The “climate guilt” narrative, which has long emphasized reducing energy use, curbing consumption, and prioritizing rapid decarbonization through intermittent renewables, is slamming into the physics of the AI era: massive, always-on computational demand that doesn’t tolerate blackouts, brownouts, or weather-dependent supply.

Global data center electricity use sat at ~415 TWh in 2024 (1.5% of world totals).

Projections have held firm or ticked higher: IEA’s Base Case still sees it doubling to ~945 TWh by 2030, with AI-accelerated servers (GPUs for training + inference) driving ~30% annual growth and nearly half the net increase.

Gartner pegs 448 TWh for 2025 rising to 980 TWh by 2030, while Goldman Sachs recently revised upward to 220% growth vs. 2023 levels—pushing global DC demand toward 1,350 TWh.

In the US (home to 45% of global DC load), consumption was 176 TWh in 2023 (4.4% of national electricity).

Forecasts now range 325–580 TWh by 2028 (6.7–12%) and potentially 606 TWh by 2030 (~12%), with data centers accounting for nearly half of total electricity demand growth through 2030.

Hotspots like Virginia could hit 41–59% of local supply. Utilities are warning of 49 GW shortfalls by 2028 in some models; hyperscalers are already delaying projects or self-generating with gas backups.

Inference (not just flashy training runs) now dominates long-term load—think everyday AI queries, agents, and enterprise deployments.

Efficiency gains in chips and cooling are real, but they haven’t offset the volume explosion.

Result: US power demand growth has flipped from flat to 1.7–2.5% annually, with gas filling near-term gaps per EIA.

The old framing—treat any new demand as sinful, push intermittents as the sole path, shame “wasteful” compute—hits hard limits when you need always-on, dense power at hyperscale.

Renewables have scaled impressively and will cover chunks of incremental load via PPAs, but:

Physics of reliability: AI clusters can’t flicker with the wind or sun. Overbuild + storage helps on paper, but transmission queues, land use, and backup needs (often gas) blunt the “pure green” story.

Growth mismatch: Total electricity demand is reaccelerating after decades of stagnation. AI is the biggest new driver, alongside EVs and industry. Sole reliance on variable sources risks blackouts, higher costs, or delayed builds—exactly what we’re seeing in 2025–26 grid congestion reports.

Big Tech’s own net-zero pledges are under pressure; some are quietly extending timelines or adding flexible “bridges.”

Emissions intensity per query keeps dropping, but absolute demand is the story.

This isn’t anti-climate—it’s recognizing that abundance (cheap, reliable watts) historically drives the innovation that cleans things up faster.

Hyperscalers aren’t waiting for narrative alignment—they’re voting with balance sheets and power purchase agreements.

As of April 2026:

Microsoft + Constellation: 20-year deal to restart Three Mile Island Unit 1 (now Crane Clean Energy Center, 835 MW). $1.6B project got a $1B Trump DOE loan in Nov 2025. Target online late 2027/2028, though transmission delays are the current bottleneck. Direct behind-the-meter power for AI data centers.

Meta: Massive Jan 2026 push—6.6 GW of nuclear projects via deals with TerraPower (up to 690 MW), Oklo (1.2 GW campus in Ohio), and others. Also extending Clinton plant life and eyeing restarts. Meta joined the “triple nuclear by 2050” pledge.

Amazon: Scaling SMRs hard—agreements with X-energy for multiple modules (aiming 5+ GW by 2039), plus Talen Energy PPA (1.9 GW existing nuclear) and Dominion. Behind-the-meter focus for campuses.

Google: Kairos Power SMR fleet—first online target 2030, more by 2035. World’s first corporate multi-SMR purchase agreement.

All four signed the World Nuclear Association pledge in 2025 to triple global capacity.

SMRs appeal for speed and modularity; restarts leverage existing sites.

Gas remains the flexible short-term fill, but nuclear is the firm, near-zero-carbon backbone AI craves. Geothermal and advanced options are in the mix too.

This isn’t greenwashing—it’s engineering reality.

AI itself will help: optimizing grids, accelerating fusion R&D, discovering better materials.

The economic upside (productivity, competitiveness) is too massive to subordinate to scarcity ideology.

Policymakers are noticing—permitting reforms, grid upgrades, and all-of-the-above supply are back on the table.

The tipping point favors outcomes over slogans: measure reliability, cost, emissions, and innovation, not guilt.

_____________________________________________________________________________________

We’re shifting from penance to power abundance, and that’s how civilization (and cleaner tech) actually advances.

As of mid-April 2026, nuclear momentum for powering the AI age has shifted from hype to concrete deals, funding, and policy tailwinds—driven squarely by hyperscalers’ need for firm, 24/7, high-density power.

Big Tech isn’t just talking; it’s underwriting restarts, funding SMR fleets, and signing multi-gigawatt PPAs.

This directly counters the intermittency limits of renewables in the face of surging data center demand.

_____________________________________________________________________________________

Small Modular Reactors (SMRs) are factory-fabricated nuclear power plants, typically under 300 MW per module (often 50-300 MW), designed for mass production, easier transport, and flexible deployment.

They contrast with traditional gigawatt-scale reactors by offering shorter construction times, lower upfront capital per unit, smaller land footprints, and enhanced safety via passive systems.

Many designs use light-water, gas-cooled, molten salt, or fast reactor technologies.