Discussion about this post

User's avatar
Ada Nestor's avatar

I always find it interesting when people conflate yesterday's data centers with the hyperscale AI campuses being proposed today. They're fundamentally different in terms of power demand, water use, footprint, and infrastructure. Pretending there's no distinction doesn't make for an honest discussion.

Albert Cory's avatar

Good analysis. I was at Google in 2006 when The Dalles (in Oregon) was being constructed, at the time one of the biggest in the world. This was long before the term "AI Data Center" became a redundancy. These were not the Nvidia GPU-heavy servers we see now.

We can ask then, why AREN'T tech companies buying up all that commercial urban real estate, whose prices are plummeting? Why are they buying up farmland instead?

Gemini (obviously Google's) notes: "A traditional data center rack uses 10 to 15 kilowatts (kW) of power. A brand-new AI rack packed with next-generation chips (like NVIDIA's Blackwell GB200 NVL72) requires 100 kW to 120 kW per single rack."

When people say, dismissively, "put it in a rural area" they ignore the fact that people live, farm, and ranch there. It's not garbage land and you don't get to fill it up with one story, windowless concrete buildings.

24 more comments...

No posts

Ready for more?