Moratoria delay only 2.3GW, but miss withdrawals: projects are moving, not stopping — resolves 2027-06-30

The debate: the research firm SemiAnalysis converted about 300 local data center moratoria and a New York State executive order into delayed capacity case by case and put the total at only about 2.3GW, which some read as proof that the bans barely matter.

Why it matters: whether a ban delays a project, kills it or pushes it elsewhere has very different effects on AI compute delivery, on siting and on local communities, so investment and siting models need the right one.

Who is on each side: one side cites SemiAnalysis's delay figure to say the effect is small, the other cites Heatmap's withdrawal counts to say it is large, and SecondSource thinks both can be true because they measure different things.

SecondSource’s reading: SemiAnalysis's 2.3GW delay figure is the right reading for 2026–27 deliveries, and it is low because of how it is built. Of the 38GW expected for 2027, 22GW is already under construction, and SemiAnalysis itself says the bans cannot reach all of 2026 and most of 2027.

The real blind spot is not the years from 2028 on but exits: its rules count a withdrawn project as zero and treat one that moves elsewhere as unaffected.

A separate tracker, Heatmap, counts at least 3.5GW of power-demand proposals withdrawn in the first quarter of 2026, with about half of challenged projects withdrawn, speculative ones included, so that is an upper bound on losses.

Put together, today's signal points to projects moving, not stopping.

The original material behind this call has not yet been linked here in a form readers can open and check.

Observation window to 2027-06-30 (set by SecondSource): if no tracker publishes withdrawal rates that are similar inside and outside moratorium areas, and SemiAnalysis's forecast of new U.S. capacity for 2028 has not been cut by more than a tenth because of moratoria or local opposition, the reading holds. Actual 2027 deliveries against 38GW are checked separately on 2028-03-31 (also set by SecondSource).

What would overturn it

(1) A tracker publishes similar withdrawal rates inside and outside moratorium areas, meaning the exit blind spot does not matter for the national total. (2) SemiAnalysis cuts its 2028 forecast of new U.S. capacity by more than 10% within the window and attributes the cut to moratoria or local opposition, so the 'signal points to moving' call is withdrawn. (3) Actual 2027 deliveries come in clearly below 38GW, with the shortfall attributable to the moratoria, so the part of this call about 2026–27 deliveries is wrong.