It looks like the generative AI bubble is finally bursting.DADO RUVIC/Reuters
Kean Birch is director of the Institute for Technoscience & Society and Ontario Research Chair in Science Policy at York University.
For the past couple of weeks, the share value of the S&P 500′s bellwether tech stocks has been falling. These include Alphabet Inc. GOOGL-Q, Apple Inc. AAPL-Q, Amazon.com Inc. AMZN-Q, Meta Platforms Inc. META-Q, Microsoft Corp. MSFT-Q, Nvidia Corp. NVDA-Q, and Tesla Inc. TSLA-Q. As these seven corporations now represent a significant proportion of the U.S. stock market, any decline in value pushes down the overall public market, and this has been causing some panic.
On Monday that panic evolved into pandemonium. Global markets tumbled, but Big Tech tumbled more. At one point, the AI standard-bearer Nvidia was down more than 7 per cent and the Magnificent Seven lost more than US$500-billion in market capitalization.
The question of why this is happening now is interesting: It looks like the generative AI bubble is finally bursting.
I wrote in April this year that generative AI technologies look like money pits with significant social costs attached: this prediction seems increasingly on point. Investors are less and less confident that generative AI technologies will provide the necessary returns on the huge investments made.
Several analyst and investor reports have come out recently making similar points. First, David Cahn at the venture capital firm Sequoia argued that generative AI needs to generate US$600-billion in revenues to pay back current infrastructure spending – and we’re nowhere near this. Then, Jim Covello, Head of Global Equity Research at Goldman Sachs, argued that, in light of future expected infrastructure investment, generative AI needs to find a “$1-trillion problem” it will solve. It’s still not clear what this could be. Finally, the hedge fund Elliott Management has stated that Nvidia is in an AI “bubble land” and that many AI technologies “are never going to actually work” or “will take up too much energy” – undermining the hype around generative AI.
All of this highlights the problem of collectively putting too much money into one technological bet when it doesn’t offer a clear case for doing so.
Ross Sandler, an analyst at Barclays Bank, argued that the investment in generative AI is large enough for 12,000 new products, which sounds like an extraordinary number – and not in a good way. Mr. Covello noted that “people generally substantially overestimate what the technology is capable of today,” especially when it comes to reducing operational costs. He also pointed out that while the AI they use at Goldman Sachs is faster, it raises costs six-fold. And the International Energy Agency has calculated that an AI-driven search application requires ten times the electricity of regular search. Consequently, generative AI not only requires a significant investment in energy infrastructure, it’s also going to have a significant knock-on effect increasing energy costs across the board. Eventually we’re going to have to pay even more for these costs if firms want to generate revenues.
Other studies are showing that generative AI is not doing what its boosters have promised, namely raising productivity. Forbes, for example, has reported on research by the Upwork Research Institute showing that “77% of employees using AI say it has added to their workload and created challenges in achieving the expected productivity gains.”
A final question all of this raises is: Why do so many people buy into the generative AI hype? I think there are at least three reasons for this; one relates to investment logic and the others to corporate strategies.
First, all bubbles are driven in part by a self-fulfilling fear of missing out. Investors know that a bubble is a bubble, but they have to participate because they still make money. The past couple of years have been very good to investors in the stock market, especially those investing in the so-called “hyperscalers” whose infrastructure investments underpin generative AI – for example, Alphabet, Amazon, Microsoft, and Nvidia. But it now looks like these corporations and others investing billions in data centres would be better known as “hype-scalers”: they’ve driven up expectations and investment but with little to show for it up to now.
Second, generative AI technologies are currently underpriced or free. As Mr. Cahn at Sequoia noted, going forward firms will need to start generating revenues from their technologies, which means charging for them. Customers are unlikely to pay for technologies with ambivalent functionality – to say the least – and investors are coming around to this perspective.
Last, generative AI technologies have also enabled some corporations to capture more revenues from their operations. Finance YouTuber Sasha Yanshin makes the point that if you look at Alphabet’s most recent 10-Q report you can see that their paid clicks through Google Search have increased 5 per cent while impressions through Google Network have fallen 13 per cent. The reason for this is that AI-driven Search keeps users on Google’s own platforms rather than sending them to their Network partners, enabling them to capture more advertising revenues. Here, generative AI looks like it’s reinforcing market concentration.