Hacker News Reader

Comment by gjsman-1000 | Hacker News Reader

$45 billion for a 3 year rental.

What would be interesting to know how much did it cost xAI to build it ? Ai says between $18-$40 billion to just build, without running cost, but no idea how close to reality this is.

Anthropic is getting capacity from Colossus 1 not Colossus 2 it sounded like. The initial colossus capex was under $5B, making that an even more astounding payoff.

Edit: S1 states both are being leased so the 20-25B initial investment probably more relevant

... and a sign Anthropic couldn't find enough compute anywhere else, so they had to bite the bullet. Interesting.

The S-1 states that it gets capacity from both Colossus 1 and Colossus 2.

Closer to 18b than 40. Running costs are 1-2b a year.

More like $25 billion since 2025, with $7 billion of spending in the past 3 months. Look at page 22 of the filing.

The AI row of the capex table in the S-1 should be a pretty close approximation.

Nobody pays MSRP at that scale

Given global demand and that they were late to the order party, they probably paid more lol

Yes and no. They paid what everyone else pays for those gpus. NVIDIA make the profit and leaves crumbs for the rest. For other components they paid less, but since the gpus are the majority of the cost…

how much did SpaceX / xAI pay for these GPUs? After 3 years they'll probably be mostly deprecated.

Are GPUs from 3 years ago being deprecated today?

I thought I saw a report from someone at Google saying that they were still running 7+ year old hardware because of demand. Even if it is not state of the art, if it generates more than the electricity costs, keep it running until it dies.

Efficiency gains have been so intense that old hardware is still viable for serving brand new mid size models.

Likely legacy products are running on the old chips because the newer generations are capacity constrained.

Hetzner is still running and selling Haswell Xeons and GTX-1080s that are 10+ years old.

Definitely not silicon waste the second something faster arrives. There's still a world beyond cutting edge LLM slop after all.

And how many of them were diverted from Tesla?

Pretty sure that Tesla didn't use Colossus. Tesla used Cortex 1 and Cortex 2 which are at the Gigafactory in Austin.

Tesla famously never got to use them https://arstechnica.com/cars/2024/06/elon-musk-is-diverting-...

The article says: In December, an internal Nvidia memo seen by CNBC said, “Elon prioritizing X H100 GPU cluster deployment at X versus Tesla by redirecting 12k of shipped H100 GPUs originally slated for Tesla to X instead. In exchange, original X orders of 12k H100 slated for Jan and June to be redirected to Tesla.”

Reply on news.ycombinator.com