• mojofrododojo@lemmy.world
    link
    fedilink
    English
    arrow-up
    110
    arrow-down
    1
    ·
    18 hours ago

    when this bubble pops it’s gonna be horrific.

    google, meta, ms, so many more leveraged out huge investments in datacenters. nvidia is propping up whole segments of the fucking economy.

    https://www.wheresyoured.at/ai-is-a-money-trap/

    it’d be fun to watch if I could isolate myself from the chaos that will ensue, but we’re all gonna get fucked by the aibros, it’s only a question of which segment of the economy blows up first.

    • skisnow@lemmy.ca
      link
      fedilink
      English
      arrow-up
      24
      ·
      12 hours ago

      A lot of startups whose entire business model relies on OpenAI’s small model API calls costing under $1/Mtok, are going to go bust when OpenAI finally runs out of money and ramps the cost up tenfold.

      • Mika@sopuli.xyz
        link
        fedilink
        English
        arrow-up
        1
        ·
        4 hours ago

        It would be just cheaper to self-host something for the whole company then? Open-source AIs are there and they are very much competitive with proprietary solutions.

        • skisnow@lemmy.ca
          link
          fedilink
          English
          arrow-up
          3
          ·
          3 hours ago

          If you want OpenAI level response times you might be surprised how expensive self-hosting gets.

        • pfizer_dose@lemmy.world
          link
          fedilink
          English
          arrow-up
          9
          ·
          11 hours ago

          Yep it’s blitzscaling. Run it at a loss until it’s a necessity, then charge whatever the hell you want. They’re blitzscaling our right to intellectual property and our right to work.

      • Tollana1234567@lemmy.today
        link
        fedilink
        English
        arrow-up
        3
        ·
        8 hours ago

        ive seen a ton of billboards of startup AI comp in west coast, i assume every new one that appears on these billboards, the old ones go under.

    • Thorry84@feddit.nl
      link
      fedilink
      English
      arrow-up
      39
      arrow-down
      1
      ·
      edit-2
      17 hours ago

      There is another factor in this which often gets overlooked. A LOT of the money invested right now is for the Nvidia chips and products based around them. As many gamers are painfully aware, these chips devalue very quickly. With the progress of technology moving so fast, what was once a top of the line unit gets outclassed by mid tier hardware within a couple of years. After 5 years it’s usefulness is severely diminished and after 10 years it is hardly worth the energy to run them.

      This means the window for return on investment is a lot shorter than usual in tech. For example when creating a software service, there would be an upfront investment for buying the startup that created the software. Then some scaling investment in infrastructure and such. But after that it turns into a steady state where the input of money is a lot lower than revenue from the customer base that was grown. This allows to get returns on investment for many years after that initial investment and growth phase.

      With this Ai shit it works a bit different. If you want to train and run the latest models in order to remain competitive in the market, you would need to continually buy the latest hardware from Nvidia. As soon as you start running on older hardware, your product would be left behind and with all the competition out there users would be lost very quickly. It’s very hard to see how the trillions of dollars invested now are ever going to be recovered within the span of five years. Especially in a time where so much companies are dumping their products for very low prices and sometimes even for free.

      This bubble has to burst and it is going to be bad. For the people who were around when the dotcom bubble burst, this is going to be much worse than that ever was.

      • mojofrododojo@lemmy.world
        link
        fedilink
        English
        arrow-up
        27
        ·
        edit-2
        17 hours ago

        yeah datacenters never really aged well, and making them gpu dependent is going mean they age like hot piss. and since they’re ai-dedicated gpus, they can’t even resell them lol.

        all this investment, for what? so some chud can have a picture of taylor swift with 4 tits?

        fucking idiots

        • howrar@lemmy.ca
          link
          fedilink
          English
          arrow-up
          3
          ·
          16 hours ago

          I don’t see why they can’t be resold. As long as there’s a market for new AI hardware, there will continue to be a market for the older stuff. You don’t need the latest and greatest for development purposes, or things that scale horizontally.

          • mojofrododojo@lemmy.world
            link
            fedilink
            English
            arrow-up
            9
            ·
            16 hours ago

            I didn’t say they couldn’t be resold, they simply won’t have as wide a potential user market like an generic GPU would. But think about it for a sec, you’ve got thousands of AI dedicated gpu’s going stale whenever a datacenter gets overhauled or a datacenter goes bust.

            that’s gonna put a lot more product on the market that other datacenters aren’t going to touch - no one puts used hardware in their racks - so who’s gonna gobble up all this stuff?

            not the gamers. who else needs this kind of stuff?

            • Passerby6497@lemmy.world
              link
              fedilink
              English
              arrow-up
              3
              ·
              edit-2
              3 hours ago

              no one important puts used hardware in their racks

              FTFY. Just about every msp I’ve worked for has cut corners and went with 2nd hand (or possibly grey market) hardware to save a buck, including the ones who colo in “real” data centers. I would not be surprised to find that we’re onboarding these kinds of cards to make a bespoke AI platform for our software customers here in a few years.

            • addie@feddit.uk
              link
              fedilink
              English
              arrow-up
              3
              ·
              11 hours ago

              I’m not sure that they’re even going to be useful for gamers. Datacenter GPUs require a substantial external cooling solution to stop them from just melting. Believe NVidia’s new stuff is liquid-only, so even if you’ve got an HVAC next to your l33t gaming PC, that won’t be sufficient.

              • mojofrododojo@lemmy.world
                link
                fedilink
                English
                arrow-up
                5
                ·
                9 hours ago

                not just those constraints, good luck getting a fucking video signal out of 'em when they literally don’t have hdmi/dp or any other connectors.

            • jkercher@programming.dev
              link
              fedilink
              English
              arrow-up
              5
              ·
              13 hours ago

              Also depends how hard the AI runs them. A good chunk of the graphics cards that were used as miners came out on life support if not completely toasted. Games generally don’t run the piss out of them like that 24/7, and many games are still CPU bound.

      • panda_abyss@lemmy.ca
        link
        fedilink
        English
        arrow-up
        10
        ·
        16 hours ago

        They’ll write this off as a loss and offset their corporate taxes

        Also china is a great example that you do not need all the latest hardware, but it does help

        • SpaceCowboy@lemmy.ca
          link
          fedilink
          English
          arrow-up
          11
          ·
          16 hours ago

          It would probably be fine if there was someone competent in Whitehouse to manage the fallout.

          But since there isn’t, you are very likely correct.

    • vector42@programming.dev
      link
      fedilink
      English
      arrow-up
      8
      ·
      13 hours ago

      Came here to see if someone had mentioned Ed Zitron’s blog. His last two pieces on the AI bubble are fantastic reads.

      • mojofrododojo@lemmy.world
        link
        fedilink
        English
        arrow-up
        3
        ·
        12 hours ago

        yeah secondary knockon effects - once nvidia realizes it’s not going to actually sell 5 gpus per human being, the datacenters for them evaporate, then the power production to feed those datacenters becomes pointless…

        an effective administration would mandate all renewable energy for this purpose, so when it implodes they could at least derive a benefit from the expanded production… but no, trump will have them build coal plants for it all. or like grok, methane powered generators fml

          • mojofrododojo@lemmy.world
            link
            fedilink
            English
            arrow-up
            3
            ·
            edit-2
            8 hours ago

            bringing old reactors online may end up an overall positive (say, if the ai bubble pops soon but the reactors still come online and displace fossil sources) but I’m dubious about smr’s still. it just seems like more chances for radionucleotides to get smeared everywhere if they become ubiquitous.

    • BetaBlake@lemmy.world
      link
      fedilink
      English
      arrow-up
      7
      ·
      15 hours ago

      Hopefully sooner rather than later, and maybe Elon can stop poisoning a neighborhood in Memphis with Grok

      • mojofrododojo@lemmy.world
        link
        fedilink
        English
        arrow-up
        6
        ·
        12 hours ago

        yeah that’s one of the more egregious examples, basically a methane factory that eats prodigious amounts of water and power, all in process of giving us MECHAHITLER.

        what’s not to love?