Stripe Acquires OpenRouter to Enhance Model Routing

Stripe has finalized the acquisition of OpenRouter, without disclosing the price. An estimate from a source places the transaction at just over $8 billion. This move comes as dynamic model routing is becoming widespread and token-based billing is becoming a cornerstone of AI applications.

Undisclosed Amount, Estimated Above $8 Billion, and Massive Traffic

Stripe and OpenRouter have not disclosed the financial terms of their agreement. However, a person familiar with the matter estimates the transaction at slightly over $8 billion. OpenRouter also claims to have a high usage scale, stating that it processes over 10 trillion tokens per day for a community of more than 10 million developers and businesses. In May, the company reported that its weekly volume had surged from five to 25 trillion tokens over six months, while serving more than eight million developers across over 400 models. These volumes face a shifting economic reality: processing costs vary by model and sometimes by the provider serving the same model. A June 2026 example cited by OpenRouter illustrates these discrepancies, with Llama 3.3 70B charged at $0.10 per million input tokens via DeepInfra compared to $1.04 via Together, and output prices ranging from $0.32 to $1.04 depending on the providers.

Dynamic Routing Expands at Snowflake, Cloudflare, AWS, and Microsoft

On August 18, Snowflake announced the arrival of dynamic model routing for Cortex AI Gateway, with a private preview planned, featuring request distribution based on quality, speed, customer preferences, and cost. Cloudflare already has a beta version of Dynamic Routing in AI Gateway, which allows users to set rules for model selection, manage quotas, and plan fallback solutions. AWS is integrating Intelligent Prompt Routing into Bedrock, while Microsoft Foundry offers routing profiles that balance quality and price. According to AWS and Snowflake, these systems can direct less complex requests to smaller or cheaper models, reserving other models for tasks requiring higher response quality or deeper reasoning. In practice, multi-model environments are becoming commonplace: a report from F5, based on over 1,100 IT decision-makers, indicates that 52% of organizations chain or orchestrate multiple models, averaging seven in use. On the provider side, Menlo Ventures noted in mid-2025 that 66% of developers had upgraded models without changing providers, while 11% had switched. However, Microsoft, via the Azure Architecture Center, warns that this dynamic selection complicates cost forecasting, debugging, and performance analysis when different requests are served by different models.

How OpenRouter Chooses the Model and Provider, with Automatic Recourse

OpenRouter evaluates each request based on complexity, price, speed, and reliability to direct it to an appropriate model. The platform adds a second layer of decision-making among providers of the same model: it continuously measures latency and throughput by model-provider combination and selects an endpoint that meets explicit cost or performance criteria. Its documentation states that clients can prioritize endpoints based on price, throughput, or latency, set price ceilings or minimum performance levels, and separate the model decision from the provider decision. The chosen provider also influences the cost, even when the model remains the same. This mechanism also serves as a safety feature: OpenRouter indicates that it can automatically redirect to other providers or models in case of failure, throttling, context length errors, or moderation refusals. Data processing constraints are also taken into account, with the option to limit routing to Zero Data Retention endpoints, exclude providers that collect data or train on prompts, and require localized processing (U.S. or EU) for enterprise clients.

Token-Based Billing: Stripe's Capabilities and Data Exposed by OpenRouter

Stripe has developed token-based billing tools for AI applications. Its service dedicated to LLMs, currently in private preview, tracks consumption for each model and each type of token, including input, output, and cached tokens when this option is available. According to the documentation, this service allows for token billing, prepaid credits, packages that include usage, or mixed plans, with automatic updates of model rates when providers change their prices. OpenRouter already provides much of the required data: its API returns the number of prompt, completion, reasoning, and cached tokens for each response, as well as the cost of the request, and logs separately the inference cost charged by the provider and the amount billed to the OpenRouter account. Accounts are settled using the native tokenizer of each model. The two companies were already collaborating before the acquisition: in January 2026, Stripe indicated that requests routed through OpenRouter could be tracked, priced, and billed by Stripe, already linking the routing layer to Stripe's measurement and billing systems. Patrick Collison emphasizes the central role of tokens as a unit for businesses building with AI and ties their economic use to the management of available computing resources. On the demand side, a Deloitte survey conducted at the end of 2025 among 515 U.S. decision-makers (all from organizations with annual revenues of at least $500 million) reported that 37% consume between one and ten billion tokens per month and 30% consume more than ten billion. By 2028, 61% expect to exceed ten billion monthly tokens, with a marked increase in loads surpassing the threshold of 100 billion tokens per month and a tripling of the frequency of these usages between 2026 and 2028. However, Deloitte warns against the notion that high consumption equates to more effective adoption, citing overly long prompts, insufficient context management, and low reuse as factors driving volume increases.

What the Deal Changes for Stripe and OpenRouter's Profile

Stripe has agreed to acquire OpenRouter, with the stated goal of adding model selection and routing to its existing work on AI and token-based billing. According to Stripe, OpenRouter aggregates over 400 models from more than 80 providers and exposes them via a single API, fitting into a broader movement of infrastructures adding model routing. Founded in 2023, the company raised capital from investors including Menlo Ventures and Andreessen Horowitz. In May, it closed a Series B round of $113 million led by CapitalG, with participation from NVentures, ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures, and Databricks Ventures.