News report (WSJ):
OpenAI recently missed its own targets for new users and revenue, stumbles that have raised concern among some company leaders about whether it will be able to support its massive spending on data centers.
Well, duh.
Katie
5 Likes
It’s almost like there is no business model.
Huh, who knew?
2 Likes
It seems that the massive expansion of data center capacity to support OpenAI, Anthropic, etc., has an aspect not covered in the general press: hardware ages out and goes obsolete pretty quickly in the scheme of things. Maybe 2-3 years or so for the GPUs, etc.? So, in addition to all the current investment, the present value of all the replacement investment could dwarf the numbers we see in the headlines. There would have to be a significantly steep efficiency curve to reduce the cost of producing AI in the out years.
Good time for a bubble burst to restore reality.
Katie
1 Like
Speaking of that …
1 Like
Weird to think that I, as an individual, am more profitable than a company with a market cap of $834 B.
Well done me! (And probably all of you too.)
8 Likes
$184 B is admittedly a big number.
But successful companies like Amazon and Uber were not profitable for many years, and they still had big valuations.
In 2000, I was more profitable than Amazon. I can only wish that that were true today.
2 Likes
OpenAI may be facing insurmountable competition.
Which means OpenAI may lose its job - to AI.
Count me among the secretly amused, although I have respect and sympathy for any hurt by either AI’s rise or fall.
1 Like
There’s not profitable, and there’s figuratively burning cash.
3 Likes
Amazon got a big boost from Prime and later AWS became its big money maker. AFAIK every new user of OpenAI increases the amount of compute that the company has to get from Microsoft, AWS, and Oracle, etc. Recently it has started shutting down some projects such as Sora. To reduce costs?
Amazon is selling both products and commodities. Gemini is part of a product strategy. OpenAI and Anthropic except for a few marginal efforts are almost pure commodity plays so far. That’s the biggest problem with the AI bubble. Massive investment but no moats to protect against switching or to capture and retain repeat customers. I suspect part of Apple’s reticence to jump in wholly is its intrinsic aversion to commoditization.
Katie
1 Like
The difference is with Uber and Amazon you could see how they get to profitability.
With AI there is no moat. DeepSeek, et al open source competitive models on a regular basis. This limits what Anthropic and OpenAI can charge.
1 Like
I’ll argue that there is at the enterprise level, which is where AI will make most of its money. Linux is free, but Microsoft is nonetheless a large and profitable concern because of its enterprise dominance. Anthropic, and now OpenAI, deliver more than just access to frontier models to their enterprise customers.
I’ve spent some time looking at the economics of all Anthropic in particular. By extension all of the AI companies. Based on what I’m paying vs usage for Max subscription I think their compute cost for inference is underpriced by a factor of 3-10.
I don’t see how they can raise prices that much.
1 Like
This assumes that the cost to provide inference—which is already lower than the cost of training—remains at its current level; it’s generally expected that it will in fact decrease significantly over time. In addition, inference is currently supply-constrained, i.e., there is more demand for it than there is supply. The enterprises that want or need access to inference will pay more for it if the expenditure justifies itself in terms of capacity, productivity, or profitability.
Consumers and smaller enterprises may be willing to roll their own AI stack using a free, open-weights model. Larger enterprises will likely do what larger enterprises do: engage a trusted third party to do the heavy lifting and take on some of the risk.
There was quite a long period when Amazon tried very hard NOT to make a profit and it still does in many countries (so it can avoid taxation), but it was fairly quickly making significant returns on investment from its innovative approach to a very well-understood business model (selling things by mail order). Selling and expanding the surplus computer services that it had built for its own use was genuinely brilliant. Having dominated the market, driving out many competitors, was another obvious path to profit.
Investors are betting on the AI companies pulling off something similar, but they don’t really have a product, they’re in profound competition for consumers and they are having to buy computing power on a vast scale from people like Amazon who’ve already dominated that market.
It’s a familiar pattern from great historical bubbles: as sensible governments warn all investors, past performance of a company or sector or type of innovation says absolutely nothing about future performance.
1 Like