Who Powers AI Right?

The AI boom is heading to Southeast Asia. Fast. Data center power demand across ASEAN is set to quadruple to 10.7 GW by 2035. Malaysia alone expects a sevenfold jump. ByteDance is pouring $2.1 billion into a Malaysian AI hub. Every major hyperscaler is in the mix.

Here’s the fork in the road: three quarters of Southeast Asia’s electricity still comes from fossil fuels. But renewables are already 13% cheaper than coal across Asia-Pacific and heading for 32% cheaper by 2030. Southeast Asia doesn’t need fossil fuels to power its data center boom. The economics say build clean.

Globally, the same split is playing out. Here’s who’s doing what.

The good-ish guys. Microsoft matched 100% of its global electricity with renewables. 40 gigawatts across 26 countries. Delivered, not pledged. Google dropped $4.75 billion on Intersect Power to co-locate data centers directly with solar, wind, and batteries. China is physically relocating compute to its western provinces, where hydro, wind, and solar are abundant, with an 80% renewable mandate for new data centers by 2030. Shanxi, a coal province, just hit a milestone: renewable capacity now exceeds coal.

The murky middle. Amazon claims five straight years as the world’s largest corporate buyer of renewables. Its own employees say only 22% of actual data center power is renewable. The rest is accounting. Montana’s AG and 15 other states are investigating. Meta is somehow both the single biggest clean energy buyer (8 GW of solar, 6.6 GW of nuclear) and a company building gas plants in El Paso and Louisiana while selling the risk to private equity. OpenAI’s Stargate project is targeting zero-emission power for its Wisconsin campus, but the full buildout needs 15 GW across 10 data centers. That’s New York City levels of power with no clear plan for how the gap gets filled.

Subscribe now

The bad guys. Elon Musk’s xAI has built two of the world’s largest data centers in Memphis, powered by rows of unpermitted gas turbines. The neighborhood has a cancer risk four times the national average. Children are developing respiratory problems. At the recent Southaven permit hearing, not a single resident spoke in support. The NAACP calls it a civil rights issue. The Southern Environmental Law Center is suing.

The strategic picture. Everyone in AI is optimizing for speed. More compute, more power, more capacity, now. But the greatest opportunity is in disrupting the old models entirely. Co-locate with clean power like Google. Move compute to where the renewables are, like China. Design for communities, not against them.

Special kudos to Anthropic, which just pledged to cover 100% of electricity price increases caused by its data centers and pay for all grid upgrades. The cost of powering AI should fall on the companies building it, not everyday Americans. That’s a new model worth watching.

A note on irony: yes, we’re aware that writing about AI’s energy footprint using AI has its own energy footprint. This edition was researched and written with assistance from Claude. We think the conversation is worth the watts.