GoKawiilAndrew Ross Sorkin; illustration by Lorenzo Gritti
In most episodes of speculative excess, the prevailing mood has been one of denial. From Amsterdam in 1636 to Silicon Valley in 2000, investors have typically insisted they were witnessing not a bubble but a revolution: a new paradigm, the dawn of an era. When challenged on whether prices had become irrational, they reached for the evergreen defense: This time it’s different.
The current boom in artificial intelligence stands apart for its lack of denial. The notion that we are in the frothy, hype-driven phase of technological speculation has become conventional wisdom. Venture capitalists and technologists openly acknowledge that valuations are inflated, expectations are overblown, and vast sums of capital are chasing both promise and illusion. Rather than contesting the bubble’s existence, they embrace it as not only inevitable but perhaps even essential to the breakthroughs ahead. This marks a subtle but significant evolution: where previous bubbles were about believing in the impossible, the current one seems to involve believing in the bubble itself.
The belief that bubbles can be useful—wasteful in the short run but transformative over time—is not entirely new. Among certain economists and historians of technology, it has become the dominant view. In her influential book Technological Revolutions and Financial Capital (2002), the economist Carlota Perez argued that major bubbles often accompany the early stages of general-purpose technologies. Speculative fevers can drive investment far beyond the potential short-term returns, but the overbuilt infrastructure eventually becomes the backbone of entire industries. The British railway boom of the 1840s laid far more track than demand justified but helped create the logistics system for industrial capitalism. The fiber-optic mania of the 1990s left behind excess capacity that later enabled the rise of the Internet.
The dot-com bubble that burst in 2000 is often cited in this light. While many start-ups failed, speculative funding helped build the infrastructure, workforce, and culture that would fuel Amazon, Google, and the digital economy as a whole. The libertarian economist Tyler Cowen, a guru to the Silicon Valley crowd, says that we shouldn’t worry about unprecedented levels of investment in AI because the benefits greatly outweigh the potential harm. “In fact, what we are seeing right now is a shortage in the AI sector’s capacity to meet demand,” Cowen recently wrote.
Major tech companies are investing in more computing capacity, but they still cannot serve all the customers who want access to AI systems. That augurs well for the future of the sector, even if there are dips and spills along the way.
Yet there are growing reasons to doubt whether the AI bubble—if that is in fact what it is—will leave behind anything as enduring as nation-spanning rail tracks or fiber-optic cables. Skeptics including the investor Michael Burry, made famous by The Big Short, and the technologist Paul Kedrosky argue that analogies to earlier industrial bubbles are dangerously misleading. The largest share of capital expenditure in the AI boom is going toward buying Nvidia chips that power large language models. But Nvidia releases dramatically more efficient chips every two years. If and when AI revenue materializes at scale, the data centers will still be standing and their cooling systems should still be working, but the chips that make up the bulk of their cost will be obsolete. That suggests not only that Open AI may soon run out of money, as the economics writer Sebastian Mallaby recently argued in The New York Times, but that the infrastructure being created is more like single-use scaffolding than underground fiber, which waited patiently for future enterprises to light it up and put it to use.
These critiques challenge the soothing assumption that even failed AI ventures will leave behind the building blocks of the future. If the lion’s share of investment is being used to buy hardware that becomes outdated every two years, then we are not in a productive infrastructure bubble akin to 2000. Instead we may be incubating something far more dangerous to the economy: a financial bubble, with no upside to speak of. The rise of circular investment structures and the growing use of off-balance-sheet financing are among the worrisome signals.
If skeptics like Burry and Kedrosky are right, we may come to see this era not as one of creative destruction but as a strange moment of self-aware self-deception. The AI bubble could end up looking less like the dot-com crash of 2000 and more like the asset bubble and systemic financial collapse of 2008—or 1929. That makes understanding past bubbles all the more urgent.
At the center of questions like these, one inevitably finds Andrew Ross Sorkin. While he is best known as a New York Times journalist—and a very good one—that label hardly captures the breadth of his presence in financial discussions. He’s also a coanchor of CNBC ’s Squawk Box and the founder of DealBook, a digital franchise he launched inside the Times in 2001. In addition to its daily newsletter, DealBook hosts an annual summit where Sorkin conducts a full day of interviews with some of the most powerful figures in finance, politics, and technology.
... continue reading