CL-OP5X 913 +8.7%
GLM-52 901 +2.2%
GROK-46H 877 +2.9%
GEM-37FH 877 +17.7%
GPT-56T 861
GLM-5 856 +9.2%
QWEN-38X 843
MUSE-SPK 841 +0.7%
GPT-6A 820
KIMI-K3X 818 +10.1%
CL-OP5H 811 +0.6%
CL-FAB5H 798 +3.1%
GPT-56SC 791
CL-OP46H 753 +3.3%
CL-OP47H 745 +3.5%
GEM-38FH 683 +0.9%
CL-OP47 594 -0.8%
INKL 531
CL-OP46 498 -0.2%
CL-OP48 492 -0.2%
CL-OP5X 913 +8.7%
GLM-52 901 +2.2%
GROK-46H 877 +2.9%
GEM-37FH 877 +17.7%
GPT-56T 861
GLM-5 856 +9.2%
QWEN-38X 843
MUSE-SPK 841 +0.7%
GPT-6A 820
KIMI-K3X 818 +10.1%
CL-OP5H 811 +0.6%
CL-FAB5H 798 +3.1%
GPT-56SC 791
CL-OP46H 753 +3.3%
CL-OP47H 745 +3.5%
GEM-38FH 683 +0.9%
CL-OP47 594 -0.8%
INKL 531
CL-OP46 498 -0.2%
CL-OP48 492 -0.2%
← Back to feed

Google Finances $5B Anthropic Data Center in Texas — 500MW Now, 7.7GW Planned

Google has moved beyond TPU compute contracts with Anthropic into physical real estate. A data center financing deal structured through Nexus Data Centers and valued at over $5 billion puts Google in the position of landlord for Anthropic’s Texas compute base — a strategic deepening of a relationship that already includes a 14% equity stake acquired through $3 billion in investments between 2023 and 2025.

The Facility

The Texas site targets 500 megawatts of operational capacity by late 2026, with a stated expansion path to 7.7 gigawatts. That ceiling puts it in the same tier as national grid infrastructure — the equivalent of a mid-sized city’s total power draw.

The facility will run on direct natural gas via on-site turbines, bypassing the public grid entirely. This behind-the-meter power model has become the industry’s answer to AI’s most acute constraint: not chip supply, but power delivery at scale. Grid-connected sites face multi-year interconnection queues in most US markets. Behind-the-meter gas generation sidesteps that bottleneck entirely, at the cost of carbon intensity and regulatory exposure.

How This Differs From the TPU Deal

Earlier reporting covered Anthropic signing a gigawatt-scale compute contract with Google and Broadcom, with TPU capacity coming online in 2027. The Texas data center financing is a distinct transaction: it is about physical building and power infrastructure, not chip access. The two deals together represent a comprehensive dependency — Anthropic leans on Google for both the physical layer (Texas facility) and the silicon layer (TPU contracts).

Google holds a 14% equity stake and is now also the landlord. That concentration of structural dependency on a single hyperscaler is notable for a lab that markets itself as an independent safety-focused AI company.

The Infrastructure Commitment Landscape

The Texas facility sits alongside a wave of hyperscaler-led AI infrastructure expansion:

ProjectCapitalTimeline
Microsoft Stargate$500B4 years
Meta US Expansion$600BThrough 2028
Google–Anthropic Texas$5B+2026 build, 7.7GW ceiling
Anthropic Separate Commitments$50BMulti-year

The pattern across all of them is the same: the constraint is not money, and increasingly it is not chips. It is power — specifically, the ability to deliver gigawatt-scale electricity reliably to a single site. Behind-the-meter gas generation is the current industry answer, with grid interconnection queues making the public utility path impractical at the timescales AI labs are working on.

What Google Gets

The financing structure gives Google compounding leverage over Anthropic. Equity stake, compute contract, and now physical infrastructure financing — each layer makes the relationship harder to exit and aligns Anthropic’s operational survival with Google’s willingness to continue. Amazon’s $4 billion Anthropic investment, structured around AWS deployment, creates a competing dependency pull. Anthropic managing both relationships simultaneously, without ceding either, is the strategic balancing act that defines its next two years.

For the data center sector, the story is simpler: the AI labs’ capital commitments are large enough to reshape regional power markets, and the projects that move fastest are the ones that own their power generation rather than waiting for grid access.