How Doubtful Hype-Driven Assumptions About AI Now Spur Stock Market

The dangers of the AI hype cycle are becoming clearer as valuations approach stratospheric and, arguably, mathematically unachievable levels. 

There are signs the market is beginning to show some hesitancy as the reality of AI performance begins to erode questionable assumptions about future potential based on current performance. JL 

James MacIntosh reports in the Wall Street Journal:

As indexes hit fresh records, four tech stocks - Nvidia, Microsoft, Apple and Alphabet - added more value than the rest of the S&P 500 put together. Behind the rises is AI. (But) there are risks: demand falls because AI is overhyped; competition reduces prices; and what if scale doesn’t matter and businesses want smaller, dedicated AI models trained on their own data, or that startups can do AI just as well as the giants? Fast commercialization of LLMs has become the baseline for investors who have pushed up other stocks, such as suppliers of electricity, cables and data centers. It is a sign of how much investors have swallowed the hype that they are betting on distant potential beneficiaries of AI

Remember all the bulls earlier this year getting excited that the rally was broadening out beyond the “Magnificent Seven” stocks, and how wider gains were a sign that the market’s rise was sustainable? Not so much.

This month, as indexes hit fresh records, just four giant technology stocks added more market value than the rest of the S&P 500 put together. A brief burst of outperformance by smaller stocks seems to have petered out again.

Nvidia NVDA -0.51%decrease; red down pointing triangle, Microsoft MSFT -2.11%decrease; red down pointing triangle, Apple AAPL 0.48%increase; green up pointing triangle and Alphabet GOOGL -1.58%decrease; red down pointing triangle between them have added over $1.4 trillion this month, more than the other 296 stocks that rose put together. Half of the gain was just one company, chip maker Nvidia.

Behind the rises in the biggest stocks this month and this year are two trends, which have already run a long way: Artificial intelligence and higher-for-longer interest rates. Any trend can always go further, but there are challenges to both. Today I’ll look at AI, with a return to the higher-for-longer narrative to follow in a subsequent column.

 

AI has had an astounding run since OpenAI unveiled ChatGPT to the world in late 2022, and Nvidia has been the biggest winner as everyone races to buy its microchips.

To see what could go wrong, note that this isn’t the usual speculative mania (though there was a mini-AI bubble last year). Nvidia’s profits are rising about as fast as its share price, so if there is a bubble, it’s a bubble in demand for chips, not a pure stock bubble. To the extent there is a mispricing, it’s more like the banks in 2007—when profits were unsustainably high—than it is to the profitless dot-coms of the 2000 bubble.

The threat to Nvidia’s share price is therefore about threats to its earnings. There are four risks:

1) Demand falls because AI is overhyped. The International Monetary Fund says AI will “transform the economy.” President Biden says AI is “the most consequential technology of our time.” And leading suppliers of AI—including the chief executive of OpenAI—said in a joint letter that “Mitigating the risk of extinction from AI should be a global priority.”  

Yet, large language models remain limited. I’ve yet to meet someone who is pleased to be faced with a customer-service chatbot—the main use case of which seems to be to try to figure out how to get past it to speak to an actual human. 

It helps programmers, but few outside the coding community got excited when, for example, GitHub transformed software development, long before ChatGPT came along. Microsoft’s Copilot can help beginners improve PowerPoint presentations, which might make office life a little less dull. And helping users with Excel is important. But again, a better “Help” function wouldn’t normally get investors’ pulses racing.

 

Don’t get me wrong. I’m deeply impressed by the progress of large language models, the technology behind ChatGPT. But all of my attempts to use it for work came with errors that took more time to fix than if I had just done it myself. It’s like having a 12-year-old help out.

Many LLMs fail basic tests, often in amusing ways—Google advises adding glue to pizza, for example. Copilot highlights the risk of being eaten by a cabbage when asked how to cross a river with a goat. Are they really ready for prime time? Perplexity is a great AI-assisted search engine, but the promise of AI goes a lot further than a better Google. The technology has yet to live up to the hype, and the risk is it takes much longer than buyers of Nvidia chips believe.

2) Competition reduces prices. Nvidia rules supreme in the high-end graphics processing units used for AI. But it faces competition from well-funded customers such as Alphabet and Meta Platforms, a slew of startups, and traditional rivals such as Intel trying to catch up. Even if they aren’t as good, they should limit Nvidia’s ability to charge pretty much what it likes.

3) Nvidia’s biggest supplier, Taiwan Semiconductor Manufacturing, might want a bigger slice. TSMC actually makes the chips. If it jacked up the price it charges Nvidia, no one else could step in to replace it in any reasonable time period. It might not want to risk its long-term relationship. On the other hand, it might want some of the cash pouring into Nvidia.

4) What if scale doesn’t matter? The designers of AI models think there is an advantage to getting big quickly. More use gathers more data, which makes the models better, which attracts more users, in a virtuous circle. The big should get bigger. Such a frenzy fuels the demand for AI chips.

If the argument is right, a lot of people who invested in the models that prove to be the losers will have wasted their money. The current boom in demand for chips should also die down once the also-rans give up. That is a way off but is a threat to future Nvidia sales.

If the scale-matters argument is wrong, and what businesses actually want are smaller, dedicated AI models trained in large part on their own data, or that startups can do AI just as well as the giants, a lot of money will have been wasted.

Nvidia Chief Executive Jensen Huang at the company’s conference earlier this year. PHOTO: JUSTIN SULLIVAN/GETTY IMAGES

Investors assume continued rapid growth from Nvidia, with the stock on a multiple of 38 times forward earnings. Superfast commercialization of LLMs has become the baseline for investors who have pushed up other stocks that should benefit, such as suppliers of electricity, cables and data centers. It is another sign of just how much investors have swallowed the hype that they are willing to bet on such distant potential beneficiaries of the AI boom.

I fear the market is paying too little heed to the risks. AI may be a big deal, but it’s unlikely to pan out the way people seem to think.