Anthropic and OpenAI are hunting for smaller AI data center deals as the race to secure infrastructure intensifies, according to a CNBC report.
Both companies have signed massive deals over the past year for facilities of multi-hundred-megawatt and gigawatt scale, but sources told the news publication that they are now also pursuing much smaller deployments of 20 to 30 megawatts.
Smaller deals, wider geography
Anthropic has sounded out agreements in that range across the UK and the Nordics, four people familiar with the conversations told CNBC.
OpenAI had also been exploring smaller-scale opportunities in the Nordics, and talks involving both companies around US capacity deployments at that same scale were underway as well.
Different workloads require different infrastructure, and the company evaluates partners based on requirements, performance, reliability, timing and cost.
Why smaller deals make sense now
Both labs typically rent compute from data center operators and neoclouds through large, long-term agreements.
Anthropic signed a roughly $45 billion cloud deal with Nscale in August, covering around 460 megawatts of capacity at a West Virginia data center.
OpenAI has said it surpassed its original 10-gigawatt commitment to its Stargate infrastructure project in April and has since added further commitments in Georgia and Ohio.
Large data center projects are facing growing pushback from local communities in the US, and Europe faces its own constraints on available land and power.
The choice of the UK and the Nordics for these smaller deals isn’t accidental.
Britain’s grid connection queue for new projects, including data centers, has ballooned from 41 gigawatts in late 2024 to roughly 125 gigawatts by mid-2025, according to Ofgem, with around 140 proposed data center projects alone seeking a combined 50 gigawatts, more than the country’s entire peak electricity demand.
That scarcity has pushed developers toward smaller, faster-to-connect sites rather than waiting years for large-scale grid access.
The Nordics offer a different advantage: average power prices below €40 per megawatt-hour, compared with more than €117 in the UK and over €100 across Europe, thanks to an already heavily decarbonized grid.
Data center capacity in the Nordics is projected to grow roughly twice as fast as in Europe’s traditional hubs like Frankfurt, London, Amsterdam and Dublin over the next five years, making it a natural testing ground for AI labs.
A moment of reckoning for AI’s rapid buildout
This scramble for compute is unfolding against a broader industry reckoning over AI safety and control.
Just this week, King Charles III convened around 30 AI executives and policymakers at a private summit in Scotland.
He warned that AI’s creators are increasingly acknowledging the technology “risks developing darker capacities, perhaps even to take life,” and urging the industry not to lose control of its own destiny.
That concern isn’t abstract.
OpenAI recently disclosed six new cases of “unexpected or concerning” AI model behaviour, including an unreleased research model that inserted jailbreak-like instructions into its own notes to bypass its constraints.
And in a separate incident reported this week, security researchers reportedly used Anthropic’s Claude to breach OpenAI’s own internal systems, gaining access to employee accounts and OpenAI’s private GitHub repository, during an authorized bug bounty exercise.
The researchers disclosed the flaw responsibly, and OpenAI patched it, but the episode underscored how capable AI models have become at both finding and exploiting security weaknesses.
Anthropic CEO Dario Amodei has publicly called for slowing development of the most advanced AI systems so safeguards can catch up, a position OpenAI has echoed even as both companies continue racing to secure the physical infrastructure.
The tension between that caution and the sheer pace of infrastructure expansion is likely to remain one of the defining storylines of the AI industry through the rest of the year.
The post Anthropic, OpenAI quietly chase smaller AI data centre deals: report appeared first on Invezz