Stripe is moving to bring OpenRouter, which handles "selection and settlement" in the AI model market, under its umbrella. Bloomberg reported on August 16 that the two companies had agreed to acquisition terms exceeding $7 billion and were nearing a finalized deal. However, neither company has publicly confirmed the transaction, and Stripe told TechCrunch that it does not comment on rumors or speculation. The reason a payments company would invest such a large sum in an AI router lies not in the models' intellectual property, but in the ability to seize, in one continuous chain, the path through which inference requests flow, along with usage metering and billing.
Over $7 Billion Is More Than 5.38x the Valuation From Three Months Ago
This latest report signals that acquisition speculation, which had continued since July, has now progressed to agreed terms. The Wall Street Journal and The Information reported on July 23 that Stripe had entered negotiations to acquire OpenRouter for approximately $10 billion. The price of over $7 billion reported by Bloomberg is lower than that earlier figure, but since the ratio of cash to stock and any contingent consideration remain unknown, this cannot be definitively characterized as a markdown.
In May, OpenRouter raised $113 million in a round led by CapitalG. While the company did not disclose its valuation, the New York Times reported that the post-funding valuation was $1.3 billion. The reported price represents more than 5.38x that valuation from roughly three months ago—a premium exceeding 438%.
Measured against revenue, the price looks even steeper. According to The Information, OpenRouter's annualized revenue stood at $50 million as of April. However, given the usage growth discussed below, that April figure cannot be used as the basis for calculating the current acquisition multiple. The price reportedly offered by Stripe prices in not so much current fee income as the future growth in inference volume and developer touchpoints.
A 200 Trillion Token Exchange for the Inference Market
OpenRouter's own published figures show that usage scale increased by another notch in the short period between its May funding round and August 17. These are all company-disclosed figures that have not been audited, and the measurement periods may not be identical, but the direction of growth is clear.
That said, 200 trillion tokens per month cannot be directly converted into revenue. Input, output, and cache tokens carry different unit prices, and even for the same model, prices vary by connected provider and processing tier. As cheaper open-weight models proliferate, a doubling in token count will not translate into a doubling of transaction value. To substantiate the acquisition price, one would need to verify, after an official announcement, the proportion of enterprise contracts and the actual revenue generated by the 5.5% fee.
| Metric | May 2026 Announcement | Official Page as of August 17 | Change |
|---|---|---|---|
| Monthly Tokens Processed | 100 trillion | Over 200 trillion | More than 2x |
| Users | Over 8 million | Over 10 million | Increase (rate not calculable) |
| Models Offered | Over 400 | Over 500 | Increase (rate not calculable) |
| Connected Providers | Not stated | Over 80 | Not comparable |
OpenRouter bundles multiple models and inference providers behind a single API. By default, it selects connections without recent major outages, prioritizing lower-priced providers among those. Developers can sort by price, processing speed, and response latency, and requests are switched to alternative providers if a connection goes down. Users can also set conditions such as Zero Data Retention (ZDR), which limits requests to providers that do not store data, or exclusions for providers that use prompts for training.
Pricing follows a pay-as-you-go plan at 5.5% of the inference price. Even if competitors could replicate the code itself, it would be difficult to acquire, all at once, a connection network of the same scale, along with observational data on outages and pricing, transaction volume, and an established user base. What OpenRouter has accumulated is not just routing functionality but market intelligence—which models are actually chosen, at what prices and speeds.
Why a Payments Company Wants an AI Router
The relationship between Stripe and OpenRouter predates the acquisition talks. In January, the two companies announced that OpenRouter uses Stripe Invoicing, Tax, and Radar to collect payments from customers worldwide. They also built an integration in which, when model providers change pricing, OpenRouter processes the inference requests while Stripe tracks usage and reflects it in pricing and billing.
Stripe subsequently launched its own AI Gateway, releasing token-based billing for LLMs in private preview. According to current documentation, the model providers directly handled by Stripe AI Gateway are OpenAI, Anthropic, and Google. External gateway integrations include OpenRouter, Vercel, and Cloudflare. Bringing OpenRouter under its wing would let Stripe instantly expand its reach to over 500 models and more than 80 providers.
Stripe AI Gateway records per-customer token usage at the same time it returns a model's response. Depending on configuration, it can reject requests before execution once a customer's purchased credits run out. If OpenRouter's routing and Stripe's usage caps were linked together, businesses could handle both the decision to switch to a cheaper model and the process of stopping overuse within a single flow. If the acquisition leads to product integration, this is the area where developers would see practical benefits soonest.
The monetization mechanisms also overlap. Stripe's token-based billing syncs model pricing, and once an AI company sets its markup ratio over cost, it handles everything from usage metering to invoice issuance. Stripe also completed its acquisition of Metronome, which handles complex usage-based billing, in January. The Revenue product suite—including Billing, Invoicing, and Tax—is projected to reach an annualized run rate of $1 billion sometime in 2026, and the total payment volume flowing through Stripe in 2025 reached $1.9 trillion.
By incorporating OpenRouter, Stripe could vertically connect the point where a model is chosen, the point where tokens are counted, the point where pricing is set, and the point where payment is collected. This is not a horizontal expansion of payment functionality—it is a move into the space just upstream of where an AI service's cost and selling price get determined.
Neutrality Conditions Developers Should Watch
Even if the acquisition closes, it hasn't been determined that OpenRouter's products or terms will change immediately. At this stage, with no official announcement yet, it makes sense to track the following four factors as observable conditions.
- Fees: Whether the 5.5% pay-as-you-go rate and enterprise discounts are maintained.
- How models and providers are selected: Whether the ranking based on price, speed, and uptime remains transparent.
- Data policy: Whether OpenRouter's own stated policy of not retaining prompts, and per-request ZDR designation, continue.
- Breadth of choice: Whether the roster of 500+ models and 80+ providers is maintained, without steering toward particular model companies or Stripe products.
OpenRouter's selling point was that it freed developers from being locked into a single company's models. But as dependence on the router itself deepens, the impact of any erosion in trust regarding its neutrality would also grow larger. Stripe's documentation lists Vercel and Cloudflare as external gateways as well, so developers still have alternative paths available. That is precisely why whether OpenRouter continues to select providers based on price and reliability will weigh just as heavily as the acquisition price itself.
What we want to confirm once there is an official announcement, beyond the consideration exceeding $7 billion and the closing timeline, is the design approach: will OpenRouter be operated independently, or integrated into Stripe AI Gateway? Whether Stripe can convert 200 trillion tokens of flow into revenue while preserving neutrality in model selection—that will be the first test of this acquisition's value.