Anthropic, OpenAI Seek Smaller Data Center Deals
Anthropic, OpenAI Seek Smaller Data Center Deals
Anonymous
Key Points
1
The smaller deployments may reflect a broader shift from centralized hyperscale model training toward distributed infrastructure for inference and serving AI models to end users.
2
Both companies generally rent computing capacity from data-center operators and specialized “neocloud” providers, while separately pursuing larger, long-term infrastructure commitments.
3
Anthropic did not respond to CNBC’s requests for comment on the reported discussions, while the sources describing the talks spoke anonymously because they involved private business dealings.
4
A source-based account characterized the smaller strategy as potentially involving facilities with hundreds of GPUs that could become operational within quarters rather than the years often required for hyperscale campuses; one infrastructure executive contrasted waiting three years for a hyperscale facility with bringing 500 H100 GPUs online in six months.
OpenAI and Anthropic are hunting smaller data-center deals of 20 to 30 megawatts as they race to deploy AI computing power faster Blockspace. The two AI labs are exploring sites in the United Kingdom, Nordic countries, and the United States—a shift from their massive, gigawatt-scale infrastructure projects that can take years to build.
Smaller facilities could go live within months rather than years, letting the companies serve AI models to users sooner TS2 Tech. Both firms continue pursuing their massive long-term deals, but these quick-turnaround sites fill a critical gap in their infrastructure strategy.
Why AI Companies Need Speed Over Scale
A 500-GPU facility can launch in six months, versus three years for a hyperscale campus Blockspace. Smaller deployments reflect a shift: AI labs now need distributed computing for inference—running trained models for end users—not just massive training operations. "Hundreds of GPUs" in modular facilities deliver faster returns on investment.
Anthropic's Global Hunt for Power and Speed
Anthropic is exploring deployments across three regions: the UK, Nordic countries, and the United States TS2 Tech. The company is simultaneously pursuing a major long-term deal—about 460 megawatts from Nscale at a planned West Virginia data center. This dual approach lets Anthropic grab immediate capacity while securing huge future infrastructure.
OpenAI's Diversified Computing Portfolio
OpenAI said it is building a "diversified global computing portfolio," evaluating infrastructure based on workload, performance, reliability, timing and cost Blockspace. Like Anthropic, OpenAI is eyeing smaller deals in the Nordics and potentially the U.S., while separately committing to massive hyperscale projects. The company generally rents computing capacity from data-center operators and specialized neocloud providers.
The Broader Infrastructure Arms Race
Both companies rent most computing power rather than owning facilities outright News SSB Crack. Crusoe, an AI infrastructure startup, has raised $3.9 billion and is expanding modular, factory-built data centers for AI inference workloads TechSpot. The move toward smaller, faster deployments signals that the AI industry's bottleneck is no longer just raw capacity—it's getting that capacity online before competitors do.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
14
Articles
13
Reach
27
A
AnonymousAuthor
