• GamingChairModel@lemmy.world
    link
    fedilink
    English
    arrow-up
    16
    ·
    1 day ago

    Once you get into things with useful generation and large context windows, or things like video generation, suddenly you need one or more $10,000+ pieces of hardware to run it.

    A Blackwell server with 72 GPUs costs about $3 million, plus requires 130 kW of power (about 3 residential homes’ max rated power through a residential 200A circuit box, for about $600-$1000/day in electricity cost).

    You’re gonna need to sell a lot of $20/month subscriptions to get that paid for, assuming that the server is good for 5 years. If it’s only good for 3 years, the economics are basically impossible.

    • SirEDCaLot@lemmy.today
      link
      fedilink
      English
      arrow-up
      2
      ·
      4 hours ago

      If it’s only good for 3 years, the economics are basically impossible.

      Also consider that as models improve, the newer frontier models that are doing actual useful work require significantly more compute and RAM than basic chat bots…

    • grinde@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      16
      ·
      1 day ago

      … assuming that the server is good for 5 years. If it’s only good for 3 years, the economics are basically impossible.

      And surely the big AI companies wouldn’t decide that hardware rated for ~3 years should be amortized over 5-6 years in order to massively inflate their value on paper.

      Surely.

      • SirEDCaLot@lemmy.today
        link
        fedilink
        English
        arrow-up
        2
        ·
        4 hours ago

        I’m sure Bernie Madoff’s fund had an ‘accounting puzzle’ too.

        I love how the publications call it a ‘puzzle’ rather than what it so obviously is- a bunch of people throwing $billions at tech they don’t understand which has no serious road to profitability anytime soon.