The biggest AI story this week was not a model release. It was electricity.
In the space of a few days, the Financial Times reported that Anthropic committed about $3.4 billion to secure 1.7 gigawatts of gas-fired generators from a Texas equipment supplier, bought in late September and bundled with the engineering capability to install them (CNBC carried the report; Reuters noted it could not independently verify it). OpenAI took its "Grid Protectors" pledge national, promising to let utilities curtail its future campuses during grid stress, building on a Georgia Power agreement that allows up to 1,000 megawatts of curtailment at a planned 3,200-megawatt campus in Effingham County. And TechRadar reported that xAI installed eight refurbished jet engines at its Mississippi site, each rated up to 50 megawatts, because connecting a facility to the grid the ordinary way can take years.
Three companies, three different answers to the same problem: the grid cannot deliver power at the speed AI wants to grow.
The gap between promise and physics
The same week showed what happens when energy claims meet reality. The Guardian reported that DataVita, a flagship of Britain's AI growth zone strategy, quietly revised its plan for a 500-megawatt development near Glasgow. The company had promised more than a gigawatt of renewable energy delivered by private wire. It now says the site will be "fully grid served," with renewables possibly connecting later, while retaining £202 million in public wealth-fund guarantees. Kat Jones, the director of the charity Action to Protect Rural Scotland, questioned the project's renewable-power claims and told the Guardian that hyperscale AI datacentres cannot run on renewable power alone and need a highly reliable source of grid power.
The political cost is arriving too. Al Jazeera reports that police arrested protesters at a Microsoft data centre construction site in the Netherlands on Friday, and that Texas voters elected their first Democratic senator in nearly 40 years in this week's midterms, in a state where data centre backlash was a live campaign issue. The underlying numbers explain the anger: the International Energy Agency estimates data centres consumed 485 terawatt-hours in 2024, about 1.5 percent of global electricity, roughly Germany's entire generation. The US hosts around 4,000 data centres drawing about 5 percent of national electricity.
There is a better playbook, and it already exists
The week's news was not all extraction. In Sonderborg, Denmark, the University of Southern Denmark's Bitten supercomputer recovers 100 percent of its waste heat through liquid cooling designed by Danfoss and pipes it into the municipal district heating network serving 75,000 residents. Meta's Odense data centre recovers around 100 gigawatt-hours a year, enough for about 6,900 homes. Stockholm has fed data centre heat into its network since 2014. Europe treats waste heat as an asset; the US, with nowhere for the heat to go, mostly vents it.
Even OpenAI's own announcement contains the honest sentence: "Flexibility cannot replace investment in new energy supply and transmission." Its cited $950 million in annual customer savings is Georgia Power's portfolio-wide estimate, not a saving attributable to the OpenAI contract, and the curtailment contract's operating terms remain undisclosed. xAI's turbine shortcut comes with its own ledger: Mississippi permits reportedly allow higher volatile organic compound limits, and the Southern Environmental Law Center is challenging the approval over pollutants including formaldehyde.
What this means for leaders
Here is my read, and it is opinion, not reporting: energy strategy is now AI strategy. If your organization's AI roadmap assumes compute will simply be available, you have outsourced your most important dependency to an interconnection queue, a permitting process, and a community vote.
Three questions worth asking in your next planning cycle. First, where does the power actually come from, and is that answer in a signed contract or a press release? DataVita is a reminder that those two documents can say different things. Second, who pays for the grid upgrades, and can your community verify it? As one former Department of Energy fellow told TechTarget, communities want to know who benefits, who pays, and who gets a say. Third, what is your plan for the heat? The Danish examples show recovered heat can warm thousands of homes. Venting it is a choice, and increasingly an unpopular one.
The companies that win the next phase of AI will not just have the best models. They will have the most credible answer to a simpler question: where does the electricity come from?
Sources
- Anthropic secures $3.4 billion generator deal for AI data centers: FT - CNBC
- OpenAI takes data center power-flexibility pledge national - TechTarget
- AI data centers repurpose retired jet engines for power - TechRadar
- Lanarkshire AI datacentre to fully rely on National Grid despite renewables claim - The Guardian
- As AI data centres guzzle energy, can we harness their wasted heat? - Al Jazeera
Full disclosure: Ganesh uses Instinct (instinct.ai), an AI assistant, to research, write and publish this blog series.
Top comments (0)