Energy

The central thesis of this KB: AI is bottlenecked by power. The AI buildout — training clusters, inference farms, hyperscaler campuses — requires reliable, always-on power at a scale that the existing grid was not designed to deliver. Renewables are cheap but intermittent; they cannot power a 500 MW data center at 3 AM. Natural gas can, but at a carbon and price-volatility cost that hyperscalers resist. Nuclear is the answer the market has converged on: dispatchable, 24/7, low-carbon, and immune to fuel-price swings. The gap between where nuclear supply is today (slow, regulated, expensive) and where AI demand needs it to be (fast, modular, co-located) defines the investment opportunity — and the constraint on the entire AI economy.

This KB covers the power/generation beat: conventional nuclear restarts, SMR development, fusion approaches, IPPs, grid infrastructure, uranium fuel cycle, and the hyperscaler power deals that are restructuring the utility sector. Battery/EV cost-curves and vehicle electrification belong to the separate electrification KB.

As of June 2026, three forces are simultaneously compressing:

1. Demand is accelerating faster than prior forecasts. Global datacenter electricity demand hit 132 GW in 2026 (up from 104 GW in 2025), with US datacenters consuming 41 GW. AI-optimized servers now account for 31% of datacenter power consumption. Goldman Sachs projects US datacenter demand doubles by 2027. The IEA projects 1,000+ TWh annual datacenter consumption by 2026 under high-growth scenarios — roughly equivalent to Japan's entire grid. (IEA Energy and AI report; Goldman Sachs AI Power report)

2. Nuclear is being industrialized by tech money. Every major hyperscaler has signed nuclear PPAs: Microsoft's Crane Clean Energy Center (TMI restart, 835 MW, 2027), Amazon's Susquehanna co-location deal ($650M campus), Meta's Clinton Power Station 20-year PPA, Google's Kairos Power + X-energy SMR contracts. Thirteen announced deals commit over 9.8 GW. GE Vernova's gas turbine backlog hit 100 GW (Q1 2026), growing $13B in a single quarter to $163B total. (SMRIntel; GE Vernova Q1 2026 earnings)

3. The supply side is structurally constrained. The US grid interconnection queue has swelled to a 2,600 GW backlog, with median wait times approaching five years. Uranium spot prices surpassed $100/lb in early 2026. Kazatomprom cut 2026 output 10%. SMR construction permits are expected from the NRC in 2026 (TerraPower Natrium, X-energy Xe-100), but first power is 2030 at the earliest. The binding constraint is not capital — it is time, permitting, and physical construction.

Concept Map

Concepts

Entities

Timeline

See timeline.md for chronological developments from 2023 through 2026.

Research Frontier

See frontier.md for active directions and knowledge gaps.

Sources

4 raw sources compiled (2026-06-24). Bodies in raw/.