We dug deep into this history on the new @stepchangeshow.bsky.social .

🎧 Listen here → www.stepchange.show/p/data-cente...

AI is so power-dense it’s breaking the efficiency curve.

The next frontier?
Doing for GPUs what these breakthroughs did for CPUs:
maximize utilization, minimize waste.

That’s why data center energy use stayed flat for almost a decade,
even as internet traffic grew 10x+.

But the miracle is ending.

Containerization (2010s)
Virtual machines were heavy. Containers were light.

Kubernetes (from Google’s Borg) let apps run anywhere, efficiently.
The final layer of abstraction.
The cloud could scale without matching energy growth.

Physical Design (2000s)
For decades, data centers were cooled like freezers.

Google re-imagined them as a warehouse-scale computer.
Hot/cold aisle containment. Smarter siting in cool+dry climates. PUE to push competition

PUE dropped from ~2.0 → ~1.1. Waste dramatically decreased.

Virtualization (1990s)
Server closets were full of idle machines.

VMware solved it by slicing one server into many “virtual” ones.
This made the cloud business model possible.
AWS didn’t rent physical boxes—they rented slices.

Time-Sharing (1960s)
Mainframes once ran one job at a time while engineers waited in line.

MIT’s CTSS system let dozens of people share a computer simultaneously.
The first leap from “single-task” to maximizing utilization.

That wasn’t luck.
It was the result of a 50-year war on waste.

Here are the 4 big turning points in efficiency that powered the modern data center—
and why that era is now ending. 👇

Didn’t he say all the same stuff last time.. and yet coal plants closed?

Thanks for diving in Peter!

Where are we in electrifying parts fertilizer production?

It seems that we could electrify all ag transport (inputs and exports). And then elements of supply chain.

From this: www.nytimes.com/interactive/...

Although source is the great @ourworldindata.org

I am one of many! Chatting with others (often parents) here: newsletter.climatepapa.com

Hiiiii!!!