The AI Crash will happen. The question is when. (Chart from Silicon Angle.)
I have spent my life being early to tech parties, the good and the bad. I started warning about the Dot-Bomb in early 1997. The problem was that if you sold out at that warning, you would have missed most of the run-up.
That’s why people don’t sell out. They keep one foot out the door, ready to bolt. Then they get run over because everyone else is doing the same thing.
In the case of AI, there are three possible break points.
- Data center growth slows, because revenue isn’t keeping up with investment.
- Application growth slows, because the models don’t meet expectations.
- An event occurs that shows the final price for AI assets.
The last is what happened with the Internet trade 25 years ago. AOL’s purchase of Time Warner (AOL shareholders got a majority stake in the resulting company) set the bar for what Internet assets could be worth. I called the deal stupid and I was right.
Something like this could happen again. If Microsoft were to buy OpenAI, for example, it would show the upper value limit for LLMs. I suspect major buyers are holding off on making bids for just this reason. Whoever makes the move second will get a better deal.
Other Scenarios
The other two scenarios can be seen in earnings and corporate announcements.
We know, for instance, that Nvidia is overvalued by any possible measure. It’s still rising, with a price to earnings ratio of 57, because it’s growing so incredibly fast, and with such incredible profits. The trouble is that any slowdown will take the whole chip complex with it, a sell-off triggered by a slowdown in data center purchases.
That’s why Microsoft keeps laying people off. That’s why the perks are gone in big tech, and those who are left are being taught the wonders of China’s “9-9-6” schedule, working from 9 to 9 six days a week. It’s what my dad did when running his TV repair shop in the 1960s. (He quit at 6 on Saturdays, the slacker.)
The canary in the second coal mine is ServiceNow. They sell AI applications, from a base in business process automation. Sellers come out before each earnings announcement. But so far, the company has delivered the goods, an annual growth rate of 23% and net income rising at double that rate.
This matters because application revenue, and productivity, are what everyone expects from AI. If those slow down, there’s less reason to buy. ServiceNow is currently selling for over 20 times its annual revenue, and its price to earnings ratio is twice that of Nvidia.
Important to note that a single AI vendor’s “miss” on earnings won’t trigger the fall, unless that vendor is Nvidia. It will take software misses across a wide front to convince traders the numbers aren’t just one side beating the other.
Let’s Summarize
A sale of OpenAI would show the top. An Nvidia miss on earnings could cause a collapse. Finally, misses on software earnings would show slowing momentum, and end the buying.
