a wallet provider that has only been in existence for five years.

This is because the wallet serves as an ideal checkpoint, allowing for an appropriate level of human intervention to ensure necessary checks and balances.

Keyrock, in its report "Who Pays the Agent," notes that the agent marketplace will "tend toward equilibrium, where agents operate with considerable autonomy within boundaries enforced by crypto, which humans can audit and revoke."

This is where Privy fits within Stripe’s technology stack. The wallet is responsible for defining the boundaries within which the proxy must operate.

Here is how the governance strategy works on this technology stack.

Privy offers two smart wallet models. In the first model, the agent has full control over the wallet and executes transactions within policy constraints without requiring human approval. This model is best suited for fully autonomous agents such as trading bots and portfolio managers. In the second model, the user retains ownership of the wallet but grants the agent limited permissions to act as a signer. Users can revoke access at any time.

Stripe's MPP follows a similar governance strategy.

MPP has introduced a feature called "Session" for high-frequency proxy tasks. In session mode, the agent is pre-authorized with a spending budget and can make continuous payments within that limit without requiring a separate request for each on-chain transaction. MPP has implemented billing below one cent for LLM inference and pay-per-query billing for data APIs.

This is a level of governance granularity that card networks cannot support.

Although Coinbase's x402 currently leads in the field of AI agent payments, Privy’s advantage is less about cryptocurrency itself and more about the distribution moat it has built through Stripe.

Coinbase has 3,900 merchants that accept proxy payments. Stripe has approximately 1,000 merchants per each of Coinbase’s proxy payment merchants. In February of this year, Privy stated that if all Stripe merchants opted to accept machine payments, proxy commerce could scale today through Privy wallets. Stripe merchants do not need to build custom cryptocurrency infrastructure.

Competition between Stripe and Coinbase is intensifying, as other traditional giants join the race for vertical expansion, aiming to grow across the entire technology stack.

Keyrock has mapped 179 items across the six layers of the agency payment stack: settlement, wallets, routing, protocols, governance, and applications.

Coinbase and Stripe each cover five of these six layers. Circle covers four. Despite its scale, Google covers only two layers, and Visa covers just one.

Over the past twelve months, existing payment giants have invested over $8 billion to fill gaps in their technology stacks. Capital One acquired the AI-native software platform Brex for $5.15 billion. Mastercard acquired BVNK for $1.8 billion. Wallet layers and AI software layers have attracted the most active acquisition activity. Stripe acquired Privy, Fireblocks acquired Dynamic, and Arbitrum acquired ZeroDev. In these cases, payment infrastructure providers have each acquired an independent wallet provider.

These transactions collectively indicate that the market has identified a hierarchy of scarce resources. Settlement fees have become cheap and interchangeable, but the value lies in production licenses, budgets, and responsibilities.

Vertical integration across multiple layers also creates a synergistic effect.

The entity that controls this checkpoint can set spending rules, intercept funds before they flow, determine which merchants, agents, and applications receive trusted access, and charge fees for enabling all of this. We see this in Privy-Stripe’s distribution moat.

Even Coinbase’s position reflects this operational model: each x402 payment generates USDC demand on its Layer 2, Base, creating yield. This yield funds additional agent tools via AgentKit, which includes built-in session limits, per-transaction caps, and allowlists to restrict transfers to audited contracts. The more agents on AgentKit, the more x402 payments occur. Each layer interacts with the others.

Investment activities by existing businesses are much more active.

Coinbase Ventures has also invested in Catena Labs, Skyfire, and Payman—three of the most prominent independent governance startups. Catena was founded by Sean Neville, co-founder of Circle, and Circle has also invested in Skyfire. a16z led the funding rounds for both companies. Visa has supported Payman and established a partnership with Skyfire.

The same companies building the payment and settlement infrastructure are funding the governance layer. The idea is that if governance functions remain part of the existing infrastructure, as Privy has structured them in both models, incumbent institutions can maximize their returns. If governance becomes a separate layer, they will profit through their portfolios.

Payment processing has never been the most valuable role, because financial systems ultimately tend to homogenize. Once this happens, profit margins shift to the环节 that determine whether transactions are permitted and under what conditions.

Historically, many industries have undergone the same commoditization process.

Think about what happened after the internet commoditized cable television. All internet service providers (ISPs) became identical and nearly interchangeable. As a result, telecommunications companies had to pursue vertical expansion to remain competitive.

India's two major telecom operators, Jio and Airtel, have begun bundling hundreds of TV channels, six OTT platform subscriptions, unlimited voice calls, set-top boxes, and free routers into a single broadband package. Similarly, AT&T invested $85 billion to acquire Time Warner, becoming a giant that combines media and telecommunications. The goal is to combine Time Warner’s premium content—such as HBO, Warner Bros., and CNN—with AT&T’s extensive distribution network to compete with streaming platforms like Netflix and Amazon.

When broadband connectivity (the underlying infrastructure) becomes the least valuable part of the package, value shifts to the content, relationships, and offers that best attract customers.

We have also seen this in the cryptocurrency space.

Settlement was originally meant to occur at the protocol level. Think of Ethereum as a shared ledger where everyone settles transactions. After Coinbase launched Base as a faster, less congested Layer-2 chain, it began charging gas fees for every transaction settled on its own chain. Today, Coinbase earns approximately $60 million annually in sequencer revenue by processing transactions on Base.

The participants in building the proxy payment system learned valuable lessons.

In the book "Active Float," we explain how to construct economic systems by controlling the stablecoin balances held by agents between transactions, enabling companies that control the wallet layer of the tech stack to diversify their revenue streams.

The governance layer has added another source of revenue, potentially a larger one.

Visa processes $14.2 trillion in payment volume annually, earning a 0.28% commission. This rate includes not only transaction fees but also reflects the management fee Visa earns through the trust it builds by preventing fraud, resolving disputes, and enforcing network rules.

Even applying a small fraction of this fee to agency trades would reveal the immense value it brings to companies built on the governance layer. McKinsey forecasts that agency trading volume will reach $3 trillion by 2030, at which point a governance fee of just 0.1%—about 35% of Visa’s fee rate—would generate $3 billion in annual revenue. For reference, Coinbase’s total subscription and services revenue in 2025 is approximately $2.8 billion. Governance layer revenue from agency trades alone could rival the combined income Coinbase currently generates from staking, custody, and Coinbase One.

Companies operating across the wallet, settlement, and governance layers of the delegated finance stack can generate revenue from idle delegated balances (floating income), settlement fees (sequencer revenue per transaction), and compliance fees (governance execution).

This is why vertical integration across the entire technology stack will be the only business model that enables a company to remain competitive in the agent era.