For two years, the AI data centers story has been a contest of size — who can sign the biggest gigawatt campus, the largest multi-billion-dollar cloud contract. This week the story quietly flipped. According to a CNBC report published Friday, Anthropic and OpenAI — the two labs behind the most expensive infrastructure deals in tech history — are now hunting for small data center deals in the 20-30 megawatt range. Yes, the giants want bite-size capacity now. And the reasons why tell you a lot about where the AI industry is heading.
01.What the report actually says
CNBC spoke to four people familiar with the conversations, who said Anthropic has sounded out 20-30 MW agreements across the UK and the Nordics, while OpenAI has explored similar-sized deployments in the Nordics. At least one source was also aware of talks involving both companies at that scale in the US. To be clear, nothing is signed — these are private commercial discussions — but the pattern is consistent across multiple sources.
An OpenAI spokesperson didn't confirm specific talks but told CNBC: "We're building a diversified compute portfolio to meet growing demand for AI around the world. Different workloads need different infrastructure." That last sentence is the whole story in seven words.
02.Why smaller suddenly makes sense
The analyst phrase doing the rounds is "speed to usable capacity." Jabez Tan, head of research at Structure Research, put it plainly: "Securing a few megawatts at an existing powered site can be more practical than waiting for a much larger block in one location." Large builds across the US are facing delays — power hookups, supply chains, community pushback. A 20-30 MW slot at a site that already has electricity can go live in months, not years.
And if your workload can be split across locations — which is increasingly true for inference, the part of AI that actually answers your questions rather than the part that trains the model — a collection of smaller deployments adds up to serious capacity. It's the same logic as microservices, but in concrete and megawatts.
03.The mega-deals haven't stopped
Don't mistake this for the big projects slowing down. In August, Anthropic signed a roughly $45 billion cloud deal with Nscale covering around 460 MW at a development in West Virginia. OpenAI, meanwhile, said in April it had already surpassed the original 10 GW commitment for its Stargate infrastructure project, and has since committed to a further 3 GW in Georgia and 8 GW in Ohio. The appetite for scale is enormous — we tracked the money when OpenAI's valuation crossed $1.2 trillion in our earlier coverage.
What's new is that the labs are adding a second, faster lane alongside the highway. Big campuses for the heavy lifting; small powered sites for quick deployment. It's a hedge, and it's also an admission: the bottleneck is no longer money or chips — it's time and electricity.
04.Crusoe's $3.9 billion bet on truck-sized data centers
The most concrete symbol of this shift is Crusoe, the eight-year-old company that built the huge Abilene, Texas data center complex used by OpenAI. On Thursday it announced a $3.9 billion Series F at a $30.9 billion post-money valuation — roughly triple its $10 billion valuation from just eleven months earlier. The round was co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, with Founders Fund, GIC, Nvidia, the Qatar Investment Authority, Radical Ventures and TPG participating.
Part of the money goes to the usual mega-campuses. But the interesting slice funds Spark — modular data centers built in a factory, transported by truck, and connected to existing power sources almost anywhere. Think shipping containers full of GPUs that plug into the grid in weeks instead of the 18-24 months a traditional build takes. Crusoe now claims more than $140 billion in contracted value, over 6 GW of gross contracted capacity with 1 GW live, a $13 billion five-year deal with Jane Street, and an inference product that has crossed $100 million in annual recurring revenue. The Wall Street Journal also reported Crusoe is investing in smaller facilities precisely because they're faster and cheaper than the big builds everyone else is stuck waiting on.
05.Why this matters for India — and for your AI bill
Here's the thing nobody's talking about: this shift is ultimately about inference — the compute that runs every chatbot answer, every AI-generated image, every agent task. When the biggest AI buyers on earth reorganise their infrastructure around cheaper, faster inference capacity, the unit economics of AI products improve for everyone downstream. That means faster rollouts of AI services in more places, including price-sensitive markets like India. We've already seen how seriously global players take Indian AI demand — remember the Google-Nvidia data center partnership for India we covered earlier.
For Indian developers and startups building on top of AI models, the practical lesson predates all this news: compute is your biggest variable cost, and how you manage it decides whether your AI side project becomes a business or a credit card bill. There's a reason the GPT-6 Astra — Zero to Hero Guide spends a full chapter on cost optimization — the same prompt done right versus wrong can be the difference between ₹8,000 and ₹80,000 a month on API bills.
Key Takeaways
- Anthropic and OpenAI are exploring 20-30 MW data center deals in the UK, Nordics and the US, per CNBC's sources
- "Speed to usable capacity" is the driver — powered sites beat waiting on delayed mega-builds
- The big deals continue: Anthropic's ~$45B Nscale pact (~460 MW) and OpenAI's Stargate expansions in Georgia and Ohio
- Crusoe raised $3.9 billion at a $30.9 billion valuation to build "Spark" — truck-deliverable modular AI factories
- The trend points to inference workloads spreading out — faster, cheaper AI capacity in more locations
Frequently Asked Questions
Why are Anthropic and OpenAI suddenly interested in small data center deals?
According to a CNBC report citing four people familiar with the conversations, both labs are exploring 20-30 megawatt agreements — Anthropic in the UK and Nordics, OpenAI in the Nordics — because of “speed to usable capacity”. A small slot at a site that already has electricity can go live in months instead of years, which suits inference workloads that can be split across locations.
Does this mean the mega AI data center projects are slowing down?
No. In August, Anthropic signed a roughly $45 billion cloud deal with Nscale covering around 460 MW in West Virginia, and OpenAI has already surpassed its original 10 GW Stargate commitment with a further 3 GW planned in Georgia and 8 GW in Ohio. The small deals are a second, faster lane running alongside the big builds, not a replacement.
What is Crusoe's Spark?
Spark is Crusoe's line of modular data centers built in a factory, transported by truck and connected to existing power sources almost anywhere — capacity that can plug into the grid in weeks instead of the 18-24 months a traditional build takes. Crusoe announced a $3.9 billion Series F at a $30.9 billion post-money valuation to fund it, and says it now has over $140 billion in contracted value.
What This Means for You
If you're building anything with AI in India — a SaaS product, an agency, a student project that outgrew free tiers — the infrastructure race happening in the Nordics and West Virginia will eventually show up in your API bill, and in a good way. More distributed, faster-to-deploy capacity means the AI services you build on get more reliable and cheaper over time, not less. But don't wait for that to rescue a badly architected product: prompt caching, model selection and workflow design are things you control today. The GPT-6 Astra — Zero to Hero Guide covers those cost-optimization techniques step by step, along with prompting, agents and business ideas — practical skills that matter far more than the size of anyone's data center.