Gartner: Data Center Power Demand Hits 565 TWh in 2026 — AI Servers Cross 31% of Total Consumption

Infrastructure Data — Gartner just quantified what Andy Jassy called “our single biggest constraint.” Data center electricity consumption will grow 26% in 2026, reaching 565 TWh. AI-optimized servers alone account for 31% — and will surpass conventional servers by 2027. Power is now the bottleneck for the entire AI stack.

The Numbers

The Crossing Point: 2027

The most significant data point: AI-optimized server power consumption will surpass conventional servers in 2027.

Conventional servers grew 1% in 2025 and will grow 1.2% in 2026. AI servers grew 84% in both years. Two curves, opposite trajectories, crossing in 2027. This is the compute layer restructuring in real time.

The Power Bottleneck

Gartner projects data center power to reach 1,200 TWh by 2030 — and explicitly states: “grid supply will be insufficient to meet the demands of future data center construction.”

This validates what Goldman Sachs projected in their $7.6 trillion AI capex analysis: power is only $358 billion of the total spend but the critical path for the other 95%. You can buy all the Nvidia GPUs you want — if you can’t plug them in, they’re paperweights.

The Map of AI Read

In the Map of AI, this is Layer 3 (Infrastructure) becoming the binding constraint for every layer above it.

Three responses to the power bottleneck are playing out simultaneously:

• SpaceX orbital data centers — move compute to space where solar power is unlimited
• Nuclear contracts — Vistra’s 20-year deal with Meta for 2,600 MW of nuclear power
• Custom silicon (Graviton 5, TPU) — more compute per watt, reducing power demand per inference

The companies that solve the power constraint — not the model constraint — will determine who wins the AI race over the next decade.

Related:
Goldman Sachs: Where $7.6 Trillion Goes
SpaceX: 1M Orbital Data Centers
AWS Graviton 5: Nvidia Tax Escape
AI Supercycle Financed by Junk Bonds

Source: Gartner Press Release (June 10, 2026)