Anthropic has a public argument with the US government about autonomous weapons. Its new Texas data center runs on gas turbines.
Google is providing financial backing for a multi-billion-dollar Anthropic data center project in Hubbard, Texas, according to the Financial Times, with Nexus Data Centers as the operator. The project could reach $5 billion in total financing — a consortium of banks is competing to provide it by mid-2026, and Google's backing improves the credit rating and lowers the cost of that financing. The campus signed its lease earlier this month, with initial capacity of 500 megawatts deployable later this year, expandable to 7.7 gigawatts at full build-out.
That part is straightforward infrastructure news. The part that is not straightforward is the energy.
The site is located near a major gas pipeline, which enables Nexus to power the facility with natural gas turbines it owns. This is not a campus that runs on solar or wind. It is a campus that runs on fossil fuel infrastructure, sized to power a frontier AI training operation at a scale that will consume power at a rate most industrial facilities cannot approach.
In December 2025, Nexus proposed a data center campus for approximately 2,000 acres in Hubbard, Texas, expected to offer 612 megawatts powered by natural gas turbines. The project Anthropic leased this month spans 2,800 acres — a larger footprint than the December proposal, alongside a buildout ceiling of 7.7 gigawatts compared to the original 612 megawatts. What changed between the proposal and the lease, and why the land footprint grew to accommodate the larger power capacity, is not explained in the available reporting.
Anthropic has spent the past two years making the case that it is different from other AI labs. It refused to allow its technology to be used for mass surveillance and autonomous weapons. On February 27, 2026, President Trump directed federal agencies to cease using Anthropic products and Defense Secretary Hegseth designated the firm a supply chain risk. That position is now the subject of an active lawsuit with the US Department of War. On March 26, 2026, Judge Rita Lin granted a preliminary injunction blocking the government punitive measures, finding the Pentagon designation likely contrary to law and arbitrary and capricious. Anthropic warned in a financial filing that the designation could cost it multiple billions of dollars and impair its ability to train future models if investors pull back. A partner with a multi-million-dollar annual contract has already switched from Claude to a rival model, eliminating an anticipated revenue pipeline of more than $100 million. 37 researchers and engineers from OpenAI and Google, including Google Chief Scientist Jeff Dean, filed an amicus brief in support of Anthropic. The company is fighting the designation in court.
The data center story and the weapons story are not unrelated. Anthropic drew a line on principle and is now operating inside an infrastructure ecosystem that runs on the opposite of principle — gas pipelines, Google's balance sheet, bank financing at terms enabled by hyperscaler credit. The company that said no to autonomous weapons is also a tenant on a Texas facility where the power comes from natural gas turbines and the financing is arranged by a consortium that needs Google's involvement to get a better rate.
The DoD has reportedly been extensively using Claude for military operations, including in target selection and analysis of missile strikes in its war against Iran — reporting that originated with the Washington Post. The company that said AI is not reliable enough for autonomous weapons has its AI in the targeting loop.
Google has invested approximately $3 billion in Anthropic across two rounds — $2 billion in 2023 and $1 billion in 2025. Court documents from last year revealed Google held a 14 percent stake. The two companies also signed a cloud capacity agreement giving Anthropic access to more than 1 gigawatt of capacity and up to one million TPUs. This is not a standard customer relationship. Google is both an investor and a financing partner for the physical infrastructure that runs Anthropic's most demanding workloads.
The 7.7-gigawatt buildout figure is worth sitting with. A large coal plant typically generates between 500 megawatts and 1 gigawatt. At full expansion, this single facility would consume power equivalent to roughly seven to fifteen large power plants. The gas pipeline feeding it is load-bearing infrastructure for a facility whose power demands will grow as Anthropic trains larger models.
According to the National Center for Energy Analytics, natural gas turbines and engines are the dominant option for AI data centers in the two-to-five-year timeframe precisely because the technical capacity exists to deploy them quickly — while grid-scale solar and battery storage at the scale AI labs need remains a five-to-ten-year horizon. SemiAnalysis puts a sharper number on the cost of waiting: the firm estimates an AI cloud can generate $10 billion to $12 billion per gigawatt annually, and getting a 400-megawatt facility online six months early is worth billions in revenue. Large AI operators, the analysis goes, treat gas turbines not as the cheapest long-run option but as the fastest path to revenue — and at this scale, time-to-revenue is measured in billions of dollars per month.
