The AI sector isn't just a bubble, says one senior market analyst: It's the single biggest bubble the markets have ever seen, the bubble of bubbles if you will, a bubble so large it looms over the entire global economy and leaves Sir Mix-A-Lot breathless.
In unrelated news, the Associated Press has just reported that OpenAI's valuation has hit $500 billion, making a company that's never turned a profit into the most valuable startup in history.
All of which is to say: there is a hell of a lot of money riding on AI producing… well, something genuinely transformative in the near future. So much money that, if the bubble bursts, the pop may herald the kind of brutal economic fallout that can define eras.
This note to investors was first reported on by MarketWatch, and written by Julien Garran (who was formerly leader of UBS’s commodities strategy team, so presumably knows what he's on about).
Garran gets to that number with some creative economising using the Wicksellian differential to calculate a GDP deficit that altogether includes AI, real estate, VC investments, and for some reason NFTs. Under this metric the misallocation in a pre-crash 2008 was around 18% of GDP: Garran estimates that this figure could now be an eye-watering 65%.
Analysts naturally find ways (and leftfield differentials) to make the numbers fit their world view, but Garran does highlight some real-world examples of how the AI productivity boom is going. He cites a study where the task-completion rate for AI at a software company was between 1.5% to 34% and, even with the tasks AI was better at, it couldn't reliably replicate that success over time. There's a chart from another economist, based on Commerce Department data, suggesting that AI pickup among big companies is declining.
"We don't know exactly when LLMs might hit diminishing returns hard, because we don’t have a measure of the statistical complexity of language," says Garran. "To find out whether we have hit a wall we have to watch the LLM developers. If they release a model that costs 10x more, likely using 20x more compute than the previous one, and it's not much better than what's out there, then we've hit a wall."
Garran further points out that the audience using LLMs the most are costing these companies more in compute power "than their monthly subscriptions". And he could've added that most of us use them for free. He then comes up with a sentence that is supposed to be a dire warning but just sounds funny, about the bubble bursting and pushing the economy "into a zone 4 deflationary bust on our investment clock." Not the investment clock dammit!
I should re-emphasise Garran is an AI critic and works for a firm that is telling its clients not to over-invest or even invest in AI. So take everything in that context. This is no truth from on high but it does feel like the mood music around this technology is shifting slightly. Perhaps AI will change the world. Perhaps not like some think.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
