The most important AI deal of the summer is not about chips or models. Bloom Energy and Brookfield have expanded their partnership to power AI data centers from $5 billion to $25 billion, a fivefold jump that says the quiet part out loud: the bottleneck of the AI boom is no longer compute. It is electricity.

Why Fuel Cells, Why Now

New AI campuses need hundreds of megawatts, and utility grid connections now carry waiting lists measured in years. Fuel cells sidestep the queue — deployed on-site in months, generating around the clock, no transmission lines required. That speed premium is exactly what data-center developers are paying for, and it explains how a fuel-cell maker landed one of the largest energy commitments of the decade.

The Numbers Behind the Land Grab

  • Data-center power demand is projected to more than double by 2030, driven overwhelmingly by AI training and inference.
  • Hyperscalers have signed nuclear restarts, geothermal pilots, and gas deals — anything that delivers firm power fast.
  • Grid interconnection queues in prime markets now stretch three to five years, making on-site generation the default plan B.

Who Profits Down the Chain

The obvious winners are power producers with speed to deploy. Behind them: electrical equipment makers, cooling specialists, grid-software firms, and the construction trades in data-center corridors. Communities are learning to negotiate too — tax revenue and jobs in exchange for land, water, and grid impact. Investors have started treating electricity as the clearest AI trade of 2026: models change monthly, but megawatts are scarce for years.

The Risk Nobody Prices

If AI efficiency improves faster than expected — smaller models, better chips — some of this capacity could overshoot demand. The counterargument: cheap intelligence creates its own demand, and every efficiency gain so far has been swallowed by new usage. Twenty-five billion dollars says the second camp is right.