US electricity prices rose 0.9% in October, 2.25 times the 0.4% rise in all consumer prices, the Bureau of Labor Statistics reported on 10 November. Two days later, PJM’s independent market monitor said putting data centers in its capacity market has caused “an irreversible increase of $29.4 billion in capacity market costs for all customers to date, with more costs to come”.
The same week, as we read it, AI builders moved to buy or fund power themselves. Anthropic’s $3.4 billion deal for 1.7 gigawatts of gas-fired generators, as the Financial Times reported it, works out to about $2.0 million per megawatt, by our arithmetic. Microsoft said “We will pay our own way”, and SpaceX applied for a tax break on a $24 billion gas plant and compute site.
Business Pill · Grid Cost Allocation
An evergreen Business Engineer explainer on who pays when the grid is extended for large loads such as data centers. It teaches the general idea only and does not discuss this week’s figures.
The key insight: As we read the week, electricity got a price the public can see, in the CPI and in PJM’s capacity bill, while the largest builders moved to buy or fund their own supply, Anthropic at about $2.0 million per megawatt, by our arithmetic.
Compounding the monthly changes from April through October, electricity is up 2.7% against 2.2% for all items, by our arithmetic from the BLS table, as our CPI piece showed. Prices excluding food and energy rose 0.3%.
The market monitor’s report gives the wholesale bill. The total cost of wholesale power in PJM rose $32.42 to $111.61 per MWh in the first nine months of 2026, up 40.9%. Without data center load in the capacity market, it says, the figure would have been $101.66. The $9.95 gap is 8.9% of the total cost and about 31% of the year’s rise, by our arithmetic (our report on the finding).
In PJM’s 2028/2029 capacity auction, the monitor calculated that forecast data center load added $5.73 billion to total revenues of $16.41 billion, or 34.9% of the total, by our arithmetic (our auction piece). The EIA’s November outlook has the commercial sector, which it says includes data centers, taking 64% of US power-sales growth from 2025 to 2027, by our arithmetic from its table (our EIA piece).
Anthropic’s $3.4 billion deal for 1.7 GW of gas-fired generators, as reported by the Financial Times, per megawatt, against the cost ranges Reuters reported for new, used and refurbished gas turbines (Aurora Energy Research; Origin Partners). The per-megawatt figure is our arithmetic; generators and turbines are different machines.
The Builders Bought Their Own
Anthropic agreed to pay $3.4 billion for gas-fired generators for its data centers, the FT reported, taking over 1.7 gigawatts of orders that OnePWR Solutions had placed with Caterpillar (our report). The $3.4 billion is about 99% of Caterpillar’s $3.421 billion of third-quarter Power Generation sales, by our arithmetic (our Caterpillar comparison).
For scale, new gas turbines cost about $2 million to $2.5 million per megawatt, according to Aurora Energy Research, and refurbished units about $1.3 million, according to Origin Partners, Reuters reported. A new turbine takes six years on average to deliver, a Wood Mackenzie analyst told Reuters (our turbine piece). Anthropic’s roughly $2.0 million per megawatt sits at the low end of the new-turbine range, by our arithmetic.
Noelle Walsh, Microsoft’s president of cloud operations and innovation, told a BloombergNEF conference in Houston “We will pay our own way”, Bloomberg reported on 11 November. Microsoft says it has contracted to add 7.9 GW of new generation in the MISO market, “more than double our current consumption”, and is paying for the on-site plant at a Pecos, Texas, campus of about 2 GW (our Microsoft piece).
SpaceX applied to the City of Starbase for a tax abatement on what the city’s notice calls a “gas-fired power plant and super-compute facility”, with initial phases put at about $24 billion; that figure covers both the power and the compute (our Starbase piece). SoftBank’s $3.7 billion loss on power-generation equipment put a price on turbine reservations, against GE Vernova’s 58 GW of slot reservations (our SoftBank piece).
Who Pays Became the Vote
The Senate voted 73-25 on 9 November to open debate on the Bipartisan American Affordability and Jobs Act, the permitting bill that would make large data centers pay more of the grid costs they cause (our vote count). By our count, local fights touched $17 billion of US data center plans in the first 11 days of November, including a $10 billion plan in Delhi Township, Michigan, that was withdrawn on 9 November (our tally).
The Structural Read
As we read the monitor’s figures, data center load is no longer a forecast in PJM’s bills: $9.95 of the $111.61 per MWh this year is its estimate of the data center effect, 8.9% of the total, by our arithmetic.
As we read the builders’ moves, buying generators, contracting generation or seeking a local tax deal for a private plant all answer the same question the Senate bill asks: who pays for the power a data center uses.
Monitoring Analytics, Independent Market Monitor for PJM, 2026 Q3 State of the Market Report, 12 November 2026
“With the completion of the 2028/2029 BRA, the inclusion of data center load in the capacity market auctions has resulted in an irreversible total increase in capacity costs of $29.4 billion.”
Three Implications
FOR UTILITIES AND REGULATORS PJM’s monitor puts the data center effect at $9.95 of $111.61 per MWh in the first nine months of 2026, 8.9%, by our arithmetic.
FOR AI BUILDERS Anthropic’s $3.4 billion for 1.7 GW, as the FT reported, is about $2.0 million per megawatt, the low end of the new-turbine range Reuters reported.
FOR HOUSEHOLDS Electricity rose 0.9% in October against 0.4% for all items, BLS data show.
The framework puts it in one line: “The structural bottleneck isn’t computing, it’s electrons.”
As we read it, this week the electrons got a public price: the CPI and the PJM monitor’s $29.4 billion. The framework does not discuss Anthropic’s deal, the October CPI or the monitor’s report.
PJM’s total cost of wholesale power per MWh, first nine months of 2025 and 2026, and the market monitor’s 2026 figure without data center load. Source: Monitoring Analytics, 2026 Q3 State of the Market Report, 12 November 2026. The shares are our arithmetic.
What Is Not Established
The $2.0 million per megawatt is a gross ratio of the FT’s reported price to the reported capacity; what the price covers, such as installation or fuel, is not given in the reporting we read, and generators and turbines are different machines. The monitor’s $29.4 billion and $101.66 are its own calculations of what costs would have been without data center load.
We read the BLS release, the monitor’s report and our earlier pieces built on each company’s and agency’s own documents; none of the companies was contacted for this piece.
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.
Electricity rose 0.9% in October, 2.25 times the CPI, and PJM’s monitor put the increase in capacity costs from data center load at $29.4 billion to date, and at $9.95 per MWh this year, 8.9% of the wholesale cost, by our arithmetic. In the same week Anthropic’s generator deal came to about $2.0 million per megawatt, Microsoft said it would pay its own way, and SpaceX sought a tax break for a $24 billion gas and compute site.
A note on sourcing. This analysis combines twelve of our pieces from 8 to 13 November, each built on its primary source, with the BLS October CPI release and the PJM market monitor’s third-quarter report, both re-read on 14 November. The Anthropic deal terms are the Financial Times’s, as reported. Ratios and shares are our arithmetic.
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.
