Nvidia and Stripe wrote nine- and ten-figure cheques for AI infrastructure and plumbing this week, even as a model small enough for a laptop matched what cloud labs still charge a premium for — so what exactly are today's biggest AI bills actually buying?
Nvidia guarantees $105bn to secure OpenAI's Ohio mega-campus
Nvidia isn't just selling chips to OpenAI anymore. It's underwriting the building that will house them.
- Nvidia will guarantee up to $105bn in lease payments for a SoftBank-backed SB Energy data centre in Pike County, Ohio, and is investing $1.5bn directly in SB Energy itself, Nikkei Asia reports.
- The campus starts at 4.25 gigawatts with an option to grow to 8GW; SB Energy plans a 9.2GW gas power plant on land once used to enrich uranium for the US nuclear arsenal, Nikkei notes.
- OpenAI's total spending on Nvidia's chips is projected to reach roughly $600bn through 2030, according to Nvidia CEO Jensen Huang.
Why it matters: Nvidia is now financing the customer that buys its own product — a circularity that will draw louder questions if AI demand cools.
Anthropic's revenue run-rate triples to $65bn as IPO looms
- Anthropic's annualised revenue run rate passed $65bn at the end of July, up from $47bn in May and just $9bn at the end of last year, Bloomberg reported, via TechCrunch.
- Investors expect the company to finish 2026 between $100bn and $120bn, according to the Financial Times; rival OpenAI's revenue has doubled to $40bn since late 2025.
- Both companies have filed confidential IPO paperwork, with Anthropic expected to list first, seeking a public valuation of $2 trillion or more, TechCrunch reports.
Why it matters: A $2 trillion target valuation resets what "successful AI IPO" means for everyone who lists after it.
A 27-billion-parameter model just ran frontier coding on a laptop
- Alibaba released Qwen3.8-27B on Hugging Face under an Apache 2.0 licence; quantised down, it fits in roughly 17GB — small enough for a high-end gaming desktop, VentureBeat reports.
- Artificial Analysis scored it 52 on its Intelligence Index — matching OpenAI's cloud-only GPT-5.6 Luna — and 51 on its Agentic Index, ahead of Anthropic's Claude Opus 4.8, per VentureBeat's reporting.
- Developer Simon Willison ran the compressed version on an M5 Max MacBook Pro and an Nvidia DGX Spark with no cloud connection at all, VentureBeat notes.
Why it matters: capability that needed a frontier lab's cluster months ago now runs on hardware nobody's vendor controls, logs, or bills by the token.
Stripe pays up to $8bn for the plumbing that routes AI traffic
- Stripe has agreed to acquire OpenRouter, the AI model marketplace, for more than $7bn according to Bloomberg, with Axios pricing the deal closer to $8bn.
- OpenRouter had raised just $164m to date, most recently in May at a $1.3bn valuation, Axios reports.
- The platform routes requests across more than 400 models for around 8 million developers, per Forbes contributor Sandy Carter's account of the deal.
Why it matters: leverage in AI is quietly migrating from labs that build models to the layer that decides which model gets the call.
Scotland lands Dell and a £300m data-centre expansion in Lanarkshire
- The UK's National Wealth Fund is providing a £202m guarantee, unlocking a £300m financing package from ING, ABN AMRO, Santander, the Scottish National Investment Bank and Siemens Financial Services, the UK government announced.
- The funding lets developer DataVita expand its DV1 data centre and build a second facility in the Lanarkshire AI Growth Zone.
- Dell Technologies will base its Scottish team at the Lanarkshire AI Innovation Park; the development is expected to support more than 3,400 jobs, per the announcement.
Why it matters: proof the UK's AI Growth Zone model can pull in private finance and a name-brand tenant outside the London-Thames Valley corridor.
The Hexalink view
Money is concentrating in infrastructure and distribution — data centres, power plants, routing layers — just as raw model capability becomes a commodity you can download free and run on a laptop. Worth noting: the same week, OpenAI quietly disbanded the team that assessed catastrophic model risk ahead of an expected IPO. Scale and safety staffing are moving in opposite directions.
If you're building an AI strategy now, don't assume next year's advantage is a better model — assume it's better infrastructure economics and better control over which model runs where. Audit compute and data-routing contracts as seriously as model selection, and don't treat "open and local" as fringe; Qwen3.8-27B suggests it's fast becoming the default for cost-sensitive, high-volume work.
We'll be back tomorrow with the next briefing — or catch the five-minute audio cut on the AI Storm Daily podcast if you'd rather listen on the commute.