Is there an AI bubble — and is it about to pop?

How much is the future worth? Usually to answer that question, you’d need to ask philosophers or economists. But if you’re a tech CEO, you have an actual number: about $1 trillion. That’s how much the tech industry as a whole is set to spend building out the artificial intelligence industry over the coming years. And even in Silicon Valley, where several companies have market capitalizations that start with “T,” a trillion dollars is a lot of money. And while you won’t find more fervent evangelists for AI anywhere than in the C-suite of companies like Google and Microsoft, eventually, all that money has to be recouped. The alternative would be an economic meltdown of the sort we haven’t experienced for years. Which may just well be in the process of happening. On Monday, the stock market continued days of heavy losses, with the S&P down 3 percent by the close of day. The blood-letting was led by many of the same tech companies that had driven the market to record highs in recent months, with AI chip maker Nvidia falling by nearly 7 percent, and Amazon dropping 4 percent. There are a lot of reasons why the bottom has at least temporarily fallen out of the market, including the possibility that the Federal Reserve has been too slow to cut interest rates in the face of a weakening US economy. Recent data around US hiring and manufacturing activity have come in weaker than expected, helping to fuel this sell-off. But there are real concerns that despite the hundreds of billions that have been spent so far to build up the AI industry, and the hundreds of billions that are projected to be spent in the years to come, AI companies themselves aren’t yet producing much in the way of economic value. And they may not for the foreseeable future. That sound you hear could be an AI investment bubble going pop. Big AI dreams No one spends a trillion dollars on something unless they really, really believe in it — and Silicon Valley really, really believes in the transformative economic potential of AI. Way back in 2018, when ChatGPT was just a twinkle in the eye of OpenAI’s Sam Altman, Google CEO Sundar Pichai famously told Kara Swisher that “AI is one of the most important things humanity is working on. It’s more profound than, I don’t know, electricity or fire.” Fire, I think we can all agree, is pretty important. You might even think of it as humanity’s first breakthrough product. But to tech leaders like Pichai, the possibility of effective, general artificial intelligence was every bit as revolutionary as the day when one of our Paleolithic ancestors rubbed two sticks together. And once OpenAI released ChatGPT in November 2022, exposing the world to the genuine magic that is large language models (LLMs), the race was on to become the company that could capture that fire. So investors rushed to fund promising LLM startups like OpenAI (currently valued at $80 billion or more) and Anthropic (estimated at $18.4 billion). In the US alone, AI startups raised $23 billion in capital in 2023, and more than 200 such companies around the world are unicorns — meaning they’re valued at $1 billion or more. All that money is in part a measure of tech’s confidence that the AI market will eventually prove titanically huge. One forecast by the consultancy PwC estimated that AI could add nearly $16 trillion — there’s that word again — to the global economy by 2030, chiefly from vastly enhanced labor productivity. Add in the fact that tech giants have plenty of cash on hand, and are actively racing against each other to be first past the post when it comes to AI. If you believe the AI industry will be worth trillions — and that the lion’s share of that value will go to the early leaders — then as Pichai said on a recent earnings call, “the risk of underinvesting is dramatically greater than the risk of overinvesting.” But the bill is rising, because generative AI is not cheap — both to build and to run. When the bill comes due Sam Altman himself has said that OpenAI is “the most capital intensive startup in history.” That’s because as models get bigger and bigger, they cost more and more to train. And that’s just the cost of making the models — running them is highly expensive as well. One analysis last year estimated that it cost OpenAI $700,000 a day to run ChatGPT, chiefly in all that compute-intensive server time. And the more ChatGPT and other LLMs are used, the higher those costs rise. While Silicon Valley may not have invented the saying “you have to spend money to make money,” it certainly lives by it. But the revenue these companies are bringing in, chiefly via subscriptions to their premium models, are just a fraction of their costs. The Information reported recently that OpenAI could lose as much as $5 billion this year, close to 10 times what it lost in 2022. That’s not a good trajectory, and neither is ChatGPT’s user numbers. Tech analyst Benedict Evans wrote recently that while many people and companies try out AI services like ChatGPT, far fewer stick with it. (Notably, ChatGPT’s usage seems to meaningfully dip during school holidays, just in case you were wondering who the power users were.) While what LLMs can do is impressive, especially compared to what seemed possible a decade ago, promises of artificial general intelligence that could replace whole classes of workers have yet to come true. As it stands now, the industry as a whole seems to suffer from a classic Silicon Valley problem: It lacks product-market fit. Chatbots aren’t yet a true product, and it’s not clear yet how big the market is for them. That’s why experts ranging from Wall Street banks like Goldman Sachs to tech VCs like Sequoia Capital have been throwing up yellow caution flags around the AI industry — and why investors seem to have begun heeding them. None of this is to say that AI itself doesn’t still have revolutionary potential, nor that the industry won’t eventually fulfill those dreams. The dot com crash in the early 2000s was in part due to overinvestment and overvaluation of startups of the era, but Evans notes that what was left over set the stage for mega-companies of today like Google and Meta. The same may one day be true for AI companies. But unless the financials improve, it may not be these AI companies. This story originally appeared in Today, Explained, Vox’s flagship daily newsletter. Sign up here for future editions.