Anthropic bought commitments for 1.7GW of gas-fired generators for $3.4bn, the Financial Times reported on November 8. That is about 4 times the 420MW of wind farms that Envision’s Galaxy AI campus in Inner Mongolia has built for its own load, according to a SemiAnalysis site report published a day earlier, by our arithmetic.
Read side by side, the reports show two ways of securing power for AI campuses as equipment: gas generators and turbines in the US cases, and wind farms sized above the load in the Chinese one. Our recent coverage also records pushback at national, state and city level.
Business Pill · Grid Cost Allocation
An evergreen Business Engineer explainer on grid cost allocation: who pays for new power lines. It teaches the general idea only and does not discuss the companies in this piece.
The key insight: The FT reported that Anthropic bought commitments for 1.7GW of gas-fired generators for $3.4bn, about 4 times the 420MW of wind farms Envision built for its Galaxy AI campus, per SemiAnalysis, by our arithmetic. SpaceX’s Starbase notice adds a gas-fired plant with no stated size, while a 73GW UK grid queue, a Bloomington ordinance and a Texas permit halt dated 21 September record the pushback side.
The FT reported that Anthropic bought the 1.7GW of heavy-duty gas-fired generator commitments in late September from OnePWR Solutions, taking over orders OnePWR had placed with Caterpillar two years ago, which we covered in Anthropic Buys 1.7GW of Pre-Ordered Gas Generators for $3.4B. By our arithmetic that is about $2.0 million per megawatt. The FT does not say where the generators will be installed.
On the supplier side, Caterpillar reported $3.421 billion of Power Generation sales in the third quarter of 2026, up 30% and mainly in data center applications according to its 10-Q, as we reported in Anthropic’s $3.4B Deal Is 99% of Cat’s Q3 Power-Gen Sales. Its firm backlog was approximately $78.1 billion at September 30.
SpaceX is on the same path at Starbase, Texas. A city notice says SpaceX “proposes to construct, own, and operate a gas-fired power plant and super-compute facility” on about 444 acres, with about $24 billion of investment in initial phases, as we reported in Starbase Notice Puts $24B on SpaceX Gas Plant, Supercomputer. A SpaceX job posting describes aeroderivative gas turbines. Neither document gives a size for the plant in megawatts.
The Chinese route looks different. SemiAnalysis says Galaxy’s two wind farms have 420MW of installed capacity, about 1.75 times the 240MW load of the campus’s first two phases, and that Envision aims to supply over 60% of that power use with its own wind, as we reported in China AI Campus Builds 1.75x Wind Power for Its Own Load. The report says Envision earlier built a 2GW off-grid wind, solar and storage system in Chifeng that feeds a green ammonia plant and a small pilot datacenter.
SemiAnalysis draws the contrast itself. In its words, “Abilene draws on the grid and on gas turbines on site, while Galaxy relies largely on wind”, and both campuses “were planned around their own power”.
Generating capacity bought or built for AI loads, as reported 6-8 November 2026, nameplate MW: Anthropic 1,700MW of gas-fired generators (Financial Times, as reported); Envision Galaxy 420MW of its own wind for a 240MW load (SemiAnalysis); SpaceX Starbase gas-fired plant, size not stated ($24B initial phases, city notice); Envision Chifeng 2,000MW off-grid system shown as a precedent, mainly a green ammonia plant. Our arithmetic: 1,700 / 420 = about 4x. Nameplate capacity, not energy delivered.
Our Number: 4x, in Nameplate Megawatts
By our arithmetic, Anthropic’s 1,700MW of gas-fired generators is about 4 times Galaxy’s 420MW of wind, and together the two come to 2,120MW of generating capacity reported for AI loads in two days. The Starbase notice gives a dollar figure, not megawatts, so it is not in the sum.
These are nameplate figures, not energy delivered. SemiAnalysis puts a good Ulanqab wind site at about 3,500 full-load hours a year. The sources give no running hours for Anthropic’s generators, so we do not compare energy output.
The Pushback Layer
Recent reporting also shows resistance. The Guardian reported 315 UK data centres queueing to connect to the grid, 73GW of demand against a 45GW national peak, which by our arithmetic is 162% of peak, and Sky reported a coalition of 75 organisations planning to march in Edinburgh on November 14 against AI data centres in Scotland, as we covered in UK Data Centre Queue 73GW vs 45GW Peak; 75 Groups to March.
In Texas, the governor’s office said on 21 September that Abbott directed the state’s environmental regulator to halt all permits sought by data centers until ERCOT completes its audit, which we covered in Developers Eye Utah, Iowa as Texas Halts Data Center Permits. In Indiana, Bloomington’s Common Council voted on November 4 to ban the distribution and deployment of superintelligent AI in the city, a measure aimed at the AI itself rather than at data centres, as we reported in Bloomington Votes to Ban Superintelligence, Sets 95% Bar.
Connecting the Coverage
This piece connects nine of our reports. The Anthropic deal and its Caterpillar supply side sit in the two pieces above. The Chinese side starts with China AI Campus Built 120 MW in a Year at a Fifth of US Cost, where SemiAnalysis says Galaxy’s first 120MW phase was delivered within a year at a fifth of US costs, and continues in China Data Center Rents Halved While Build Costs Fell 25%, where SemiAnalysis puts rents at about $40/kW/month and capex at under $3 million per megawatt. What’s new here is the side-by-side count of generating capacity across the US and Chinese cases.
The Business Lever: Power as a Procurement Line
As we read the reports, the scarce input in each US case is equipment with a delivery date. The FT writes that the engines were ordered “years earlier before prices surged”, and Caterpillar’s filing shows Power Generation sales up 30%. In the Chinese case, as SemiAnalysis describes it, the company that makes the turbines, batteries and power electronics also builds the campus.
The Structural Read
As we read the reports, the US cases buy gas generators and turbines, while the Chinese case sizes wind farms above the load and links them through a dedicated line.
As we read the pushback in our coverage, it comes at three levels: a national grid queue in the UK, a state permit halt in Texas dated 21 September and a city ordinance in Bloomington.
As we read the figures, nameplate megawatts measure what was bought or built, not what will be delivered, which is why the piece does not compare energy output.
SemiAnalysis, We Visited China’s Stargate in Inner Mongolia
“Abilene draws on the grid and on gas turbines on site, while Galaxy relies largely on wind”
Three Implications
FOR AI LABS The FT reported that Anthropic took over OnePWR’s Caterpillar generator orders, engines ordered years earlier before prices surged, per the FT.
FOR EQUIPMENT MAKERS Caterpillar reported Power Generation sales of $3.421 billion in the third quarter, up 30%, and firm backlog of about $78.1 billion, per its 10-Q.
FOR THE NUMBERS TO WATCH The Starbase city commission will consider a tax abatement for the SpaceX project on December 9, and the UK march is set for November 14, per the city notice and Sky.
The essay says “The structural bottleneck isn’t computing, it’s electrons.”
As we read it, the week’s gas generators, gas turbines and wind farms are three ways of buying those electrons as equipment. The essay does not discuss Anthropic, SpaceX or Envision.
What Is Not Established
The FT does not say where Anthropic’s generators will be installed or whether they will run off the grid. The Starbase notice gives no plant size in megawatts, and the city has not approved the abatement. The Chifeng system mainly feeds a green ammonia plant and is shown only as Envision’s earlier project. Nameplate megawatts are not energy delivered, and the 4x comparison is our arithmetic.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Anthropic’s 1.7GW of gas-fired generators, reported by the FT, is about 4 times the 420MW of wind Envision built for its Galaxy AI campus, by our arithmetic, and SpaceX’s Starbase notice adds a gas-fired plant with no stated size. Separately, a 73GW UK grid queue reported on November 5, a Bloomington ordinance passed on November 4 and a Texas permit halt dated 21 September record the pushback side.
A note on sourcing. This analysis is built only from figures in nine earlier FourWeekMBA pieces, each read on 8 November 2026 and each attributing its own sources: the Financial Times on Anthropic, Caterpillar’s third-quarter 10-Q, the City of Starbase notice and a SpaceX job posting, the free section of SemiAnalysis’s report on Inner Mongolia, The Guardian and Sky News on the UK, the Bloomington ordinance, and the Texas governor’s office. The 4x and 2,120MW figures are our arithmetic on nameplate capacity.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
