The race over how AI will spend our money is no longer hypothetical — it is a full‑scale industrial contest. In the past 12 months two very different strategies have emerged:
Google and a 60‑plus partner coalition publishing an open protocol that stitches AI into existing payment rails, and
Stripe building a vertically integrated stack — from stablecoin orchestration to wallets and its own payments primitives — designed to make agentic commerce run inside its lanes. The result is not just a standards fight; it is a battle for the tollbooth on trillions of future transactions.
Background: why payments matter when agents buy for us
AI agents are rapidly moving from product discovery and recommendation to
transaction execution. That change breaks a core assumption in payments infrastructure: today’s systems assume a human clicks “buy.” When an agent acts on behalf of a person, we need new ways to prove authorization, manage credentials, settle funds, and resolve disputes — and those choices determine who captures recurring revenue from millions of tiny, automated purchases.
Two competing plays emerged publicly in 2025:
- An industry coalition centered on Google released the Agent Payments Protocol (AP2), a payment‑agnostic framework built around cryptographic mandates to show an agent had user authorization.
- Stripe, working with OpenAI, released the Agentic Commerce Protocol (ACP) and rolled it into live commerce features inside ChatGPT via an “Instant Checkout” experience; Stripe paired protocol work with acquisitions and a multi‑layered product strategy that covers stablecoin accounts, wallets, and even a purpose‑built blockchain incubated with Paradigm.
Those moves are the first visible stages of a much larger contest: who controls the
entry point (the interface and standard that AI agents use to initiate payments), and who owns the
settlement layer (the rails where transactions ultimately land and get reconciled).
Overview: AP2 vs ACP — different philosophies, similar stakes
AP2: coalition, mandates, payment‑agnosticism
Google’s approach centers on an
open, industry coalition and an interoperability philosophy. AP2 focuses on verifiable, tamper‑evident
mandates — cryptographically signed records that attest to the user’s instructions and agent authority. That mandate model is engineered to satisfy traditional risk, compliance, and audit requirements: payments providers, banks, and card networks can rely on explicit proof of consent rather than inferred model intent.
Key characteristics of AP2:
- Payment‑agnostic: designed to support cards, bank transfers, real‑time rails, and stablecoins.
- Mandate‑first: cryptographic proof of approval travels with the transaction.
- Coalition model: more than 60 partners spanning card networks, banks, payment processors, marketplaces, identity and security vendors.
- Open orientation: public repos and reference implementations intended to accelerate cross‑industry adoption.
The coalition model buys Google leverage with incumbent financial institutions: by centering existing trust architectures and compliance flows, AP2 targets enterprise comfort and regulatory acceptance.
ACP: product integration, tokens, and “Shared Payment Tokens”
Stripe and OpenAI took a different tack: they co‑designed a protocol that maps cleanly onto Stripe’s long‑standing merchant and platform APIs and shipped a live experience (Instant Checkout) inside ChatGPT. ACP uses a new payment primitive — a
Shared Payment Token (SPT) — that permits an AI surface to request a charge without ever exposing raw credentials. ACP is merchant‑friendly and tightly integrated with Stripe’s fraud, tax, and fulfillment primitives.
Key characteristics of ACP:
- Product‑first, live deployments: powering Instant Checkout inside ChatGPT with Etsy (and broader Shopify merchant rollouts).
- SPT primitive: scoped, time‑limited tokens that reduce PCI exposure and facilitate delegated transactions.
- Merchant control: merchants remain merchant‑of‑record and keep fulfillment, returns, and tax flow under their systems.
- Developer ergonomics: single integrations, hosted ACP endpoints, and clear migration paths for existing Stripe customers.
ACP’s live deployments gave it a head start: working commerce, product feedback loops, and a revenue model embedded in an enormous consumer surface (ChatGPT’s weekly active user base).
Timeline and tactics: how Stripe “built the road” and Google “built a coalition”
The last 18 months can be read as a sequence of deliberate moves by each camp.
- Stripe’s acquisition play and vertical build
- Stripe acquired Bridge, a stablecoin orchestration platform, and completed that deal in early 2025. That purchase gave Stripe custody, issuance, and flows for dollar‑pegged stablecoins.
- Stripe launched Stablecoin Financial Accounts (available to businesses in 101 countries), enabling firms to hold and move dollar‑pegged stablecoins across fiat and crypto rails.
- Stripe acquired Privy, a wallet infrastructure company that enables embedded on‑chain wallets inside applications, closing a wallet gap in its stack.
- Stripe and Paradigm incubated a payments‑focused Layer‑1 blockchain (Tempo), designed for sub‑second settlement and ultra‑low fees; Tempo raised a large Series A and opened public testing.
- Stripe co‑developed ACP with OpenAI, shipped Instant Checkout inside ChatGPT, and packaged an Agentic Commerce Suite for merchants.
- Google’s coalition and AP2
- Google published AP2 as an extension of its agent protocols, inviting banks, card networks, market places, and Web3 companies to align on mandate semantics and agent identification.
- The coalition includes major financial incumbents and identity vendors, deliberately anchoring agentic commerce within legacy trust frameworks.
- Google introduced merchant‑facing features inside Search and Gemini (and later Universal Commerce Protocol extensions), enabling in‑interface checkout on select retailer partners.
Stripe’s moves look like a “build‑every‑layer” play — own the orchestration, custody, issuance, wallet UX, and a blockchain tuned for payments. Google’s play is a standards coalition designed to preserve the role of established payment networks and issuers while enabling agentic flows.
The real beneficiary: why Circle (USDC) is the stealth winner
There is a second, quieter dynamic everyone should notice: no matter which interface or protocol wins,
settlement is migrating toward compliant stablecoins, and one issuer sits squarely in the center.
Circle’s USDC is a reserve‑backed digital dollar with institutional integrations and public financial disclosures. As players design agentic checkout, they need a settlement asset that satisfies enterprise audits, AML controls, and reconciliation practices. Circle’s model — transparent reserve reporting, institutional custodianship, and public corporate governance — makes USDC an obvious preference for enterprises building programmatic payments.
Put simply:
- Both sides support USDC in practice. Stripe’s stablecoin financial accounts include USDC; OpenAI and other merchant platforms route settlement via compliant stablecoins; Google’s AP2 explicitly accommodates stablecoin rails.
- Institutional comfort with reserve backing and auditability favors USDC over less‑transparent alternatives.
- Whoever wins the entry point (AP2, ACP, UCP, or platforms yet to come) will likely still settle a meaningful share of volume into a small set of compliant stablecoins — and Circle’s USDC is positioned to capture settlement volume.
That means the
interface battle may determine transaction fees and platform economics, but the
settlement economics — interest on reserves, custody fees, and treasury yields — channel predictable revenue to stablecoin issuers.
Technical contrasts: mandates, SPTs, and the plumbing of trust
Understanding where friction will arise requires a closer look at the primitives both camps propose.
- Mandates (AP2): cryptographic attestations of a user’s explicit authorization. Good for audit trails and dispute defensibility. Mandates align with banking and card‑network risk practices because they provide deterministic evidence of intent, not probabilistic inference from model outputs. Mandates help incumbents preserve liability models and apply step‑up authentication when risk signals demand it.
- Shared Payment Tokens (ACP): a merchant‑scoped token issued by a payment provider that authorizes a charge for a specific cart and amount. SPTs reduce PCI scope for the AI surface and let the payments processor retain fraud detection and charge orchestration. SPTs optimize product integration and merchant experience.
- Stablecoin settlement (Tempo, Bridge): on‑chain settlement enables real‑time finality, programmable fees under a token model, and near‑zero per‑transaction costs when engineered to scale. Tempo and similar L1 designs aim for sub‑second confirmations, deterministic settlement, and stablecoin‑native gas mechanics.
- Wallet UX (Privy): embedded wallet APIs let apps give users on‑chain balances without the cognitive load of external wallets. For mainstream adoption, wallets must behave like bank accounts: intuitive UI, secure key management, and clear recovery flows.
Each primitive trades off between developer ergonomics, regulatory friendliness, and operational control. Mandates maximize enterprise trust; SPTs maximize integration velocity; stablecoin settlement minimizes settlement friction and fees — if regulators and institutions accept those coins.
Business stakes: why this is a trillion‑dollar fight
Agentic commerce isn’t just another payment product. Analysts project explosive growth in AI‑mediated transactions. Conservative industry forecasts estimate agentic commerce could be worth hundreds of billions to trillions by the end of the decade as agents handle routine purchases, subscriptions, and microtransactions on behalf of users.
Why that equals trillions of dollars of value — and recurring revenue — for infrastructure owners:
- Each transaction carries a small fee. Multiply tiny margins across high frequency, low friction purchases and the economics compound.
- Merchants will pay to be surfaced by agents; AI platforms will monetize seamless checkout flows.
- Settlement and treasury functions generate ongoing interest and float revenue when stablecoin reserves are managed centrally.
- Marketplace dynamics: once an interface becomes the default entry point for agents, it captures high‑value merchant relationships and data flows that enable cross‑sell and ad monetization.
Control of the
checkout determines short‑term merchant fees and revenue splits; control of the
settlement layer determines who earns recurring treasury and custody economics. The synergies are why a standards announcement or an acquisition can move markets.
Regulatory and consumer safety risks
Agentic commerce accelerates value, but it magnifies risk vectors that regulators and consumers already worry about.
- Authorization and dispute complexity: when agents act, disputes morph from “did a human click buy?” to “did the agent operate within user policy?” Mandates help, but attribution models, consent granularities, and liability chains must be standardized. Consumers could face purchases their agent “thought” were correct but that the human never intended.
- Fraud at scale: AI systems can be targeted for automated card‑testing, credential abuse, or model manipulation. New fraud patterns will emerge when agents have delegated authority to spend on repeated routines.
- Consumer protection and refunds: current refund and chargeback frameworks assume a human initiated purchase. Agentic commerce needs new contractual and operational solutions for merchant returns, automated refunds, and liability arbitration — especially for asynchronous or subscription purchases an agent initiates while the user is asleep.
- Concentration risk and gatekeeping: if one platform becomes the dominant agent entry point, it gains outsized control over pricing, discoverability, and merchant economics — replicating past platform‑power dynamics inside agentic flows.
- Regulatory acceptance of stablecoins: for stablecoin settlement to scale, regulators must retain confidence in reserve practices, custody models, and AML controls. Public‑company issuers with transparent reporting and third‑party custody will have advantages, but regulatory frameworks still vary by jurisdiction.
- Privacy and profiling: agents personalize purchases using deep personal data. Marketplaces and protocols must define how consented context travels across agents and merchants without enabling opaque profiling or nondisclosed data monetization.
Regulators are already responding to aspects of this: the GENIUS‑style stablecoin frameworks, public accounting obligations for issuers, and payment network rule adaptations show that policy is catching up — but the pace of innovation outstrips rulemaking.
Merchant and developer playbook: practical steps today
Merchants, platforms, and payments teams must treat agentic commerce as an imminent channel, not a futurist thought experiment. Practical readiness steps include:
- Inventory readiness
- Ensure real‑time product feeds, accurate inventory levels, and clear return policies.
- Prepare product metadata in machine‑readable formats so agents can compare and choose reliably.
- Payments and compliance
- Support multiple settlement options (cards, ACH, and compliant stablecoins).
- Evaluate partnerships with both Stripe ACP pathways and AP2‑aligned payment gateways where available.
- Fraud and dispute workflows
- Update fraud rules to account for delegated tokens and cryptographic mandates.
- Define clear refund SLAs and automation for agent‑initiated purchases.
- UX and data
- Maintain merchant‑of‑record controls: preserve brand voice in agent interactions and always display merchant support channels.
- Decide how much personalization to surface to agents (e.g., saved addresses, spending caps).
- Multi‑protocol strategy
- Recognize that multiple protocols may coexist (ACP, AP2, UCP, others). Adopt an integration approach that can route and reconcile orders across agent channels.
Being “agent‑ready” will be a competitive advantage. Early adopters who retain ownership of the fulfillment and customer experience will capture repeat value even if the AI surface captures initial discovery credit.
What to watch next: chokepoints and likely winners
- Adoption velocity of protocols. Live deployments matter. ACP’s early production usage inside ChatGPT gave Stripe and OpenAI crucial real‑world data that will inform fraud models and merchant economics. AP2’s coalition gives it deep banking and card network credibility — and that matters for enterprise adoption.
- Tempo and other payment‑optimized blockchains. If L1s like Tempo deliver on sub‑cent fees and near‑instant finality with strong custody and compliance primitives, transaction settlement economics change fundamentally. High‑frequency agentic flows favor blockchains designed for payments.
- Regulatory clarity. New national frameworks that codify reserve requirements, disclosures, and issuer responsibilities will favor compliant stablecoin issuers. Market consolidation is likely: trusted issuers with strong governance will capture settlement volume.
- Merchants’ multi‑protocol strategies. Nobody wants to alienate a channel that channels millions of buyers. The path to scale will be multi‑protocol interoperability and merchant tools that syndicate catalogs across agent surfaces.
- Consumer protections and dispute law evolution. Practical rules for refunds, unauthorized AI purchases, and agent misbehavior will shape user trust and therefore adoption.
Conclusion: two futures, one unavoidable truth
The contest between Google’s coalition and Stripe’s vertical stack embodies a larger design question:
should AI commerce be grafted onto existing financial trust architectures, or should we build new rails optimized for programmatic agents? Both answers have merits — and both are being built in parallel.
- Google’s AP2 prioritizes regulatory alignment and institutional trust through mandates and coalition governance.
- Stripe’s ACP plus acquisitions prioritize end‑to‑end product experience and optimized settlement economics through stablecoin accounts, wallets, and a payments‑centric blockchain.
But one thing is clear:
AI will spend money, and when it does, money will almost certainly flow on‑chain via compliant stablecoins. That reality elevates a different class of winners: not the interface owner alone, but the settlement providers who manage reserves and custody. In that contest, a transparent, audited, institutional stablecoin has enormous leverage.
For consumers, the upside is undeniable: seamless shopping, lower friction, and smarter automation. The downside is equally tangible: disputes that don’t map cleanly to human intent, concentrated platform power, and new fraud vectors. The technology train has already left the station; our policy, merchant operations, and consumer safeguards must now sprint to keep pace.
Expect the coming 12–24 months to be decisive. Protocols will iterate. Live deployments will teach new lessons. Regulators will test assumptions. And the company that combines broad developer adoption, merchant economics, and settlement credibility will capture the lion’s share of this emerging market.
For now, watch the lanes closely:
AP2,
ACP,
UCP, and independent blockchains like
Tempo are not just technology experiments — they are blueprints for who will collect tolls in the age when machines buy for people.
Source: 深潮TechFlow
The Covert Battle of AI Payments: Google Brings 60 Allies, While Stripe Builds the Entire Road