Energy infrastructure
The Data Center Buildout
Speed to power has overtaken LCOE. Seven questions reshaping how generation gets built and who gets served first.
Climate week, rewritten by load growth.
Data center demand is reshaping the energy system.
The buildout is the largest private capex boom since the railroads. By 2030, US data centers are projected to consume the energy equivalent of 8 New York Cities. Hyperscalers can turn one dollar of power into twenty dollars of compute. And it's changing everything.
How do they build faster?
Speed to power has surpassed LCOE as the deciding factor in new generation. Hyperscaler planning cycles run in months. Utility and regulatory cycles run in years. The core debate: can Silicon Valley move fast and break things, or will it move at the speed of regulation?
What energy mix will power them?
Renewables, gas, advanced nuclear, geothermal, long-duration storage. Hyperscalers are funding all of them. One speaker put it bluntly: after decades of natural gas as a bridge fuel, renewables are now a bridge to gas.
Can you bring your own power?
BYOP. BYOC. BYOE. The acronyms need to stop. Hyperscalers want speed, and they're done waiting on utilities. Behind-the-meter, co-located, off-grid, microgrids, take your pick. Many energy experts are calling BS on fully off-grid solutions. More likely, off-grid is a bridge, not a destination. The question isn't will they. It's can they.
Will data centers operate flexibly?
Data centers were built on uptime at all costs, but that formula is cracking. Flexible compute is the new frontier. Workloads that can shift across time, geography, or priority don't just ease grid stress. They get to power faster. A flexible three-nine developer beats a rigid five-nine developer stuck in queue until 2032.
How can they avoid community pushback?
"Data center" is now a four-letter word. Communities are organizing, suing, and winning. Community sentiment programs are becoming line items. Speed to power, it turns out, sometimes means getting an entire town to agree first.
Will consumer prices go up?
Utilities pass costs through to consumers, and energy prices determine elections. The headline nobody wants: grandma in Ohio subsidizes Microsoft's training run. The optimistic view: hyperscalers are paying premiums that fund innovation and eventually lower costs for everyone.
Can AI solve its own power generation crisis?
The strangest question of the week. The same workloads breaking the grid might also be what fixes it. AI-driven workload scheduling. Virtual power plants made of compute. The bull case: we're not in a power crisis, we're in a power coordination crisis, and AI is uniquely good at coordination. This could be a tech version of "the markets regulate themselves." But every hyperscaler and VC in the room hopes it's true.
The hottest take of the week: data centers will eventually push consumer energy prices lower, not higher. Time will tell. What I do know: the new winners won't be the cheapest electrons. They'll be the fastest ones.
Next insight
Abundance Is the Climate Strategy