Data Centres & the Grid
The compute build-out is a physical-infrastructure story pretending to be a software story. The projection below concerns electricity, and the tail scenarios beside it concern what happens when a grid under that much load is attacked or hit.
Live projections
US data centers consumed 4.4% of national electricity in 2023 and are projected at up to 12% by 2028 (LBNL); the 2024 PJM capacity auction cleared ~10x its prior price before a single gigawatt of that demand peaked. Either generation build-out accelerates to wartime tempo, or electricity prices become the next inflation politics — most likely both.
Wrong if: through 2028, average US retail electricity prices rise no faster than CPI, no major grid operator declares data-center-driven resource inadequacy, and interconnection queues shorten.
Window: 2026-2029.
Tail scenarios
State actors have already burrowed into Western critical infrastructure — CISA’s 2023–2024 Volt Typhoon advisories describe implants pre-positioned for disruption, not espionage. Ukraine’s grid was switched off remotely twice (BlackEnergy, 2015; Industroyer, 2016). Modern grids expose OT/SCADA control layers to networked attack surfaces, so a coordinated strike could black out regions simultaneously and hold them down.
A coronal mass ejection on the scale of the 1859 Carrington Event induces geomagnetic currents that overload extra-high-voltage transformers — bespoke units with 12–24 month lead times and almost no spares. In July 2012 a Carrington-class CME crossed Earth’s orbit just nine days ahead of the planet (Baker et al., 2013). The US National Academy of Sciences estimated a severe event could cause $1–2 trillion in first-year damage, with a multi-year recovery.
More than 70% of trades are already algorithmic, and AI increasingly governs power dispatch, logistics, air traffic and defensive triage. Each system is tested alone; the interactions between them are not. A shared dependency, an adversarial input, or an emergent feedback loop can propagate across domains faster than human operators can intervene. The 2010 Flash Crash — a trillion dollars gone in six minutes (SEC/CFTC, 2010) — was the small preview.
Other subjects on the record
Every claim above is also listed in the full projection registry and the calibration ledger. Scoring is explained in how a prediction is made.