Google Backs $15 Billion Anthropic Texas AI Campus with 1.6GW On-Site Power
With global AI infrastructure spending now exceeding $1 trillion, the race to scale compute is increasingly constrained by power availability rather than chip supply.
The story
In a significant move to secure future AI compute capacity, Google is set to back a $15 billion AI campus for Anthropic in Hubbard, Texas. The project, currently in advanced financing talks led by Morgan Stanley, will feature a substantial 1.6 gigawatt on-site natural-gas power plant.
Google will guarantee Anthropic's lease and power-payment obligations, acquiring approximately 20% equity in the data center and power generation venture. This initiative highlights the critical pivot by hyperscalers towards direct investment in energy infrastructure to meet the escalating demands of AI workloads.
Anthropic intends to equip the facility with its custom TPUs, co-designed with Google and Broadcom. The deal underscores the industry's shift as power and physical infrastructure become the primary bottlenecks, with server rack loads for AI inferencing surging up to 150 kW.
Silicon
Instinct MI455X / MI430X
Maker: AMD
What: Next-gen data center GPUs with HBM4 memory, optimized for frontier AI/AI factory deployments (MI455X) and sovereign AI/HPC (MI430X).
For Whom: Hyperscale AI factories, sovereign AI, and scientific computing centers.
Helios Rack-scale Solution with Cerebras Wafer-Scale Engine
Maker: AMD & Cerebras Systems
What: Disaggregated AI inference solution designed for ultra-low latency and increased throughput.
For Whom: Advanced AI applications requiring high performance and efficiency.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Data Center Capacity Agreement | AMD and Core Scientific | Up to 2.5 gigawatts (initial 500+ MW starting 2027) | United States |
| Manufacturing Expansion | United Microelectronics Corp (UMC) | $2 billion capex increase for 2026 | Taiwan (Tainan campus) and Singapore |
| AI Data Center Campus | CleanCore Solutions, Inc. (ZONE) | 55 MW utility power capacity, 40 MW critical IT load, $800M initial contract value | Strategic U.S. markets |
Supply & policy signals
TSMC increased its 2026 capital expenditure budget to $60-64 billion.
Implication: Reflects stronger-than-expected structural demand for AI and high-performance computing, with 70-80% allocated to leading-edge processes.
Taiwan's semiconductor power demand is projected to reach 52 billion kWh in 2026.
Implication: AI-related expansion is significantly increasing electricity demand for wafer fabrication, memory, and advanced packaging.
Power delivery and electrical equipment supply chain are the primary constraints for U.S. data center growth.
Implication: Lead times for large power transformers average 128 weeks, with prices up 77% since 2019, delaying capacity deployment.
What we'll be watching
- Cipher Digital Inc. is scheduled to release its second-quarter 2026 financial results on August 4, 2026.
- Nvidia's Rubin architecture, the next-generation AI chip platform, is scheduled for release in 2027.
- Intel's 2027 capital expenditure is expected to be "significantly above 2026 levels."
- The public preview for Microsoft's agentic security platform, Project Perception, opens on August 3.
Reporting + analyst voices: grounded via Google Search at publish time.