Silicon Valley leaders argue AI's current investment bubble is beneficial despite risks, comparing it to historical bubbles like railroads and dot-com that accelerated technological progress. They acknowledge potential economic crashes and job losses but believe the long-term innovation justifies the short-term financial pain.
Key Takeaways
•Tech billionaires and investors view the AI bubble as 'good' because it accelerates innovation, similar to historical bubbles like railroads and dot-com that ultimately benefited society.
•Massive AI investment ($650B planned this year) has already driven rapid progress in AI capabilities, talent acquisition, and startup formation, but risks overbuilding and economic collapse.
•Bubble defenders acknowledge downsides including potential catastrophic wealth loss ($35T estimate) and recession, but argue technological advancement outweighs financial volatility.
•The 'good bubble' narrative justifies continued high-risk investment in unprofitable AI companies, potentially increasing financial system exposure and retirement account risks.
•AI's transformative potential and bubble risks coexist - whether the bubble is 'good' depends largely on one's financial position and ability to withstand potential crashes.
An AI crash could bring down the economy. Some in the tech world think that's the price of progress. llustration by Alisa Gao / The Atlantic The tech billionaire Hemant Taneja admits that AI is a bubble. In fact, he welcomes it: “Bubbles are good,” Taneja, the CEO of General Catalyst, a venture-capital firm, told me in an email. If AI comes crashing down, it will lead to “some spectacular failures,” he said—companies will go under and people will lose their jobs—but that’s a price worth paying for “enduring companies that change the world forever.”
His view is widespread in Silicon Valley. Some, such as Nvidia CEO Jensen Huang, reject the notion that their companies are overvalued. But many of the wealthiest and most powerful people in tech are embracing the idea of an AI bubble. Jeff Bezos has argued that AI might be a “good” kind of bubble. Sam Altman has made similar comments, predicting that AI will be a “huge net win for the economy” even if “a phenomenal amount of money” is lost along the way.
Indeed, a phenomenal amount of money is at stake: OpenAI, which is still far from profitable, is currently worth more than Toyota, Coca-Cola, and Disney combined. This year, Big Tech plans to spend some $650 billion on the AI build-out—a sum that far exceeds the GDP of most countries. Investors are banking that AI will spur a productivity boom and deliver unimaginable corporate profits, but that future could be far off. If the spending dries up first, the bubble could pop—perhaps dragging the rest of the economy down with it. Nonetheless, Silicon Valley thinks that the present mania will eventually pay back its returns through scientific discovery and economic growth. “Stop trying to make bubbles go away,” as the entrepreneur James Thomason recently wrote. “The benefits of innovation outweigh the costs of volatility.” In other words: Be grateful for the bubble.
Silicon Valley did not invent the idea that bubbles can be worth the pain. Various economists have made the argument for decades. But as the AI boom has exploded, a book by two investors, Tobias Huber and Byrne Hobart, has helped formalize tech’s pro-bubble ideology. Boom: Bubbles and the End of Stagnation was a hit in Silicon Valley when it came out in 2024, praised by the tech billionaires Peter Thiel and Marc Andreessen.
The authors argue that there are essentially two kinds of bubbles: good ones (dot-com, the railroads) and bad ones (the 2008 housing crisis). Both cause damage when they burst, but the good bubbles accelerate the development of new technologies, which ultimately benefits society as a whole. In a bubble, a “set of investments that you could never underwrite otherwise suddenly makes sense,” Hobart told me.
Bubble defenders such as Hobart point to the railroads as one example of how exuberant speculation can end up paying off. They acknowledge that the development of the railroads in the late 19th century led to multiple devastating depressions—but they also point out that the country got, well, railroads that transformed the fabric of American life. The United States “has some of the world’s best freight rail infrastructure thanks to what in the 19th century was excess capacity,” Hobart and Huber write. (Commercial rail travel in the U.S. is another story.) They also look to the early days of the internet, when overzealous investing resulted in the dot-com crash. Yes, it was bad when the bubble burst, but the froth also financed a massive build-out of fiber-optic cables that helped shape today’s internet. Without a bubble, the thinking goes, the modern web would have developed much more slowly.
Even people outside the tech industry seem convinced by the idea that bubbles can have positive elements. “If investors remained dispassionate,” Howard Marks, the billionaire investor who famously anticipated the dot-com crash, told me, “it would take a lot longer for a new unproven technology to be adopted.” Of course, this idea is premised on the notion that widespread adoption is in the public’s best interest.
Either way, though, bubble defenders see the same thing happening with AI: Conscious machines might sound mythical, but if excited investors throw enough cash at the problem—giving entrepreneurs the space to pursue risky, experimental work—superintelligence just might become reality. “There is both froth in parts of the AI ecosystem and real breakthroughs,” as the investment firm KKR wrote last fall. “Past overbuilds in rail, electrification, and fiber seeded critical economic change.” Even Mary Daly, the president of the San Francisco Fed, has suggested that AI is a “good bubble,” noting that “even if the investors don’t get all the returns that the early enthusiasts think when they invest, it doesn’t leave us with nothing.”
Indeed, the technology has already advanced significantly since the arrival of ChatGPT—thanks, in large part, to the spending frenzy. More investment has meant more computing power to throw at training AI models, which, in turn, has led to more capable AI systems. The mania has also sucked talent into the industry and birthed an explosion of start-ups experimenting with new approaches to building the technology. Without such intense investment, it’s hard to imagine so much progress over such a short period.
Less clear is whether the current AI-infrastructure build-out will prove fruitful in the long run. As Silicon Valley continues to pour unfathomable sums into data centers, there’s a risk they will overbuild. Unlike railroad tracks and fiber-optic cables, which can last for decades, computer chips, which power data centers, quickly become obsolete. Still, some bubble defenders argue that all this construction will have lasting value. For example, AI’s seemingly limitless appetite for electricity could also spur a boom in clean-energy generation, as the tech analyst Ben Thompson has written, bringing new sources of nuclear and solar energy online. This, of course, is an optimistic vision: Right now, data centers are driving a gas boom.
Even if Silicon Valley is correct that the bubble is accelerating AI progress, that doesn’t make it unilaterally appealing. “The investor doesn’t say, ‘Well, yes, I lost my money, but thank God it advantaged society,’” Marks said. Accepting short-term financial pain as the cost of technological progress might be easy for tech titans with truckloads of money. It’s a much harder sell to the rest of America. Who cares about better chatbots if you’re about to retire and a crash wipes out your 401(k)?
The freight-rail system might seem great from today’s vantage point, but the Panic of 1893 was among the most severe financial crises in our nation’s history, causing unemployment to spike to more than 10 percent for half a decade. The situation was so dire that J. P. Morgan—who himself was enriched by the railroads—helped bail out the federal government. After the dot-com bubble burst, the U.S. entered a recession. If the AI bubble were to collapse, the fallout could be “catastrophic,” Carlota Perez, the author of a seminal book on bubbles and innovation, told me. The flood of investment is the eye “of a much larger hurricane that involves the whole financial world,” she said. According to one estimate, an AI crash could wipe out some $35 trillion in global wealth.
Inside of tech, many bubble apologists acknowledge the downsides. “There will be people who will have just really unfortunate outcomes from this,” Hobart said about a potential crash. Still, the industry’s mindset seems to be that innovation is worth whatever costs are incurred along the way. If Meta ends up “misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously,” Zuckerberg said last fall. “But what I’d say is I actually think the risk is higher on the other side.”
What makes the narrative of a “good bubble” concerning is that it provides justification for investors to keep pumping money into AI, regardless of whether it really makes sense to do so. As the cash keeps flowing, the risk of a debilitating crash seems to only be increasing. Both Anthropic and OpenAI are racing to go public, reportedly as soon as this year. Such high-status public offerings could ratchet up the mania, and increase the potential for financial contagion, as more people’s retirement accounts and investment portfolios get tied up in still-unprofitable AI companies.
Two things can be true at once: AI is a generational technology that will transform the world, and people are going to lose large amounts of money along the way. A bubble is good only if you’re the one who wins.
