Hacker News Reader

Comment by ftufek | Hacker News Reader

A 8xH100 server sells for like 300k+$ (the A100s go for half, still very expensive), I'd imagine the margins are pretty high on that. The demand for ML models (therefore servers) will likely grow now that startups seem to switch from crypto hype to ai hype and big tech derives huge value from ml workloads.

View on HN

Yep. If Nvidia can sell their datacenter cards (not even full servers) the revenue is basically 10x for nearly equivalent silicon (>$12k direct b2b price vs $1600 through retail channels per card). All they have to do is switch their semiconductor orders to the server chips and for the 8x servers build some interconnect boards.

The reason you can’t find 4090s is because Nvidia no longer cares about retail consumers.

Is this possible: the "crypto hype" and the "ai hype" and this nvidia quarterly earnings all intersecting at the need for high-end chips running on a bunch of servers.

In a supposed down economy c. early next year, offering processor power and hard disk space from home machines as a source of passive income could be the next OnlyFans.

Vast.ai already does that, it's fine if you're working on public datasets, but big no-no if you're working with private/expensive datasets imo.

I don't see how crowdsourcing compute would be possible without giving access to all the data being crunched, which would be a complete deal-breaker for many (most?) businesses, I would imagine. So unless someone can figure that part out, it's probably a non-starter.

With disk space, you have the issue of liability—what happens when someone decides to store child pornography on my hard drive? That's a more solvable problem, but I'm certainly not envious of anyone who wants to tackle it.

Reply on news.ycombinator.com