The AI bubble has burst. Here's how we know.

The AI hype we've seen since ChatGPT exploded on the scene echoes previous tech bubbles, all the way back to the dotcom era.

When you live in tech bubble central, signs of a tech bubble become easier to spot every time. Drive to Silicon Valley on any of the Bay Area's main arteries right now, and you'll notice nearly every billboard pumping a product "driven by AI."

On the same drive five years ago, you'd see the same scene with the word "blockchain." Ten years ago: "big data." Twenty-five years ago: literally any word followed by ".com." Each one in turn, for all its promise, became a punchline.

It's not a question of whether the Silicon Valley machine was wrong on any of these technologies. Especially not the dotcom thing. Heck, the entire internet had just dropped into our laps in the 1990s; you can't blame anyone for dreaming about creating all the stuff we now take for granted. It's a question of impatience: all the investors, startup shysters and panicked CEOs that rush in when a promising new technology emerges are eager for immediate results.

Then reality dawns, as it has evidently done for AI investors this summer. Amid a general stock market downturn Monday driven by the worst one-day crash the Japanese stock market has seen in decades, AI stocks like Nvidia — which delayed its next-gen AI chips, and has reportedly been scraping "a human lifetime" of videos every day for an unreleased AI product despite internal concerns — are taking a beating. Nvidia stock has lost a trillion dollars of valuation, 30% of the total, since its 2024 high.

If history is any guide, and it usually is, there's no coming back from a tipping point like this. The next part will be really painful for a lot of people — and yet more beneficial for longterm tech progress than the bubble mentality could ever be.

How the AI bubble burst

What accounts for the AI vibe shift?

Take your pick. Maybe it was a study revealing that consumers are so turned off by the term "AI," they are less likely to buy a product that uses it.

Maybe it was Sam Altman and his weird unforced error with Scarlett Johansson, or Elon Musk trying to sell a flawed AI chatbot while pushing AI deepfakes.

Maybe it was one of the largest and oldest hedge funds in the world telling clients that AI products are "never going to be cost-efficient, never going to actually work right, will take up too much energy, or will prove to be untrustworthy."

Or maybe it was a Goldman Sachs report on generative AI titled "Too Much Spend, Too Little Benefit?" (That minces words just about as much as one of the best-titled academic papers of 2024, an inquiry into the growing phenomenon of AI hallucination: "ChatGPT is bullshit.")

Tweet may have been deleted

Whatever it was, the signs of panic are everywhere. Tech stocks are getting hammered no matter if they're planning to increase their expenditure — Mark Zuckerberg announced $5 billion in "aggressive" new AI spending on his earnings call last week, then Meta stock dropped as much as 15% the next day — or decrease it to make up for all their AI investments — as was the case with Intel, where $10 billion in cost-cutting tanked the stock 25%.

"Generative AI itself won’t disappear," wrote Gary Marcus, a noted AI skeptic, on his blog over the weekend. "But investors may well stop forking out money at the rates they have, enthusiasm may diminish, and a lot of people may lose their shirts. Companies that are currently valued at billions of dollars may fold, or stripped for parts."

Marcus had previously written that the crash would happen in 2025, but now believes it has arrived months ahead of schedule.

Rather than find revenue on their own, AI startups are suddenly keen to be gobbled up by one of the big fish. A chatbot maker called Character.AI, founded by a couple of AI enthusiasts who left Google because they were frustrated by the bureaucracy, just licensed its product to Google — who will fold the founders and main researchers back into its bureaucracy. In recent weeks, Amazon ran the same acqui-hire playbook on an AI startup called Adept, while Microsoft did it to Inflection.

Could it be any more clear that we've reached an inflection point?

AI's biggest hallucination is being solved

Clearly, there is some promise in generative AI. It's just that much of that promise, as we've said before, lies in the small and boring business tasks: saving a bit of time on code development here, writing the first draft of a document there.

The vision of vast, exponential growth in AI tech, to the point where it is an imminent threat to humanity, should not trouble us much longer. Sam Altman's OpenAI has spent the last two years hyping up this threat; increasingly, Altman is seen as the boy who cried wolf. Large Language Models like ChatGPT have run out of stuff to train on, and the more they are trained on "the internet," the more the internet contains a body of work written by AI — degrading the product in question.

Our excitement about AI art has also popped like a bubble. Everyone has access to the tools — including your family members on Facebook — therefore no AI art is special anymore. Indeed, we're more alert than ever to the shady theft and energy drain involved in making this stuff. AI-generated video is even worse on both counts, and has a harder problem leaving the uncanny valley.

Back in Silicon Valley, what this means is that AI companies and products must have something special to survive the coming correction. Look at the startups that were still standing after the dotcom collapse in 2000 and you'll see some familiar names, like Google and Netflix.

Only if it happens slowly, over many years, without so much venture capital investment creating so much froth, will we finally arrive in the AI future today's startups can only dream about.

This column reflects the opinion of the author.