I was recently reminded of a deal Yahoo did during the dot-com bubble where it invested $30m into a start-up. The start-up promptly used the $30m to buy banner ads on Yahoo. It didn’t survive the year but Yahoo got to book the sales as revenue while writing down the investment to zero.
I was thinking about this because of the spate of self-dealing in the broad AI space. We’ve seen Nvidia investing in Mistral (the French large language model [LLM]), CoreWeave (an AI cloud-computing company) and Intel, and recently putting $100bn into OpenAI. Then we see OpenAI do a deal with Oracle and another with AMD that will make it one of AMD’s largest shareholders.
Is OpenAI essentially using Nvidia money to buy a competitor’s chips? Perhaps more importantly, OpenAI does not have the money it is promising even though its most recent valuation, to enable staff to sell shares, is at $500bn.
If you’ve been wondering if AI is a bubble, the above says yes. Spend like crazy, don’t worry about revenue and do lots of deals between yourselves to keep everybody excited.
That said, even if it is a bubble, it doesn’t mean we all need to head for the hills in double-quick time. The point is, where are we in the bubble? Bubbles can last longer than anybody expects and AI in the form of LLMs arrived on the scene via ChatGPT only three years ago next month.
Spend like crazy, don’t worry about revenue and do lots of deals between yourselves to keep everybody excited
I suggest that two important points will keep this bubble going. First, most AI build-out is coming from free cash flow at the largest tech companies and, second, the demand, via data centres, remains absolute. There is no overcapacity, rather there is a shortage with every new data centre immediately filled with clients.
Then there’s the self-dealing. This is a bad sign, but there is some logic for the chip-makers — artificial general intelligence (AGI). This is the holy grail of AI and when (if) it is achieved the demand for AI chips will be off the already extreme charts. So chip-makers are pushing out as many chips as possible in any way possible to try to advance the day AGI arrives.
More LLMs will be extremely useful — OpenAI recently hit 800-million weekly users. I am one of those (I’m even paying every month) using ChatGPT all day, especially for some of the boring grunt work I do.
So, yes, AI is bubble territory but I’m not sure it will necessarily pop any time soon. I am watching the hyper-scalers to ensure they continue to spend on the capex and watching demand for new data centres as they’re being built. Weakness in either of these would have me concerned, but for now both data points remain strong.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.