Summary: OpenRouter announced a $113 million Series B on May 26, 2026, led by Alphabet’s CapitalG; The New York Times put the post-money valuation at about $1.3 billion, while OpenRouter did not state a valuation. The company reported about 8 million users and roughly 100 trillion tokens monthly, up from about 5 trillion tokens weekly six months earlier. The funding reflects production AI traffic and demand for routing, though revenue, margins, and long-term gateway concentration remain open questions.
OpenRouter announced on May 26, 2026 that it had closed a $113 million Series B, with CapitalG — the growth-equity fund inside Alphabet — leading the list of investors. Per The New York Times, the deal leaves the AI gateway service at about $1.3 billion after the money — more than twice where it stood a year back. A three-year-old outfit whose product is, at bottom, “picking the right model for you” being worth that much says something plain: AI applications have moved out of demos and into production, and the layer that picks models has turned from a convenience into infrastructure.
One Round, Twice the Valuation
Numbers first: CapitalG led a $113 million round; OpenRouter said nothing about valuation, and the roughly $1.3 billion figure is the Times’. The previous mark, per PitchBook, was a $547 million post-money — the outcome of a $40 million Series A in June 2025 on which Andreessen Horowitz and Menlo Ventures took the lead, Sequoia joining as well. One year, one round: better than double.
Founded in 2023, the company runs an AI gateway — a single interface from which developers and companies reach more than 400 models supplied by Anthropic, Google, OpenAI, xAI, DeepSeek, and others, choosing per task where to trade cost against reasoning depth and accuracy. The wide shelf is the selling point: no vendor alone fields that much model coverage, and OpenRouter packs the whole market behind a single interface and a single bill.
A Monthly Load of 100 Trillion Tokens
More telling than the check size is the traffic. OpenRouter now serves a global base of about 8 million users, with roughly 100 trillion tokens flowing through each month — about 25 trillion a week, a number the outfit’s own press release stuck straight into its headline. Wind back six months and the weekly count was 5 trillion. Half a year, five times over.
Revenue went unmentioned. Still, for a gateway, throughput is the least fudgeable size gauge there is: each token moving through the pipes is one routing call, one billing line, one usage-management event — the raw feedstock the business eventually charges against. The timing deserves a note as well: this is precisely the stretch in which agent software began reaching users at scale, and an agent working in many steps burns many times the tokens of a single prompt. Some rough division makes the point — 25 trillion weekly tokens over 8 million users is something like three million tokens per person per week, a load that reads as agents grinding all day, not people pecking out prompts now and then.
What Made Routing Worth So Much So Fast
TechCrunch frames the round as a structural hand-off: the industry’s workload has already swung “from training to inference to, now, agents.” The moment a product runs around the clock instead of answering one prompt at a time, tool calls and subtasks each pull toward a different model — the budget ones soak up routine steps, and the premium ones are spent only where judgment gets hard. The model fades into an unseen, interchangeable engine.
Staying un-locked-in is the other motive, and the industry read the report cites is blunt — companies “have no plans to get locked into a model vendor as they did with their various SaaS providers,” and the same crowd declares “the multi-model future is already here.” On that road, the toll booth belongs to the routing layer. It echoes what we wrote at the start of 2026 in our opening outlook — models shipping faster and vendors splintering further end up rewarding whoever holds the abstraction stacked above them.
The Developer Takeaways
One: the unified API has fresh capital-markets certification — handing integration to one abstraction instead of wiring into each provider now stands as the default architecture among young companies and mid-sized teams, and nobody calls it a lazy shortcut anymore. Two: CapitalG belongs to Alphabet itself, and its decision to get out front of a router that spans vendors means the cloud giants, too, have accepted that models can be swapped — deflating the lock-in on which vendor pricing has long rested. Three: in the agent era, keeping costs down will happen at the routing layer, where per-task model choice has gone from nice-to-have to table stakes for staying alive. The loose thread is margin: a gateway only works as a business while models stay commodities, and history has not been kind to commodity layers trying to hold pricing power.
The open question from here is whether gateways concentrate into a content-delivery-network style market — a few neutral middlemen through which every serious product sends its traffic — or whether the biggest model vendors start bundling the routing function themselves, crowding out the independents. That $1.3 billion price tag is, in the end, a wager that the first scenario wins.