Google is closing in on a multibillion-dollar deal to finance a data center in Texas leased to Anthropic, the company behind Claude. The initial phase of the project could exceed $5 billion, with Google expected to provide construction loans for Nexus Data Centers, the operator of the site, according to the Financial Times via Cointelegraph.
The deal underscores how much raw compute infrastructure the current generation of AI agent platforms requires. Anthropic’s Claude now runs computer-use capabilities, autonomous desktop agents, and enterprise-grade API workloads. Those are all inference-heavy, long-running tasks that consume significantly more compute per session than a standard chatbot query.
The Project Details
The 2,800-acre campus is already under construction. Anthropic signed the lease earlier this month, and the site has received early-stage debt from asset manager Eagle Point, LiveMint reported. A consortium of banks is competing to provide financing by mid-year.
The facility is expected to deliver roughly 500 megawatts of capacity, enough to power approximately 500,000 homes, with eventual expansion plans reaching 7.7 gigawatts, according to Times Now. Google’s involvement through its parent Alphabet should allow the project to raise financing at lower rates due to the company’s strong credit rating, per the Financial Times report.
Energy Strategy
The Texas location sits near major gas pipelines operated by Enterprise Products Partners, Energy Transfer, and Atmos Energy. Nexus plans to run its own gas turbines on-site, reducing reliance on grid power and avoiding peak-hour pricing, LiveMint reported.
This behind-the-meter approach is becoming standard practice. Roughly 50 gigawatts of behind-the-meter projects were announced in 2025, according to research provider Cleanview as cited by the Financial Times. Elon Musk’s Colossus data centers in Tennessee and Mississippi use off-grid gas turbines, and both Amazon and Microsoft have announced large behind-the-meter projects, per Times Now.
What $5 Billion Says About Agent Infrastructure
The scale of this investment reflects a specific reality about agent workloads. A chatbot query completes in seconds. An autonomous agent session — browsing a desktop, executing multi-step workflows, maintaining context across long task sequences — can run for minutes or hours. That translates directly into GPU-hours per user session, and it multiplies quickly as enterprise deployments scale.
The Electric Reliability Council of Texas estimates that data center energy usage in the state could reach 78 gigawatts by 2031, about 36% of total state power demand, according to the Financial Times report. This facility is one piece of that growth.
For competing agent platforms, the Anthropic-Google deal sets a clear benchmark: running autonomous agents at enterprise scale requires infrastructure commitments measured in billions, not millions. Any agent platform that expects to compete on reliability, latency, and availability will eventually need compute capacity in the same range.