Nvidia Lost OpenRouter to Stripe, Then Spent $140B in Weeks

By Daniel Rao

The fastest-growing layer of the AI stack just slipped through Nvidia's fingers. Not the chips, not the models — the toll road in between. And the kicker? Nvidia knew it wanted OpenRouter, told the startup so, and still managed to lose it. What followed is one of the most colossal spending sprees in corporate history.

The Information reported this week that back in July 2026, Nvidia learned OpenRouter — the three-year-old AI model marketplace that has become developers' default way of routing traffic between hundreds of models — was about to be acquired. CEO Jensen Huang wanted in. Nvidia executives told OpenRouter's leadership they were prepared to make a generous offer, according to reporting by The Information's Valida Pau and Phoebe Liu. There was just one catch: they needed more time to evaluate the deal.

OpenRouter's co-founder and chief executive, Alex Atallah, would not wait. Stripe closed the deal for more than $8 billion in cash and stock, and Nvidia never submitted a formal offer. Databricks, incidentally, had also been talking to the startup. It is a rare thing to watch the most valuable chip company on earth get outmanoeuvred on speed — the one currency it normally prints itself.

The bid that never was

Let us be precise about what The Information actually established, because this story is built on unnamed sources and should be read with the attribution pinned to the wall. In July, Nvidia told OpenRouter it was interested in competing for the marketplace but needed more time, the outlet reported. A person familiar with the negotiations said Nvidia executives were prepared to make a generous offer but required more time for evaluation before tabling a formal bid. Atallah declined to slow down. Stripe got the signatures.

There is a particular kind of sting here. This was not a failure of imagination — Nvidia wanted OpenRouter — and it was not a failure of capital; the numbers coming next will make that absurdly clear. It was a failure of tempo. Deals get done by the decisive, and in July 2026, a payments company moved faster than a company that sells time itself by the clock cycle.

It is also worth remembering who else was sniffing around. Databricks had been in talks with OpenRouter too, per the same reporting — which tells you this was not a Stripe-only romance but a contested asset that multiple infrastructure giants had done the maths on.

Why everyone wanted the router

Step back for a moment and ask why a payments processor and a chipmaker both wanted a routing startup. OpenRouter's pitch is almost comically simple: instead of writing and maintaining separate integrations for every AI provider, developers write one integration, point it at OpenRouter's endpoint, and let the platform handle routing, failover and unified billing across hundreds of models.

The numbers explain the feeding frenzy. OpenRouter announced a $113 million Series B in May 2026, led by CapitalG — Alphabet's growth fund — at a post-money valuation of about $1.3 billion. The round was stacked with strategic money: NVentures (Nvidia's own venture arm), ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures and Databricks Ventures all joined, alongside existing backers Andreessen Horowitz and Menlo Ventures. NVentures, note: Nvidia was already an investor in the company it then tried and failed to buy outright.

Since May, weekly token volume had surged fivefold, from 5 trillion to 25 trillion tokens. OpenRouter claimed 8 million users in May and, by the time of the Stripe announcement, a community exceeding 10 million developers, handling more than 10 trillion tokens per day across more than 400 models. The multi-model future — no lock-in, best model for every task — is not a thesis any more. It is traffic, and traffic gets priced.

And here is the detail I cannot get over: OpenRouter's Atallah had described his own company, back in May, as "the equivalent of Stripe for AI". The Stripe-for-AI company got bought by actual Stripe. If you wrote that as fiction, an editor would send it back for being too neat.

$7.5 billion, $8 billion — pick your number, none of them official

Readers deserve the honest mess on the price, because it is a shoddy-looking piece of journalism if a paper just picks one figure and moves on. The August announcement cycle produced a spread: Bloomberg first reported a deal of more than $7 billion; SiliconANGLE reported a ~$7.5 billion deal on August 19; the Wall Street Journal had earlier floated a roughly $10 billion figure in July — meaning the final price landed about 30% below the first rumour.

Then Reuters reported the transaction at slightly more than $8 billion, and Axios separately put it above $8 billion. The Information's own headline calls it "$8B". Neither Stripe nor OpenRouter disclosed terms, and Stripe's standard line — "we do not comment on rumors or speculation" — stayed firmly in place.

We cross-checked the figures cited in this article against Reuters' August 17–18 wire reporting and the companies' own statements before writing; wherever the numbers diverge, the attribution tells you exactly which source said what. Either way, OpenRouter sold for more than five times the $1.3 billion valuation it held just three months earlier. That is not a price. That is a verdict on what the distribution layer is worth in 2026.

Photo: Nvidia CEO Jensen Huang at Supercomputing 2018. (Image: Raysonho via Wikimedia Commons, CC0)

The consolation spree: $140 billion in eight weeks

Nvidia did not sulk. Over the two months after the OpenRouter miss, the company finalised transactions totalling more than $140 billion. Read that again. Most companies never do $140 billion in acquisitions in their entire existence; Nvidia did it in roughly the time it takes to get a passport renewed.

The anchor is real and independently verified. In August, Nvidia guaranteed up to $105 billion in lease payments to help OpenAI secure a sprawling campus at the PORTS-Pike Technology Campus in Pike County, Ohio — a reindustrialised former uranium enrichment site where SB Energy, a SoftBank subsidiary, will build, own and operate data centres under a 20-year lease to OpenAI. Nvidia becomes the exclusive AI-compute provider, and it put $1.5 billion directly into SB Energy alongside SoftBank and OpenAI.

The scale is genuinely hard to process: 8 gigawatts of IT capacity, 10 gigawatts of new generation, 35,000 construction jobs, phased from 2028 through 2032. Huang, announcing it, said: "We are securing long-lived infrastructure for NVIDIA compute so OpenAI can deploy the most productive AI factories that can be upgraded repeatedly." OpenAI chief executive Sam Altman added: "This is going to be a huge site, with enough computing power to help millions of people use AI to do things we can only start to imagine today."

Reuters, which first nailed the figure, also noted what everyone in finance is whispering: deals like this raise questions about circular funding flows between the chipmaker and its biggest customers. Nvidia finances the customers who buy Nvidia chips. That is either a flywheel or a ouroboros, and it is fair to say the market has not fully decided which.

Beyond Ohio, The Information reports that by the end of July Nvidia held nearly $100 billion in equity investments and carried another $25 billion in future investment commitments. That is balance-sheet warfare — part of Nvidia's expanding portfolio of AI bets — and it explains why the OpenRouter whiff barely registered inside Santa Clara.

Hugging Face: the bigger catch

Here is the twist that reframes the whole story. In September, Nvidia agreed to acquire Hugging Face for $12.93 billion — its largest acquisition ever — and it had been negotiating for weeks while the OpenRouter story was still playing out. Eighteen million developers. Three million models. The default first address of the open-source AI world, now owned by the chipmaker.

Huang moved fast to calm the obvious fear: "Hugging Face will remain an open platform for the entire AI ecosystem," he said at the announcement, adding that developers would remain free to choose their models, frameworks, cloud providers and computing platforms. The deal structure tells its own story — roughly $11.9 billion for the company plus up to $1 billion in equity-based employee retention, and six-year retention terms for the co-founders. Nvidia was not buying code. It was buying trust, and it priced it like a commodity that cannot be replaced.

Which makes the next detail curious, and we would be doing readers a disservice to smooth it over. The Information's latest reporting lists Hugging Face — along with Perplexity — as among Nvidia's further targets. But Nvidia announced the Hugging Face acquisition on September 3, a full five weeks before The Information's piece. Either the "targets" framing refers to follow-on moves in the same orbit, or the reporting is stale on this point. Flag it and move on — but flag it honestly, because gloss is how thin stories pretend to be thick ones.

For the US audience, the OpenRouter connection to our earlier reporting on Chinese models dominating developer traffic is worth keeping in mind: Stripe now owns the router through which an enormous share of that traffic flows, including a large volume of Chinese-origin models. That makes the toll-road ownership question geopolitical as well as commercial.

What comes next: Figure, Perplexity, and a $10 billion question mark

The Information's forward-looking read says Nvidia is in discussions over a new investment in Figure, the humanoid-robotics company, as part of a push into physical AI — robotics, autonomous driving, and models that run on local devices rather than in the cloud. Multiple bankers, lawyers and investors close to Nvidia reportedly confirm the direction.

Now, a word of caution we refuse to skip. A widely circulated version of this story says Nvidia is discussing a "$10 billion investment" in Figure. We could not corroborate that figure against The Information's actual reporting — the most specific secondary read puts the figure at roughly $1 billion more, with Figure valued around $38 billion going into the round. The one hard "$10 billion Nvidia" figure we could verify belongs to a different story entirely: Reuters reported in September that Nvidia was considering up to $10 billion as an anchor investor in Anthropic's prospective IPO. Conflating the two would be exactly the kind of error this column exists to avoid, so treat any "$10B for Figure" claim as unverified until a primary source confirms it.

Perplexity as a target is likewise single-sourced to The Information. Interesting, but file it under "watch", not "fact". What is well established is that Nvidia is actively hunting across robotics and on-device models amid the wider AI-chip funding rush — a sensible hedge for a company whose own disclosures show just three customers accounting for 44% of sales in the six months through July. Concentration that extreme is precisely why the toll roads matter: owned infrastructure diversifies the customer base without selling a single extra chip.

Photo: Nvidia's headquarters in Santa Clara, California. File photo — not the Pike County, Ohio campus backed by Nvidia's $105 billion credit support. (Image: Coolcaesar via Wikimedia Commons, CC BY-SA 4.0)

Our take: owning the toll road is the whole game

Step back and the pattern is unmissable. In roughly six weeks, buyers committed around $21 billion to two companies — Hugging Face and OpenRouter — that do not build the best models. They sit between the model makers and the developers who pay. These are toll-road purchases: payment for owning the point where AI usage flows, in exchange for the power to collect from everyone who crosses.

Our judgment is blunt: Stripe got the better of this exchange. Nvidia is betting that owning distribution will funnel the world onto Nvidia silicon; Stripe is betting that owning distribution is the business, regardless of whose silicon sits underneath. One of those bets has to be wrong, because neutrality is what made these platforms valuable — and strategic owners convert neutrality into captive traffic. If Nvidia turns Hugging Face into an Nvidia showroom, or Stripe turns OpenRouter into a Stripe payments funnel, the developer flight will be swift. Open-source developers have long memories and short loyalty lists.

And Nvidia's failure on OpenRouter itself? Losing an $8 billion deal because you needed "more time to evaluate" is the kind of dithering that would get a startup CEO fired — except Nvidia then went and closed $140 billion of other deals, including its biggest acquisition ever. The honest reading is not that Nvidia is slow. It is that Nvidia, for the first time in its imperial phase, faced a seller who would not wait for it. Atallah looked at the most feared negotiating counterparty in tech and said: too slow, moving on. That takes some nerve — and it tells you everything about how badly everyone wanted this particular toll road.

Sources

Primary reporting: The Information (Oct 2026), via TheFly/TipRanks summary and the AI Weekly alert "Nvidia Lost OpenRouter to Stripe, Then Closed $140B in Deals". OpenRouter funding: TechCrunch (May 26 and Aug 16, 2026); The New York Times via TechCrunch on the $1.3B valuation. Stripe deal: Bloomberg, SiliconANGLE (Aug 19, 2026), Reuters, Axios via VentureBeat (Oct 2026). Nvidia–Ohio campus: Reuters (Aug 17–18, 2026), Nikkei Asia, Futurum Group, BetaNews. Hugging Face acquisition: Cointelegraph (Sept 2026), Law360. Anthropic IPO talks: Reuters (Sept 12, 2026). Nvidia's robotics/Figure discussions: The Information via a secondary read (Oct 7–8, 2026).

Daniel Rao covers the business of AI infrastructure for TechPulse USA — the deals, the data centres, and the toll roads in between.