AI's growth constraint has moved from capital and code to concrete and copper. Physics, not software, is the bottleneck.

You can commit $200 billion in capital expenditure. You cannot will a power plant into existence.

This is the single most important sentence in AI strategy right now, and almost nobody in the technology conversation is saying it. AI’s binding constraint has flipped. It’s no longer capital or code. It’s concrete, copper, and cooling. Physics, not software, is the bottleneck.

The numbers are stark. Available data centre capacity worldwide sits between 11 and 14 gigawatts. Demand in the interconnection queue: 40 to 50 gigawatts. Grid interconnection wait times have ballooned to three to five years. Each gigawatt of data centre capacity supports roughly $10 billion in annual recurring revenue. Do the arithmetic: the gap between capacity and demand represents hundreds of billions in stranded revenue.

Hyperscaler requests have escalated from 100 megawatts — a large facility — to 1 gigawatt, the output of a nuclear power station. Cooling system suppliers are backordered to 2030. AI labs are deploying on-site gas turbines to bypass the grid entirely, because waiting for grid connection means waiting years for revenue.

The consequence? Technology companies are becoming energy companies. Alphabet acquired Intersect Power. Meta launched “Meta Compute” — a division dedicated to gigawatt-scale energy infrastructure. Microsoft signed the largest corporate clean energy deal in history. These aren’t PR gestures. They’re existential strategic moves. If you can’t power your data centres, you can’t serve your customers. Full stop.

But here’s where the story gets interesting — and, for those paying attention, optimistic.

Energy itself is on a learning curve. In 2026, 99% of new US electricity generation capacity comes from solar, wind, and storage. Battery storage costs fell 27% in a single year, to $78 per megawatt-hour. Solar has followed Wright’s Law for decades — every doubling of cumulative production drops costs by roughly 20%. This isn’t policy-dependent optimism. It’s manufacturing economics.

The pattern is global and accelerating. China has recorded 21 consecutive months of flat or falling CO₂ emissions — not because of climate virtue, but because renewables are simply cheaper. In Pakistan, 22 gigawatts of rooftop solar were installed in a single year, at one-third the cost of grid electricity. Consumers aren’t waiting for government policy. They’re opting out of centralised infrastructure because the economics are irresistible.

What does this mean for business leaders?

First, energy strategy is now AI strategy. If your organisation is planning significant AI deployment and hasn’t secured its energy supply chain, you’re building on sand. The companies that lock in energy infrastructure today — through power purchase agreements, direct investment, or strategic partnerships — are building moats that will compound for decades.

Second, the AI-energy nexus is the most underleveraged strategic opportunity in the market. Very few leaders understand both domains. Most AI strategists ignore energy constraints. Most energy executives underestimate AI demand. The gap between these two conversations is where enormous value will be created.

Third, the physical bottleneck is temporary but decisive. Learning curves will eventually close the gap — solar, wind, and storage costs will continue to fall, new nuclear designs will mature, grid infrastructure will catch up. But the next three to five years represent a window where energy access determines competitive position in AI. Miss the window, and the cost of catching up rises exponentially.

The AI revolution won’t be won by the company with the best model. It will be won by the company that can keep the lights on.

Azeem explores this topic with leadership teams. Begin a conversation.