Anthropic PBC has unveiled a $50 billion initiative to construct data centers across the United States.
The AI company plans to open new facilities in Texas and New York next year, with additional sites slated for future rollout. According to Anthropic, these data centers will feature custom-designed infrastructure optimized specifically for its AI workloads.
Anthropic is partnering with Fluidstack Ltd., a relatively low-profile AI infrastructure startup, for this ambitious project. Weeks ago, Forbes reported that Fluidstack had raised approximately $25 million in funding. Additionally, The Information revealed that the company secured over $10 billion in credit lines from a consortium of institutional lenders.
Notably, Anthropic opted for Fluidstack over larger AI infrastructure providers like CoreWeave Inc., which went public in April. One key factor behind this decision appears to be Fluidstack’s proprietary software tools. The company claims its platform automates much of the manual effort traditionally required to manage GPU clusters.
Fluidstack developed a custom operating system called Atlas OS, enabling administrators to deploy GPU clusters with just a few clicks. The system also handles ongoing maintenance tasks—such as server reboots—automatically. Users can further enhance Atlas OS by integrating orchestration tools like Kubernetes and Slurm.
Another core offering from Fluidstack is its Lighthouse software, which monitors GPU clusters for technical issues and initiates automatic repairs. In cases where a server cannot be restored, Lighthouse seamlessly migrates its workloads to another machine.
While Anthropic hasn’t disclosed which specific chips will power its new data centers, the company’s website indicates it currently offers AI clusters based on high-end NVIDIA GPUs such as the GB200, B200, and H200—suggesting its upcoming facilities will likely rely on NVIDIA hardware as well.
Anthropic has not yet detailed how it intends to finance the $50 billion investment. The company remains unprofitable and has raised roughly $33.7 billion from investors to date. It may seek additional equity or debt financing to support this infrastructure expansion.
“We are getting closer to AI systems that can accelerate scientific discovery and tackle complex problems in ways previously unimaginable,” said Dario Amodei, CEO of Anthropic. “Realizing this potential demands infrastructure capable of sustaining continuous advancement at the frontier.”
This announcement follows closely on the heels of Amazon Web Services’ recent launch of Project Rainier—a $11 billion data center campus built specifically for Anthropic. By year-end, the site is expected to house one million AWS Trainium2 chips. Over the long term, AWS plans to construct 23 additional buildings on the campus, scaling its total computing capacity to 2.3 gigawatts.
