META

The numbers behind Meta’s new AI site are honestly absurd

AI data centres are getting so big, their power use now matches entire U.S. states. Meta’s new Hyperion AI data centre is one example.

Once finished, it is expected to use as much electricity as South Dakota.

To power it, Meta plans to back 10 natural gas plants in Louisiana.

Together, they are expected to produce around 7.5 gigawatts of electricity, slightly more than South Dakota’s total power capacity.

The wider data centre project is expected to cost $27 billion.

That stands out because Meta has spent years talking about its climate goals.

It regularly shares sustainability reports and has invested heavily in renewable and nuclear energy.

This project raises new questions about how those goals hold up as AI infrastructure grows.

Natural gas is often framed as a temporary energy source while cleaner options like solar, batteries, and nuclear scale up.

But that argument is getting harder to defend.

What you should know:

  • Meta’s new AI data centre could use as much electricity as South Dakota.

  • The company is backing 10 natural gas plants to support it.

  • Those plants could push Meta’s emissions much higher.

Green-ish at best

Renewable energy has become cheaper, while gas equipment has become more expensive, which makes Meta’s decision more striking.

According to TechCrunch, the Louisiana plants could produce 12.4 million metric tons of CO2 each year based on Department of Energy data.

That is about 50% more than Meta’s total carbon footprint in 2024.

And that number likely does not include methane leaks from the gas supply chain.

That matters because methane is a much more powerful greenhouse gas than carbon dioxide.

Even small leaks can make natural gas far more damaging than it appears on paper.

Meta’s latest sustainability report does not mention methane or natural gas.

But if these plants move ahead, gas could become one of the company’s biggest sources of emissions.

For an industry obsessed with optimisation, this feels wildly unoptimised.- MG

Keep Reading