What goes up must come down, which is to say bubbles that inflate eventually pop, with the end result being a recession and lots of bankrupt companies. And, not to spoil the story, that will almost certainly happen to the AI bubble as well. What is important to keep in mind, however, is that that is not the end of the story, at least in the best case. Bubbles have real benefits.
We are, of course, in an AI bubble, and billions of dollars of investment will evaporate when the bubble bursts. However, as Ben Thompson points out, bubbles aren’t inherently bad. They drive overexuberant investment into both infrastructure and innovations that are left over to power the next wave of technology. The dot-com bubble gave us the fiber that became the backbone of today’s Internet, and it brought hundreds of millions of people online, creating a market for consumer Internet firms and a population capable of working on the Web. In the case of AI, although the money spent on short-lived Nvidia GPUs may be wasted, the investment in chip fabs and power generation will be the foundation of an increasingly digital and electrified society.
The article (and your comment) reminds me of a comment I’ve made many times in the past.
At various times, after Intel would make a press release talking about how some great breakthrough they were expecting didn’t happen (or didn’t happen on schedule), you’d always see some pundit writing something like “well that’s the end of Moore’s Law”.
To which, I’d always respond something pointing out Moore’s law is an observation about overall industry trends, not about how any particular company is or is not able to keep pace with the rest of the industry.
If Intel doesn’t double CPU power every two years, but AMD or TSMC (or any of their other competitors) do, that doesn’t invalidate the observation.
I see this as something similar. When a bubble bursts, lots of companies will lose a lot of money. There will be bankruptcies and mergers. But the companies that survive (and there will always be a few, unless the tech was complete vaporware) will become the power players of that new industry, and the tech itself will form the basis for the “next great thing”, whatever that may turn out to be.
I’m a skeptic of the current crop of machine-learning systems. I don’t think they’ll replace programmers or other skilled workers. But I do think they can assist skilled workers in interesting ways, as I found when trying to use an LLM chatbot to solve some coding problems. I think they can be better at some customer support jobs than phone center clones reading from a script, in that LLMs can have access to a lot of information (history, tips, tricks, and hacks) than the scripted support workers do – possibly better than even the very best, well trained customer service manager. I’m currently putting speech recognition on my PBX, and the current machine-learning models are leagues ahead of the algorithmic models of yesteryear, even running on my decidedly low-end servers. But a “revolution” where LLMs replace people in a wholesale way? I think C-suite jobs are probably a better fit for “AGI” than the average wage earner.
(Hopefully that link works for everyone). Any way the gist here is that electricity is the main cost for AI so the most profitable thing AI can do is to make AI use less electricity. So if AI starts to use less electricity, then the rest of the rate payers will be stuck paying for all the sunk costs. That’s even if there is not an AI bubble burst. If the AI bubble bursting means a sharp drop in electricity use then same thing can play out. I’ve seen this story already in my area with the paper mills. They used to need a lot of water so municipalities upgraded the water systems, then they developed processes that used half the water. If the bursting AI bubble takes down electrical utilities, then that would be a worse crisis than the financial meltdown.
Bubbles happen when the price of some product (classically tulips) gets wildly exaggerated relative to the rest of the market and the economy realizes that its exaggerated is bogus and dumps it. The AI bubble is being recognized now because the market capitalization of AI stocks is grossly exaggerated. Today the market capitalization of NVIDIA is $4.62 trillion dollars. That is higher than the nominal gross domestic product of all nations except the US and China. GDP by Country - Worldometer
That’s not to say that Market capitalization is equivalent to gross domestic product, but it is an indication that the market cap of AI companies is seriously distorted with respect to the rest of the market. Some people made serious fortunes from the Tech bubble by selling their companies to greater fools at the peak of the bubble. I had a couple of good years because sales of my book on fiber optics boomed. The boom/bust cycle also was entertaining. But overall billions of dollars were vaporized when valuations of companies like Worldcom, Lucent, Nortel and JDS-Unipahse cratered.
I’m not a big fan of Apple right now but when it comes to AI they have done a lot right.
They have added an efficient and powerful NPU to all their devices and exploited this with their operating systems.
They have decided to outsource the AI component saving a bundle in capital expense while losing little to nothing in revenue and profits.
While I strongly dislike Apple’s built-in RAM as expensive and limiting, built-in RAM is the way to go with AI development and many of Apple’s MiniPC configurations are fully competitive on price and performance with all the other companies – big or small – at the moment.
As far as the issues with power mentioned in the article. these issues do not come up for those who choose to self-host at least some of their AI resources. In fact all the MiniPCs competed will on power consumption and with fiber to the door, what’s the problem?
While the idea of good comes from the remnant of bad is true, there is the small matter of the in-between faze to contend with that happens when the bubble actually bursts, especially when it comes to jobs and people’s basic incomes.
It’s not just jobs inside the AI industry that are going to be highly affected by a bust, but more specifically and mainly those outside it are, by proxy. Your spending is my income, my spending is your income – when the amount of overall spending shrinks heavily in a recession (and this will surely be the largest ever), the amount of immediate harm caused to people is immense, both financially and psychologically.
Unfortunately, this seems to be an inherent feature of a mature global economy’s economic cycle, while mitigating against such things seems to be an anathema or impossible by govts.
If the use of “bubble” in this thread means a stock price/financial asset price bubble, I regard cryptocurrency to be a bigger threat to individuals than Generative AI.
Why? Key reasons include:
Most of the aggressive spending on gAI is by multinational companies with diversified businesses that are not solely reliant on gAI activity.
Except for Meta, excess proprietary data center capacity created by a gAI “crash” can be redeployed to commercial cloud services, such as AWS and Azure.
Crypto products and trading venues are unregulated and opaque. This makes failure cascades (similar situations are the LTCM crash in 1998 and the subprime mortgage crash in 2008) much more likely.
Crypto speculation is mostly driven by individuals, who often have high percentages of their net worth committed to crypto holdings.