Hubert Sim

ComEd's own filing puts a price on it: connecting 30 data centers will cost $3.1B. $1.6B in transmission, $1.5B in distribution. Developers have paid $721M in deposits so far. Who covers the rest is now a fight before Illinois regulators.

Newly announced data center projects have run at $400B+ a month globally since April. The US has 53 GW of data center load actually running. Those two numbers sit on very different timelines, and most of the risk in this cycle lives in the gap between them. The figures come from Industrial Info Resources (IIR), reported by Data Center Knowledge this week. The money side. Microsoft, Meta, Amazon and Google spent a combined $64B on data centers in 2018. Their announced spending is expected to pass $700B this year. Cumulative announced data center capex has passed $7 trillion globally. The US and Canada account for about $4.75 trillion of it, across more than 4,250 planned or in-progress projects. More than 400 GW of new load is in planning, and almost 100 US sites are being designed for at least 1 GW. The physical side. US operational data center load is 53 GW, up from 23 GW in 2023. Another 42 GW is approved or under construction. Together that's roughly 95 GW that is real or close to real. Against 400+ GW in planning. For every GW running or being built, more than four are still on paper. I've spent most of my career in capital project governance, and the first thing that discipline teaches you is that an announcement is not a project. A project becomes real when it clears five gates: site control, permits, an interconnection agreement, long-lead equipment with a delivery date someone will sign for, and committed financing. Most of that 400 GW hasn't cleared all five. A lot of it hasn't cleared one. The gate that has moved the most is equipment. GE Vernova, Mitsubishi Power and Siemens Energy are sold out of larger gas turbines through 2029, with some orders stretching into 2031. All three are expanding. Fuel cells, reciprocating engines and aeroderivative units are being offered as faster alternatives. Meanwhile IIR tracks $228B of new gas power plant projects in play, all drawing on the same factories. So the critical path for a 1 GW campus increasingly runs through a turbine factory's order book. In schedule terms, a turbine reservation for 2028 may be worth more than the land it will sit on. If you're evaluating a project, ask to see the slot: who holds it, what deposit secured it, and what happens to it if the project slips a year. IIR's Shane Mullins put the developer logic plainly: past 300 MW, with a utility queue that says 2034, your own gas plant is "the fastest and most certain path to power." Fast and certain, maybe. But the bottleneck just moves. The queue at the utility becomes a queue at the OEM, and on-site generation brings its own permit gate. New Jersey recently fined one data center $1M over 123 MW of unpermitted generation. Then there is the forecast problem. In 2022, US grid planners expected 24 GW of additional demand over a decade. By 2025 that number was 166 GW, nearly seven times higher. IIR says data center demand could go from 35 GW in 2024 to 125 GW by 2030. Utilities build transmission and generation against forecasts like these, and ratepayers usually fund it. When a forecast moves 7x in three years, the honest question is how much of it is double counting. One campus can sit in three utility queues while the developer shops for the best deal. Texas has paused new grid interconnections while it works through about 200 GW of requests. And opposition is no longer a side story. IIR counts more than 500 data center moratoria across 42 states, led by New Jersey with 41 and Michigan with 36. New York has temporarily banned new data centers over 50 MW. About $130B of projects were rejected in Q1 2026 alone, more than in all of 2025, and the number of opposition groups has more than doubled in a year. None of this has slowed the money yet. May alone saw more than $650B in new announcements, and IIR reads twelve straight months of elevated spending as a multiyear buildout. I don't doubt the demand. What I'd watch is conversion: how much of the announced pipeline turns into energized megawatts, on what schedule, and who carries the cost of the part that doesn't. Three things I'd want on one page for any large AI campus right now: 1. Interconnection status in writing, with the utility's date, and whether service is firm or conditional. 2. The long-lead equipment schedule (turbines, transformers, switchgear) with contractual delivery dates, and penalties if they slip. 3. A clear answer on who pays for grid upgrades if the load never shows up. If a project can't produce that page, I'd treat it as an announcement, not yet a project. Question for people inside these deals: of the 400+ GW in planning, what share do you think actually energizes by 2030? And which gate do you trust most when you make that call?

Developers have announced 110+ GW of behind-the-meter gas at data centers, about 20% of all US gas capacity installed today. Why? Grid hookups take years. On-site gas takes less. RMI just put numbers on what that speed costs. 🧵

What I find interesting: BTM gas and non-firm service aren't either/or. On-site generation sized to cover curtailment hours, not full load, could be what makes non-firm service bankable. Smaller plant, fewer run hours, earlier power.

Jacobs surveyed 600 senior energy leaders in 5 countries. Only 20% in the U.S. say they consistently deliver major projects on time and on budget (30% globally). Capital isn't the bottleneck: 71% say it's available. Only 29% say ready-to-invest projects are too.