Stratechery — Data Center Discontent, Understanding the Opposition, Fixing the Problem (2026-05-18)
Why this is in the vault
Thompson's central argument: data center opposition has gone mainstream and bipartisan in a way nuclear never reached. Three data points he leads with:
- Hill County, TX (Dallas SW): commissioners voted 3-2 for a yearlong moratorium on data center + power plant construction in unincorporated areas, citing up to 8 planned data centers (many with own power plants).
- Gallup poll (March 2026): 71% of Americans oppose a local data center (48% strongly), vs. 53% opposed to a local nuclear plant. Bipartisan: Dems 56% strongly opposed, Reps 39% strongly opposed, indies 48%. Data center opposition now exceeds the historical high for nuclear opposition (63%, set since 2001).
- DeForest, WI (his hometown reference): village rejected a planned 1.6 GW data center.
His diagnosis of why opposition is uniquely strong (vs. the pre-AI data center era when local governments competed for them):
- AI's job-loss framing kills the message. Best-case tech narrative = "takes everyone's jobs," worst-case = "ends humanity." Both make terrible local pitches.
- Globalization-payback dynamic: rust-belt workers had no say when factories moved to China. Data centers require permission. People who never got a vote on globalization now get a vote on AI and are using it.
- Tech has no partisan constituency: left opposes Big Tech reflexively, and the Trump base (despite some tech-aligned leadership) also hates Big Tech. Data center fights are decided by bases, not elites.
His proposed solution: direct cash payments to residents. Not amorphous tax benefits routed through (low-trust) local governments. Not "workforce training." A check in the mailbox every year. His worked example for the rejected DeForest 1.6 GW facility:
- 1.6 GW → ~$3B annual operator revenue (his estimate)
- DeForest pop. ~11,500
- Pay every resident $10,000/year → 3.8% of annual revenue
- He bets that would have been approved; operators could pass cost to data-center customers
He frames this explicitly as AI-derived UBI delivered through the most American kludge possible — analogous to EMTALA being de-facto universal healthcare and SSDI/Section 8/SNAP being de-facto welfare.
Sponsor / bias notes: Stratechery is subscription-funded, no third-party sponsor in this update. Bias disclosure embedded in piece — Thompson is openly pro-data-center build-out and self-identifies as an AI optimist; he's grinding an axe against the opposition, not steelmanning them deeply. His DeForest reference is personal (it's near his home). The piece is short on opposition-side primary voices.
Mapping against Ray Data Co
Strong relevance to the active power-cycle v1.1 paper trade. The thesis (memory-cycle-shaped power-cycle: GEV/TLN/VST/CEG/CCJ + ERCOT interconnection-queue anchor) has been built almost entirely on the supply side: hyperscaler capex commitments translating into long-dated power demand, gas turbine OEM bottleneck, nuclear restart optionality, uranium fuel cycle. Thompson's piece is a demand-side risk update on the same trade.
The honest read: the v1.1 thesis prices in demand materializing on schedule. If local moratoria spread — Hill County TX is now the leading edge; Gallup at 71% opposition suggests it scales — the interconnection-queue conversion rate falls, projects slip from active to stalled, and the gap between announced capex and operating MW widens. That gap is exactly the thing the ERCOT anchor was supposed to surveil.
Three concrete watch-items this surfaces:
- Track moratorium count as a leading indicator, similar to how the smart-money watch tracks 13F drift. Hill County TX is now N=1 publicly. North Carolina pulling back sales-tax exemptions ($57M/yr per Stein) is a related signal. A moratorium dashboard belongs in
01-projects/investing/anchors/power-cycle-v1alongside the queue tracker. - Differentiate the long names by siting risk. TLN (Susquehanna) and CEG (nuclear fleet) sit on already-permitted, already-loved nuclear sites — the NIMBY risk is much lower than for greenfield gas+DC behind-the-meter campuses. GEV (gas turbine OEM) is somewhat agnostic to which DC gets built but does need some to get built. VST (gas+nuclear mix, ERCOT-heavy) sits exactly in the Hill County TX firing line. The "DC-discontent factor" splits the basket.
- Thompson's UBI-via-DC proposal is itself an investable hypothesis. If the political resolution genuinely is direct payments to residents, capex per MW goes up by ~3-5% (his DeForest math) — meaningful but not thesis-killing. The bigger consequence: operators with the balance sheet to write the checks (Meta, MSFT, GOOG, AMZN) widen their lead over the long tail. That reinforces hyperscaler concentration, which is bullish for the picks-and-shovels names (GEV especially) and neutral-to-positive for the merchant-power operators (TLN, CEG, VST) selling into that demand.
Adjacent: MAC product positioning doesn't move on this. The Sanity Check newsletter could potentially riff on the UBI-via-kludge framing — but per the no-derivative rule, only if there's an original re-frame, not a restatement. Possible RDCO original angle: "data-center-as-UBI is the same operational pattern as our COO-agent-as-distribution-of-knowledge-work — the kludge that delivers a thing we can't admit we want." That's a Sanity Check candidate but needs more cooking; not filing it as a pitch yet.
Related
- [[2026-05-18-power-cycle-v1.1]]
- [[2026-05-17-power-cycle-v1]]
- [[2026-01-08-stratechery-interview-power-for-ai]]
- [[2026-01-09-stratechery-weekly-ai-power]]
- [[2026-02-18-not-boring-power-age-of-intelligence]]
- [[2026-01-09-not-boring-a16z-power-brokers]]