It seems obvious that we’re in the middle of an AI bubble. Here’s a good post from David Cahn at Sequoia:
In September 2023, I published AI’s $200B Question. The goal of the piece was to ask the question: “Where is all the revenue?”
At that time, I noticed a big gap between the revenue expectations implied by the AI infrastructure build-out, and actual revenue growth in the AI ecosystem, which is also a proxy for end-user value. I described this as a “$125B hole that needs to be filled for each year of CapEx at today’s levels.”
This week, Nvidia completed its ascent to become the most valuable company in the world. In the weeks leading up to this, I’ve received numerous requests for the updated math behind my analysis. Has AI’s $200B question been solved, or exacerbated?
If you run this analysis again today, here are the results you get: AI’s $200B question is now AI’s $600B question.
In other words, in order to justify the capex spent on AI, the AI industry needs to generate about $600 billion in revenues. While it’s likely that AI will eventually generate that much revenue and more, it seems unlikely that it will do so any time soon. (I don’t agree with the claims of people like Leopold Aschenbrenner who foresee AGI arriving in 2027 or 28.)
Cahn continues his argument:
One of the major rebuttals to my last piece was that “GPU CapEx is like building railroads” and eventually the trains will come, as will the destinations—the new agriculture exports, amusement parks, malls, etc. I actually agree with this, but I think it misses a few points:
Lack of pricing power: In the case of physical infrastructure build outs, there is some intrinsic values associated with the infrastructure you are building….
Investment incineration: Even in the case of railroads—and in the case of many new technologies—speculative investment frenzies often lead to high rates of capital incineration….
Depreciation: We know from the history of technology that semiconductors tend to get better and better…This will lead to the rapid depreciation of the last-gen chips….
Winners vs. losers:…Founders and company builders will continue to build in AI—and they will be more likely to succeed, because they will benefit both from lower costs and from learnings accrued during this period of experimentation.
There are a few observations we can make:
AI infrastructure is capital intensive and very competitive. There is not much entrepreneurial opportunity to be had here. There was opportunity, probably when OpenAI was first founded, but the commodification of large language models suggests that only the most well-capitalized hyperscalers will be able to play in this market going forward. (There may be opportunities in small, niche language models, though it remains to be seen whether much larger language models generalize across all domains.)
Consumers are as fickle as ever. They benefit because the cost of artificial intelligence is declining towards zero, which leads to a surfeit of products aimed at their interests. But it seems hard to build a sustainable company on the backs of fickle consumers, for whom an alternative is just a click a way.
Enterprises—by which I mainly mean legacy non-tech companies—desperately want to understand AI, but suffer from a lack of expertise and resources. I think there is significant opportunity here. The rest of this post provides some thoughts about this opportunity.
I see a lot of legacy industries looking at AI technology. The general reaction I see can be described roughly like this: “Yeah, I’ve played around a bit with ChatGPT, and it’s a really interesting technology, and I can sort of kind of see how AI might help my company do [thing] more efficiently but I don’t know how to get from that fuzzy notion to actual execution.”
And I think that’s where the opportunity lies. I think that entrepreneurs who want to build AI products for the enterprise would do well to pick a particular industry vertical, and build for customers in that vertical. Become the recognized industry leader in applying AI to a vertical. Sure, the initial TAM is smaller. But if you build a scalable and repeatable customer acquisition process, nothing stops you from expanding your market by pursuing adjacent or complementary industry verticals.
One way that this might look is that you build a product to help an industry manage its implementation of AI and layer some consulting on top of that product. For example, maybe chemical manufacturers want to use AI to optimize their factories’ operations but they need special tooling to sit on top of OpenAI’s or Anthropic’s LLMs. An entrepreneur can build that tooling and provide consulting services to the chemical manufacturing industry to help them use the tooling. As for adjacent industries that our entrepreneur can expand to once she has developed a repeatable sales process: these run the gamut from pharma & biotech, to agriculture, to oil and gas, to plastics, and many more.
