How Stripe Is Reshaping Payments with AI Agents, and What Its Strategy Really Is

Author: Insights4VC

Compiled by TechFlow

TechFlow Editor's Note:As artificial intelligence moves from being a mere conversational tool to an autonomous “actor” that can execute tasks, payment infrastructure is undergoing a fundamental reshaping. Stripe has acquired Bridge, Privy, and Metronome, and is seeking to acquire OpenRouter, assembling a new closed-loop business model from the ground up to serve autonomous machine-to-machine settlement.

However, behind the polished narrative, agentic payments face a harsh reality check in actual business arithmetic. The dilution of inference costs, the true retention rate of stablecoins in the settlement chain, and the defensive counterattacks by traditional card networks in identity and risk controls all determine that the machine economy's explosive growth will not happen overnight.

This article looks beyond the hype around acquisitions and product launches to deeply deconstruct how Stripe is trying to capture value both before and after payments occur, and to clarify the economic gaps that must be bridged before agentic commerce can be commercialized.

On October 26, Instinct launched a prepaid-enabled Shopping Agent for eligible U.S. users.

Users can fund it with their Link account balance and set spending limits for each purchase. Within those limits, the agent can autonomously complete checkout without seeking user consent again. New York State was excluded from the initial rollout markets.

The day after Instinct's launch, on October 27, Stripe Crypto disclosed the underlying payment infrastructure behind it. [1]

In Instinct's onboarding flow, Stripe's onramp converts USD funded by debit card into stablecoins on the Tempo chain.

The funds are then deposited into a non-custodial wallet provided by Privy. Link is responsible for strictly enforcing the user's spending rules, while Stripe Issuing provides virtual debit cards for final checkout.

Merchants still receive standard card payments without needing to integrate with any cryptocurrency system. Link's terms explicitly state that non-custodial agent wallet balances use USDC, even though OUSD is the default option in other contexts for Stripe's stablecoin products. [1] [2]

Figure: Instinct's valuation surged about 300x in one year (Source: @rdominguezibar)

Purchase requests that exceed the limit must receive the user's approval on a transaction-by-transaction basis; within the limit, Instinct can place orders directly under preset permissions.

This means that an agent could well buy the wrong item within the budget limit. Link's user terms generally make users bear the consequences of these purchasing decisions, while treating technical errors at the payment-execution level separately and preserving applicable rights.

Stripe also provides protection mechanisms for compliant agent purchases, initially through Muse. However, this protection depends on the specific agent and transaction type, and the official announcement did not confirm whether Instinct orders are covered. [2]

Instinct integrates Privy wallets with Stripe's card-issuing infrastructure into a single checkout flow, while at the merchant end, cards still play the core role in checkout.

For Stripe, the key business question is: how much of the pre-checkout activity can be converted into chargeable services? This covers the entire process of model selection, billing and metering, spending authorization, and fund transfers.

YC Is Looking for Crypto Startups

Y Combinator has included agentic commerce, stablecoin applications, trading systems, and institutional-grade financial products among its priority crypto tracks for funding.

The Request for Startups (RFS) published by guest partner Nemil Dalal provides a useful counterpoint to Stripe's strategy: the underlying infrastructure is being assembled rapidly, while specific applications built on top of it still need to prove real market demand.

The regular application deadline for YC's Winter 2027 batch is 8:00 p.m. Pacific Time on November 2, 2026.

Figure: Key focus areas in Y Combinator's Crypto Request for Startups (CRFS)

Figure: Y Combinator's priority exploration tracks for the next six months (capital raising, stablecoins, trading, and agent infrastructure)

Stripe: The Acquisition Playbook Behind the Strategy

Stripe disclosed that its total transaction volume in 2025 reached $1.9 trillion, up 34% year over year. A February 2026 employee tender offer valued the company at $159 billion.

The first figure measures total payments processed, while the second reflects equity valuation; neither is actual revenue. Axios reported that first-half revenue rose 41% and free cash flow rose 43%. However, these announcements did not disclose specific profit data for individual product segments. [4] [34]

Stripe's recent series of acquisitions covers the entire transaction stack. Bridge, whose acquisition closed on February 4, 2025, handles stablecoin issuance and orchestration.

Privy also joined in 2025, bringing embedded wallets, signing, and programmable control software. The Metronome acquisition, which closed on January 14, 2026, added usage metering and billing capabilities.

On August 19, Stripe agreed to acquire OpenRouter, a routing provider that aggregates large-model requests. As of the October 28 research cutoff, the parties had announced an agreement, but final closing had not been officially announced. [3] [5]

Reuters, citing people familiar with the matter, reported that the acquisition consideration for OpenRouter was slightly above $8 billion. Neither party publicly disclosed the exact transaction amount.

In May, OpenRouter had just announced the completion of a $113 million Series B led by CapitalG, with media reports valuing it at approximately $1.3 billion at the time. Public financial disclosures lack details on OpenRouter's profitability, making it difficult to fully explain the surge in consideration. [8] [33] [35]

The proposed technology stack begins with model requests. OpenRouter selects preferred inference providers; Metronome and Stripe Billing record usage and bill customers.

Payments and Checkout handle payment collection; Connect coordinates platforms and sellers, while Link handles buyer authentication. Privy provides wallet and signing controls, Bridge provides fiat-to-stablecoin conversion, and Treasury manages balances and settlement.

Agentic Commerce Suite helps merchants distribute product catalogs and integrate agentic checkout; Stripe MCP allows authorized agents to call financial tools under permissions strictly scoped to the account.

Even so, Stripe still depends on external model providers, banks, card networks, and merchants. Tempo and open payment protocols add more infrastructure, but Stripe does not control every participating node. [3] [4] [5] [10] [11] [12] [20] [23] [30]

As early as April, OpenRouter had already provided services through Stripe Projects, including account opening and billing. Broader integration remains underway.

On October 15, Patrick Collison said substantial work remained on integration across Privy, Bridge, and Stripe. John Palmer noted that the PartyDAO team had been at Stripe for six months, working on Link balance functionality. [9] [11] [32]

OpenRouter and inference costs

Acquiring OpenRouter would position Stripe firmly at the core node where AI companies incur model inference costs.

When the acquisition was announced in August, OpenRouter disclosed more than 10 trillion inference tokens per day, a community of more than 10 million developers and enterprises, and aggregation of more than 400 models from 80 providers.

By October 28, its website showed more than 700 trillion tokens per month, aggregating more than 500 models, 10 million users, and 80 providers. Note that token usage reflects model consumption only and is not directly equivalent to revenue. [5] [6]

OpenRouter's routing update launched on October 28 can weigh trade-offs among providers offering a given model, considering price, speed, and stability.

In official tests, system performance improved significantly while costs rose by 6%. If faster responses and lower failure rates reduce retries and customer support, enterprises are fully willing to accept this trade-off.

Conversely, if cheap model requests require manual intervention to correct errors, the total cost is higher. Connecting routing and billing data helps enterprises accurately calculate the actual ROI per customer or feature.

Take a hypothetical AI company with $100 in revenue and $60 in model inference spend: if model costs fall by 20%, inference spend drops to $48.

Excluding other costs, contribution margin jumps from $40 to $52, a 30% increase. This simple business arithmetic clearly shows that optimizing model selection has a far greater impact on margins than small improvements in payment success rates.

Metronome founder Scott Woody wrote on October 1 that some customers have already selected models upstream and passed token consumption data to Metronome to manage margins.

He believes internal usage metering need not mechanically determine how customers are billed: products can sell credits per business operation while flexibly switching the underlying model calls. Inference routing and billing share the same economic objective and do not need to be forced onto a blockchain. [10]

OpenRouter charges standard customers a 5.5% top-up fee and enterprise customers 8%, while passing through underlying model prices at cost without markup.

Therefore, a $100 standard credit top-up can generate $5.50 in platform fees before OpenRouter's own costs. Direct provider agreements and custom enterprise deals may change the actual rates. Mechanically applying public rates to all disclosed tokens would significantly overestimate actual revenue and profit. [7]

Stripe could have been content remaining a payments and billing tool provider, leaving model routing to independent providers and thereby avoiding huge integration costs.

A controlling stake in OpenRouter helps improve product distribution and consolidate cross-system data, even though large customers may still have the capability to build their own gateways. OpenRouter's Alex Atallah has committed to remaining provider-neutral; if future commercial interests favor certain models, that commitment will face scrutiny. [5]

Granting Agents Spending Authority

Prudence is also needed when evaluating Privy's disclosed growth data. In October, Collison said its monthly transaction volume was close to $25 billion and cited that since the acquisition, assets had grown about 25x, revenue 9x, and monthly active developers 3x, but did not provide a baseline.

Privy disclosed in September that the wallet count exceeded 160 million, up from the 150 million accounts mentioned in August. These metrics use different definitions, do not distinguish unique active users, and the $25 billion transaction volume includes substantial non-agent spending activity. [11] [12]

Privy focuses on embedded wallets, delegated signing, and enterprise control tools, including whitelists, transfer limits, and multisig approval workflows, and supports pricing based on active users, signing frequency, or transaction volume.

Once an enterprise has built authorization rules around a wallet, switching providers involves migrating both business processes and funds. Fortunately, exportable private keys and open standards limit lock-in. AWS's AgentCore, previewed in May, supports both Privy and Coinbase wallets as well as budget controls.

On October 13, Privy announced a partnership with Ripple covering XRPL wallet creation, transaction signing, and agent payments.

The announcement listed policy controls and gas fee sponsorship as future development work and did not provide a clear timeline for a full-featured production launch. [12] [13] [28]

The true value of agent wallets lies in authorization controls: who can spend, whom they can pay, spending limits, and when human approval is required. This logic also applies to banking and card payments.

Instinct's Link balance uses a self-custodial architecture, while Privy also supports Bridge's custodial solution. These balances are not protected by bank deposit insurance, and self-custodial users also bear key recovery risk.

Although on-chain transfers are irreversible, subsequent card transactions remain protected by card networks' refund and chargeback dispute processes. [1] [2]

For ordinary consumers delegating an agent to buy $100 sneakers, precise authorization and reliable fulfillment matter far more than minor cost differences between settlement rails.

Traditional bank cards have long built ubiquitous merchant networks, credit lines, and dispute arbitration systems. Instinct leverages exactly this: agents hold stablecoin-funded balances, generate virtual cards at checkout to complete payments, and keep the crypto mechanics hidden behind ordinary retail transactions. [1] [2]

But in scenarios where agents procure API calls, data, or inference compute, the commercial logic reverses. If a single call is priced at just 1 cent, a fixed 30-cent card processing fee would be 30 times the purchase amount.

If 1,000 calls are bundled into a $10 invoice, the fixed fee is diluted to below 3%. Traditional usage-based billing and cryptographic payment channels can both effectively spread per-transaction costs, making batch settlement a highly competitive alternative to individual on-chain payments.

The Machine Payment Protocol (MPP), jointly developed by Stripe and Tempo, launched alongside the Tempo mainnet on March 18. The blockchain was co-incubated by Stripe and Paradigm.

MPP allows services to request and verify payment before delivering resources, and Stripe's implementation supports both bank cards and stablecoins. Its supported Payment Sessions let users pre-fund and aggregate multiple requests, greatly reducing separate on-chain settlements, although funds locked in a session cannot be used elsewhere until released. [14] [15]

Coinbase's x402 protocol also relies on HTTP payment challenge mechanisms and focuses on stablecoins; retail checkout uses a separate protocol.

Stripe and OpenAI's ACP protocol coordinates checkout and authorization; Google's UCP covers product discovery and post-transaction flows, while AP2 provides signed intent credentials. MCP connects agents to tools, A2A connects agents to each other, and settlement occurs at the underlying layer.

Tempo listed more than 100 services at launch, but directory entries cannot prove actual payment frequency; repurchase rate and net revenue are the solid metrics for validating results. [14] [16] [17] [18] [19]

For enterprises, the biggest appeal of stablecoins lies in cross-border transfers outside banking hours, provided recipients can hold or convert back to fiat at reasonable cost.

For domestic vendor settlements, traditional bank accounts, credit lines, and centralized netting often perform better. A comprehensive cost comparison must include foreign exchange spreads, prepaid working capital, compliance and risk controls, withdrawal slippage, and failed retry costs.

If contractors can spend USD balances directly through linked cards, they may avoid immediate withdrawals; but vendors needing local fiat deposits have no choice. Agents must still verify payees and invoice authenticity.

On October 21, Stripe launched Treasury services for agents. Authorized tools in ChatGPT and Claude can directly check balances, pay bills, transfer funds, and exchange currencies.

Six days later, Stripe announced Stripe Pay, offering free, real-time business-to-business transfers via dashboard or CLI. As of October 29, the CLI preview supports same-currency transfers among USD, GBP, and USDC for compliant businesses in the United States and the United Kingdom.

Payers need a pre-funded Stripe Treasury account, while payees can use Treasury or a linked bank account. Agent-assisted CLI payments require human confirmation, and official documentation does not list Tempo as the settlement and clearing layer. [20] [21]

Stripe Pay gives enterprises a strong reason to keep funds within the Treasury system after collecting payments.

Enterprises can identify vendors via Stripe Handle and initiate payments, which are executed after human approval. If vendors also hold Treasury accounts, more funds remain within Stripe between transfers, supporting demand for its financial software services even when transfers themselves are free.

However, because recipients can still link external bank accounts to withdraw funds, Stripe Pay cannot ensure that funds never leave the platform. Its commercial value depends on whether enterprises use it frequently and pay for adjacent ancillary services. [20] [21]

Privy officially launched Treasury Management in full on September 17, providing strategy and approval workflow tools for on-chain assets. Tempo launched Zones privacy partitions on October 14, establishing Deel as its first official client.

Zones provide enterprises with a degree of transaction privacy protection, but operators retain visibility, and activity at public on-ramps and off-ramps may still leak information. Payroll settlement providers also need reliable mechanisms for handling payment failures and confidential records. [22] [27]

OUSD's economic mechanics

OUSD introduces interest income from reserve assets into Stripe's broader stablecoin strategy. Open USD is issued by Bridge Building Inc. and was fully integrated into Stripe on September 30.

Open Standard is responsible for governance and distribution, with Stripe, Coinbase, Mastercard, Shopify, and Visa as initial founding members. BlackRock, Lead Bank, and The Bank of New York Mellon (BNY) manage reserve assets.

The issuer dashboard separately discloses cash and short-term U.S. Treasury positions. Bridge's issuer role and the reserve-management functions of the aforementioned institutions are clearly separated from Stripe's own platform-distribution role. [23] [24] [25]

Open Standard said participating enterprises can share reserve returns based on their supply volumes and ecosystem activity. OUSD uses the same dollar accounting unit as USDC or USDT; the core differences lie in governance structure and distribution incentives.

Other issuers also subsidize distribution channels. The real commercial test is whether OUSD's terms can attract sticky long-term retained funds and recurring real payments at high frequency.

Enterprises may generate massive transaction volume while retaining very little OUSD; conversely, large retained balances can earn reserve interest but may not be accompanied by active transactions. If interest rates decline in the future, these reserve returns will shrink substantially. [25]

Stripe has set OUSD on the Tempo chain as the default configuration while retaining support for other assets and networks. Even so, Instinct still selects USDC in Link balances.

If OUSD successfully scales, Treasury, distribution settlement, wallets, and card-issuance programs will form its massive distribution channel. In October, Stripe expanded its stablecoin card to 65 countries and regions, with plans to reach 100 by year-end.

Even if funds originate from stablecoins, end merchants can still seamlessly receive local fiat currency. [1] [23] [26]

Even when multiple options coexist, default choices still profoundly influence users' asset-holding preferences. On October 5, Henri Stern emphasized system interoperability and customer choice, a commitment that becomes critical when Stripe and OUSD have aligned incentives.

Customers always have reasons to choose the lowest-cost or most reliable path, including continuing to use traditional bank wires when they perform better. Stripe's Treasury documentation shows that its checking accounts are provided by Fifth Third Bank; although Stripe controls the software interface, deposit infrastructure still relies on traditional banking partners. [20] [28]

The Gap Between Usage and Actual Demand

Stripe said its Link users reached 250 million in June. In September, its agent-initiated spending transaction count surged 38x month over month, but it did not disclose the initial base or transaction amounts.

Link's massive coverage gives Stripe a powerful distribution springboard, yet this surge does not reveal how many real users have already become accustomed to fully delegating purchasing decisions to agents. [29]

Allium estimated that from January to August, real stablecoin payments ranged from $401 billion to $527 billion, while raw on-chain transfers during the same period totaled $85 trillion, adjusted to $4 trillion after cleaning.

This statistic excludes internal hedging by exchanges, nested DeFi structures, and infrastructure arbitrage, separating payments from speculative trading. Identifiable business-to-business (B2B) payments reached $137 billion to $153 billion, forming the largest core base.

The latest briefing on October 15 showed total payments of $482 billion over the first nine months. Different methodologies and statistical classifications are difficult to convert directly, and most of this aggregate volume was not initiated by AI agents. [36]

Analysis of 2026 Stablecoin and Payment Economics

Stablecoin markets are moving more capital, but retained balances in their own asset pools have not expanded at the same rate.

Chainalysis documented an artificial inflation of x402 transaction data driven by speculative token activity. Bitquery's audit of August transactions showed that 88% of transaction volume matching broad-authorization characteristics was concentrated in a single cross-chain bridge.

This methodology also captured a large volume of routine operations unrelated to agentic commerce, so the ratio applies only to a specific sample and cannot be directly equated with market share across the entire x402 market. On-chain transaction records alone cannot verify who initiated payments or what actual services buyers received. [37] [38]

Card network giants are aggressively advancing their own agentic payment products. Visa launched a live-network pilot on October 20, connecting BILL buyers and Billtrust suppliers through tokenized cards, agent identity verification, and invoice review.

Mastercard's Agent Pay, upgraded in September, focuses on identity authenticity, transaction intent detection, and anti-fraud filtering. Google, OpenAI, Coinbase, and major mainstream banks also occupy critical links in the transaction chain.

Even if underlying protocols are interoperable, core institutions controlling user distribution channels, credit risk management, default liability, and anti-fraud arbitration retain the decisive leverage in commercial competition. [13] [17] [18] [19] [39] [40]

Compared with uncontrolled open consumer spending, standardized invoicing and multi-level approval policies make enterprise settlement an ideal starting point for agent deployment. Automation can reduce administrative compliance costs but may not directly expand total payment volume.

Especially under bulk billing models, this may generate more sporadic transactions with thin per-transaction fees. When transfer fees are driven toward zero, Stripe's core profit will shift toward collaboration software and workflow orchestration services.

Free account-to-account transfers provided through Stripe Pay help retain funds and customers but do not directly generate transfer commissions. Whether adjacent enterprise services can offset this profit compression remains to be verified by financial data. [21]

Stripe's April roadmap plans a non-public test of Privy agent policies in Q4 and full rollout of Link stablecoin settlement in Q1 2027.

Privy Treasury Management fully launched on September 17 but is a separate product line. Access rights across global regions also depend on specific integration methods.

As of October 29, SaaS platforms using Connect to allow agents to sell products must be based in the United States and subject to a review waitlist; merchant checkout documentation shows support for the U.S., Canada, and European markets. Shared payment tokens in Europe are expected to enter public preview in Q4. [22] [30] [31]

For Link and Instinct, user repurchase frequency, average order value, return rates, and fraud-related bad debt are far more persuasive than flashy user growth multiples.

The OpenRouter acquisition remains pending final closing, and its potential profit benefits must be objectively weighed against customer-built routing solutions. Planned treasury risk controls and multi-region support must move beyond limited testing and waitlists.

Stablecoin payments face a common-sense reality check: after accounting for exchange spreads, prepaid capital costs, and withdrawal friction, are they truly cheaper and more resilient than traditional payment solutions? Subsidy-driven false prosperity will eventually recede once rewards end. If rising transaction volume is accompanied by continued gross margin erosion, the foundation of this business model will be undermined.

Instinct now integrates crypto wallets, spend authorizations, and debit card issuance into a single purchase flow. But at a macro level, the overall system remains a patchwork of discrete products at different integration stages, and Stripe has not publicly disclosed specific margins for agentic commerce.

The strategic premise hinges on whether it can capture value before and after a payment is completed—that is, in model distribution, usage metering, permissions management, and fund movements.

Bank cards, bank wires, and stablecoins will coexist and complement each other over the long term; Stripe's ultimate moat depends on whether customers are willing to pay for its autonomous choice among these three rails.

References

[1] Noah Shinn, Instinct launch; Stripe Crypto, underlying architecture and Link balance revealed, October 26–27, 2026.

[2] Link, agentic wallet balances, FAQ, and terms of service, official online documentation.

[3] Stripe, acquisition of Bridge (February 2025); acquisition of Metronome (January 2026).

[4] Stripe, 2025 annual update and financial results report, February 24, 2026.

[5] Stripe, OpenRouter acquisition agreement; OpenRouter, official announcement of joining Stripe, August 19, 2026.

[6] OpenRouter, real-time core platform metrics.

[7] OpenRouter, official documentation on pricing and billing.

[8] OpenRouter, official announcement of Series B funding, May–August 2026.

[9] OpenRouter, Stripe Projects integration documentation, April–June 2026.

[10] Scott Woody (Stripe), 'Why I Tried to Kill Per-Token Pricing', October 1, 2026.

[11] Patrick Collison, Privy integration progress update, October 15, 2026.

[12] Privy, institutional wallets, crypto orchestration, control framework, and pricing plans, August–September 2026.

[13] AWS, AgentCore Payments preview (May 7); Privy, XRP Ledger agentic commerce, October 13, 2026.

[14] Tempo, mainnet launch and Machine Payment Protocol (MPP), March 18, 2026.

[15] Stripe, Machine Payment Protocol; Tempo, MPP payment session mechanism, March–April 2026.

[16] Coinbase, official announcement of x402 protocol launch, May 6, 2025.

[17] Stripe and OpenAI, Instant Checkout and Agentic Commerce Protocol (ACP), September 29, 2025.

[18] Google Cloud, Agent Payment Protocol (AP2), September 16, 2025.

[19] Google, Universal Commerce Protocol (UCP), January 11, 2026.

[20] Stripe, Treasury services for agents, October 21, 2026.

[21] Stripe, official launch of Stripe Pay (October 27, 2026) and official CLI technical documentation.

[22] Privy, full launch of treasury management features, September 17, 2026.

[23] Henri Stern (Stripe), OUSD becomes the default stablecoin, September 30, 2026.

[24] Bridge, official launch of OUSD and reserve assets dashboard, September–October 2026.

[25] Open Standard, OUSD launch and integration details; Solana Foundation, OUSD on Solana, September 2026.

[26] Neetika Bansal (Stripe), A New Paradigm for Money Movement at Fintech Companies, October 16, 2026.

[27] Tempo, Privacy Partition Feature Officially Launched, October 14, 2026.

[28] Henri Stern, Steering the Stripe Crypto Strategy, October 5, 2026.

[29] Stripe, AI Spending Data and Link-Based Personal Agent Shopping Analytics, June to September 2026.

[30] Stripe, Official Documentation for Agentic Commerce and SaaS Agent Sales.

[31] Stripe, Sessions 2026 Conference Announcements and Product Evolution Roadmap, April to October 2026.

[32] John Palmer, Link balance Development Diary; Jeff Weinstein, Privy and Tempo Architecture Deep Dive, October 26, 2026.

[33] Reuters, Exclusive Report on Stripe's Proposed Acquisition of OpenRouter, August 19, 2026.

[34] Axios, In-Depth Coverage of Stripe's OpenRouter Acquisition and Financial Performance, August 19, 2026.

[35] TechCrunch, OpenRouter Valuation Surpasses $1.3 Billion, May 26, 2026.

[36] Allium, 'State of Stablecoins and Payments Report', Public Summary, September to October 2026.

[37] Chainalysis, Adoption of x402 Agentic Payments, June 3, 2026.

[38] Bitquery, On-Chain Audit Report on x402 AI Agent Payments, September 6, 2026.

[39] Visa, Live Network Pilot for BILL and Billtrust Agentic Payments, October 20, 2026.

[40] Mastercard, Agentic Commerce Trust and Intelligence Services, September 30, 2026.

Source: 深潮 TechFlow