Anthropic to invest $50 billion in U.S. AI data centers with new sites in Texas and New York

Anthropic said Wednesday it’s investing $50 billion into building AI data centers across the United States, starting with two major custom sites in Texas and New York, according to a press release.

The facilities, which are part of the company’s wider push to stay competitive in the AI infrastructure war, are being built in partnership with Fluidstack, an AI cloud firm that already powers GPU compute for clients like Meta, Midjourney, and Mistral. The buildout will kick off immediately, with both locations expected to go live in 2026.

This isn’t some small pilot project. Anthropic is building facilities designed to handle gigawatts of power, focused entirely on training and running its AI models at scale.

The company is targeting its enterprise clients, the ones responsible for most of its revenue, and these new sites are built to match the size of that demand.

Dario Amodei, the company’s CEO, said the goal is to push forward on developing systems that can “solve complex problems” and “drive scientific discovery,” but that none of it is possible without physical infrastructure to support that growth.

“These sites will help us build more capable AI systems that can drive those breakthroughs, while creating American jobs,” Dario said.

Texas and New York break ground as Anthropic’s AI race accelerates

The plan includes 800 permanent jobs and more than 2,000 roles in construction, with even more locations expected to follow later.

By launching these new facilities inside the U.S., Anthropic is responding directly to Washington’s growing demand for domestic compute power.

There’s been rising pressure from federal officials and industry watchers to bring more AI infrastructure back onshore, especially as concerns grow about tech dependence on overseas suppliers.

While Anthropic is expanding, its rival OpenAI is already deep into its own buildout. That company has lined up more than $1.4 trillion in long-term infrastructure deals with Nvidia, Broadcom, Oracle, Microsoft, Google, and now Amazon.

But that much spending has raised new questions in D.C. and across the industry; namely, whether the country’s power grid and construction capacity can even keep up.

Some analysts have even asked if the AI industry is ballooning into a financial bubble. Anthropic, for now, is staying quiet on that debate and focusing on building what it says is “needed to support frontier-level AI.”

The company currently supports over 300,000 businesses, with the bulk of that usage coming from enterprise accounts. Those accounts spending more than $100,000 annually have jumped nearly seven times higher in the past year alone.

Internal projections reviewed by The Wall Street Journal show that Anthropic expects to break even by 2028, which puts it ahead of OpenAI, which is still projecting $74 billion in operating losses in that same year.

Indiana site goes live as government funding debate heats up

To stay on track, Anthropic brought in Fluidstack to design and build custom facilities built specifically for its AI workload types.

The company says Fluidstack can deliver on short timelines, with enough power to support GPU‑intensive model training and deployment.

That relationship is already showing results. In Indiana, Amazon has opened a $11 billion, 1,200-acre data center campus exclusively for Anthropic.

That site is already operational, giving the company a head start on real-world deployment, while many competitors are still only drafting proposals.

Beyond the private buildouts, there’s a growing debate over how much the U.S. government should be involved in financing the AI infrastructure boom.

Just last week, OpenAI formally asked the Trump administration to expand CHIPS Act tax credits to cover AI data centers and key power grid parts like transformers.

That request came days after Sarah Friar, OpenAI’s CFO, floated the idea of a government backstop to protect some of its big compute investments.

The backlash was quick and sharp, forcing the company to walk back the comments. But the damage was done, and the fight over who pays for America’s AI buildout is just getting louder.