Hacker Newsnew | past | comments | ask | show | jobs | submit | keyworkorange's commentslogin

How are these models so cheap?


They are just not overcharging. Nvidia's top AI chip Rubin sells in 72-GPU racks for about $3.5–7.8M. A rack running Xiaomi's MiMo V2.6 Pro generates roughly 150–300B tokens a day, worth about $130–260k at Xiaomi's API price.

That's a payback of the infrastructure in a few weeks in theory. After a few weeks or a month, the only cost is electricity, and whatever they make after that is pure profit. This is why they can charge normal prices. Not $50 for 1M tokens.


Seriously! Same! Relieved to know I'm not alone in having this quirk.


This reminded me of a few subplots in Murderbot- (Do yourself a favor and check it out if you haven't, it's a fun, quick read)

But seriously, one would assume there's a reward system of some sort at play, otherwise why do anything?


Maybe you're thinking of subnet relays? https://tailscale.com/kb/1019/subnets



Always looking to conserve cloud spend, I'll give this a shot.

Heads up- there's an error on your landing page's Calendly link. "This Calendly URL is not valid."


My understanding is that these electronics need to be radiation hardened (eg; RADHARD) to prevent them from misbehaving in ways far worse than the Belgian bit flipping incident (https://radiolab.org/podcast/bit-flip). If you want to use commercial off the shelf (COTS) parts, you need to install them in triplicate and have them "vote". (https://apps.dtic.mil/sti/citations/ADA391766)


I wonder if we couldn't "harpoon" debris with lower cost tethers like the ones tested on TEPCE, but optimized for drag rather than power generation?

[PDF Warning] https://apps.dtic.mil/sti/pdfs/AD1161973.pdf


I could use one!


Dang, bummer. Those went fast! Thanks for posting though xenospn


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: