707
'At some point you've got to make money': Goldman's top AI skeptic warns the clock is running out ahead of OpenAI and Anthropic IPOs (fortune.com)
submitted 2 months ago by to c/technology@lemmy.world
147 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] 21 points 2 months ago (1 child)

I think the price point is where your theory falls apart. The AI companies are spending far more than Netflix and they will need to charge more than the average family can afford in order to turn a profit. (Especially in the US where people are struggling to pay for gas and groceries right now.)

Take a look at the new token-based business services. People are blowing through their monthly allotment in a couple of days doing the same things they had been doing previously. Businesses won't be able to afford increasing their AI spending by 10x to keep up.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] -1 points 2 months ago (3 children)

    I think it will have to consolidate down to a few players - I certainly don’t expect that there is a market for 100 different systems. And I think the personal market is a supplement to the executive suite market, so getting a few people addicted for cheap both brings in a small revenue stream and dumbs a population down enough that they have no choice but to function without it (imagine telling someone in 2001 that they would one day pay $1700 for a cell phone - now it’s just considered the cost of living)

    The businesses who overdid it on tokens have a choice to make: hire back workers or switch to a different AI company. Maybe one day they will look at humans (although salary may have to increase due to negative reputation), but I think there are a good number that will just shop around to another AI company for a better deal.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] 11 points 2 months ago (2 children)

    There's just no way to pay for the cost of these services, though.

    When someone constructs a 100 MW data center (now considered a smaller one for new construction), that's about $2 billion in total costs to outfit the whole operation. And then once it's on, we're talking something like $10-20 million/month in electricity alone, and a few million in other costs. How many $20 subscriptions do you need to sell just to break even with your operating expenses? How many $100/month subscriptions do you need to sell to make a dent on your interest payments on the construction? Will there be a market for $1000/month subscriptions from millions of customers? If not, how's this all going to be paid for?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 6 points 2 months ago

    They certainly cant make their money back writing personal emails and doing kids homework, but I dont think thats where they are aiming in the long run. We may end up with big business and military paying big money for the real frontier models and everyone else using lightweight local models on their own hardware, cheap models integrated into existing applications, and watered down frontier models on subscription.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 1 point 2 months ago (2 children)

    You may be right (I hope you are). Do you think government funding would offset this?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 3 points 2 months ago

    I don't think government funding can actually offset the crash in consumer and business demand being insufficient to cover the cost of the most expensive models on the most expensive GPUs. But if you look through my comment history I've made the comparison to supersonic flight, because I genuinely believe there's a possibility that governments fund the expensive branch of this technology for their own military or surveillance or law enforcement purposes without the benefits necessarily actually spilling out into normal commercial applications.

    We've hit the point where training a model (both pre training and post training) isn't the expensive part, and the expensive part is actual inference, which makes it hard to scale the most expensive models to where it's useful for a lot of people. So it might be that the companies and governments that can afford to operate an expensive model might be the only ones to do it. And they'll be able to, without necessarily the public being able to have access to the same tech.

  • permalink
  • fedilink
  • source
  • parent