Sunday, November 22, 2026Vol. III · No. 326Subscribe
The Mining, Energy & Technology Wire

The Grid Is AI's Real Bottleneck

Power access, not chip supply, is deciding where AI capacity gets built and who pays for it. The IEA, a fight over PJM's capacity auction, Meta's emissions and a Schneider warning all point the same way.

The Grid Is AI's Real Bottleneck
PhotographPower access, not chip supply, is deciding where AI capacity gets built and who pays for it. The IEA, a fight over PJM's capacity auction, Meta's emissions and a Schneider warning all point the same way.Photo: Janek Valdsalu / Unsplash

The market sold the metals of the grid hard in its latest session. COPX, the copper-miners ETF, declined 4.98% in its latest session and closed at $56.33 on Friday. URA, the uranium ETF, declined 6.56%. The broader market fell sharply with them, so one day proves little about copper or uranium specifically. But the longer run of closes is blunter. COPX closed February at $95.70 and stands at $56.33 now. URA has gone from $55.12 in October 2025 to $26.79.

That sits awkwardly beside what the International Energy Agency published this month. Its World Energy Outlook 2026 says the next ten years need at least 25 million kilometres of new power lines, and that 15 million kilometres of existing lines need replacement or upgrades. It adds that the copper, transformers and cables involved are, for now, in tight supply.

The queue for electrons

That is the story under the AI boom. Chips can be ordered. Electricity has to be connected, and the connection is where projects stall. Investors will pour some $4 trillion into data centers globally by 2030, after $565 billion in 2025, according to the IEA, Energyconnects reported. The expected growth is concentrated in facilities built to train and run AI, rather than in cloud storage or video streaming.

The scale is easier to grasp in power terms. Bloomberg reported that fewer than 100 major AI projects would carry combined power needs of 50 gigawatts, roughly the peak electricity demand of Great Britain. The IEA's satellite monitoring of 93 major AI data centers under construction has shown capacity rising steeply in just two years. Planned developments in some big growth markets would exceed current peak demand if all of them were built.

The agency is not only pessimistic. It says AI could help optimise the electricity system and cut costs. For now, though, it says surging data-center demand is amplifying strains in some markets and supply chains. Energyconnects reported that the IEA found data-center carbon emissions are set to rise 70% by 2030, driven by power from gas- and coal-fired plants.

Who pays

The fight over cost is playing out in PJM, the grid operator serving 67 million consumers. Representatives from seven states, including Pennsylvania, Ohio and New Jersey, wrote to PJM's board asking it to postpone the capacity auction for the 2029/30 delivery year, scheduled for Dec. 9 to Dec. 15, Utility Dive reported. They want the delay "until data center loads, both real and speculative, can be removed from the load forecast."

Their evidence is a Nov. 12 report from the market monitor, Monitoring Analytics. It warns that PJM's forecast includes data centers that have been canceled or are unlikely to reach projected demand, which could cost consumers an additional $5 billion to $10 billion. Including existing and forecast data-center demand has already raised capacity costs by $29.4 billion for all customers, the monitor says.

PJM is resisting. Spokesman Jeffrey Shields told Utility Dive: "Postponing the auction this close to its execution could destabilize the market and increase the risk of power interruptions." He also said demand is growing faster than new supply can be built. Any delay would need approval from the Federal Energy Regulatory Commission. PJM has proposed its own two-part plan: bilateral contracts outside the auction, plus a backstop procurement for large loads that lack such contracts. The advocates asked for a written reply from the board by Nov. 23.

Utility Dive also reported that PJM's current rules effectively stop data-center companies from paying for their own power supply and infrastructure. That is the grievance in a sentence. Speculative load enters the forecast, and everyone's bill carries it.

The builders feel it too

Meta's sustainability report shows the strain from the other side. Ars Technica reported that its executives are rethinking the goal of net-zero emissions by 2030, because, in their words, "demand for AI infrastructure has grown faster than almost anyone projected." Ars Technica said Meta posted a record percentage increase in greenhouse gas emissions among Big Tech rivals building AI data centers.

Trellis put Meta's net footprint at 11.9 million metric tons of CO2 equivalent, up 45 percent last year. The location-based figure was 21.7 million mtCO2e, well above the net number. Electricity-related emissions grew almost 38 percent between 2024 and 2025, alongside a 35 percent rise in power use. Trellis reported that Microsoft's footprint jumped 25 percent, Amazon's rose 16 percent and Google's rose 18 percent. Meta alone among the four has said it will reconsider its timelines.

Meta says it matched 100% of its electricity use with clean energy for the sixth straight year. Its executives Rachel Peterson and Blair Swedeen blame long interconnection and permitting queues, and slow supply chains for low-carbon materials, for stalled clean-energy deployments, Trellis reported.

Europe's warning

The pattern is not only American. Schneider Electric's chief digital officer, Peter Weckesser, told Reuters on Nov 19 that high energy costs, thin grid investment and fragmented regulation risk slowing Europe's AI build-out and pushing capital elsewhere. He said the United States is clearly ahead, followed by China, while Europe lags and regions including the Middle East are expanding faster. Energy costs feed directly into the cost of computing, he said, so higher European power prices could redirect investment. Europe needs major grid investment and a more consistent regulatory framework, he added.

Read together, these accounts describe the same constraint from four angles. The IEA counts the wires, PJM's states count the bills, Meta counts the carbon and Schneider counts the capital that may leave. The latest sell-off in the miners says little about any of them. The lines, though, still have to be built from copper that, by the IEA's account, is scarce, and whoever gets connected first will decide where the models run.

Original reporting and analysis by the Stake & Paper editorial team. See linked sources within the article.

Share this story

More from Stake & Paper

Was this article helpful?

ClaimWatch

Mining claims intelligence — from query to report, in minutes.

Every unpatented mining claim across all twelve BLM states. Leadfile audits, due diligence, site selection, regional prospecting, entity investigations, and AOI monitoring — delivered as complete report packages.

4.4M+
Claims Tracked
12
BLM States
7
Report Types
Request a Sample Report
Stake & Paper AM

One morning brief. The whole energy sector.

Original analysis, the day's most important wire stories, and market data — delivered before your first cup of coffee. Free.