Of Course we're in an AI Bubble
The trillion-dollar question isn't whether we're in a bubble anymore. It's what happens when it bursts, and whether that's necessarily a bad thing.
About 40% of US GDP growth in the last 12 months came from AI investment. If that investment suddenly stops, whether because investors lose faith or the economics don't work, the macroeconomic consequences could be severe.
When Pat Gelsinger, former CEO of Intel and a 40-year veteran of Silicon Valley, was asked on CNBC whether we're in an AI bubble, his response was refreshingly blunt: ‘Of course. We're hyped, we're accelerating, we're putting enormous leverage into the system.’
No hedging. No corporate speak. Just a simple admission that the smartest people in technology have been dancing around for months: Yes, we're in a bubble. The question that should keep investors, employees, and policymakers awake at night isn't if there's a bubble, it's what kind of bubble we're in, and what it leaves behind when it inevitably pops.
The Anatomy of Denial
For the past two years, the AI investment frenzy has followed a predictable script. Whenever concerns about unsustainable valuations surface, they're immediately dismissed as pessimism or "not understanding the technology." Critics are told they're missing the forest for the trees, that AI is fundamentally different from past technological manias.
But something fascinating happened in August 2024. Adrian Cox at Deutsche Bank Research Institute noticed a spike in Google searches for "AI bubble", triggered by Sam Altman's admission that investors were "overexcited" and an MIT study showing 95% of AI pilots weren't working out. The concern was real, measurable, and growing. Then it vanished.
By September, searches for "AI bubble" had plummeted to just 20% of their peak. The fear evaporated faster than morning fog in Silicon Valley. Cox's analysis reveals something profound about market psychology: "The only thing we have to fear is a lack of fear itself. It is the lack of fear that can really lead to a bubble.". When everyone stops worrying, that's exactly when you should start.
The Numbers don't Lie…But they're Confusing
Let's start with what even the optimists admit: the valuations are insane.
OpenAI, despite generating "only" $13 billion in annual revenue, believes it needs $1 trillion in investment. That's a 77:1 ratio of required capital to current revenue. During the dot-com bubble, widely considered the most spectacular example of irrational exuberance in modern history, companies rarely presented such lopsided numbers and survived scrutiny.
Jack Selby, who manages $181 billion at Thoma Bravo, does the math: "You cannot value a $50 million ARR company at $10 billion." For that company to merely double an investor's money, it would need to eventually generate $1 billion in annual free cash flow. "That's a tall order managerially, even if the product is right and the market is right." Orlando Bravo, managing partner at the same firm, is even more direct: "Valuations in AI are at a bubble. You cannot value a $50 million ARR company at $10 billion."
Meanwhile, Jared Bernstein, former chair of President Biden's Council of Economic Advisers, published research showing that AI's share of the economy already exceeds internet-related investments during the 1999 peak, by one-third. The IMF's managing director, Christine Lagarde, warns that valuations are "getting close to the level of internet companies just before the dotcom crash." These aren't pessimists or luddites. These are serious people with serious money making serious warnings.
But this Time really is Different…at least sort off!
GeHere's where the debate gets genuinely interesting: the bubble deniers have some compelling arguments.
Real Technology, Real Revenue: Unlike the dot-com era, when companies went public with business plans written on napkins, today's AI investments are backed by the most profitable corporations in human history. Microsoft, Google, Amazon, and Meta aren't burning through venture capital, they're funding AI development from massive free cash flows. ChatGPT generates 5-6 billion visits monthly. That's not speculation; it's measurable demand. AI is already transforming search, advertising, and recommendations that generate hundreds of billions in revenue.
Documented ROI: when Apple removed attribution data from Meta, the company lost hundreds of billions in market cap overnight. Meta deployed AI powered by Nvidia GPUs to solve the problem, and recovered the valuation. That's not a promise of future returns; it's documented, auditable ROI happening right now.
Infrastructure Reality: as Nvidia CEO Jensen Huang argues, this isn't just about new companies. AI is transforming hundreds of billions in existing revenue. The infrastructure transition alone, moving from classical CPUs to AI computing, represents "hundreds of billions of dollars" in necessary investment, regardless of whether OpenAI or Anthropic succeed.
Legitimate Barriers In 1999, you could "start an internet company by hanging a sign out." Today, AI requires billions in infrastructure and specialized talent. Gelsinger's Intel story is instructive: even with vast resources, the company made "bad decisions over 15 years," lost technical leadership, and now needs "five-plus years" and tens of billions to rebuild. These aren't barriers that disappear overnight.
The Tale of Two Bubbles
Jeff Bezos, who watched Amazon's stock crater from $113 to $6 during the dot-com collapse, offers perhaps the most useful framework for understanding what's happening. "We need to define bubble," he says. "If what we mean is like tulips in the Netherlands, that they eventually looked back and said 'What the heck? There was nothing there. Those were just tulips.', no, that's not where we are."
But if you mean it's like the internet bubble, "where in the end something very profound happened, the world was very different, some companies succeeded, but a lot of companies were kind of me-too, fell behind, burning capital companies... absolutely."
Bezos explains that during Amazon's darkest days, when the stock price suggested imminent bankruptcy, every single business metric was improving monthly: customers increased, gross profits grew, operational efficiency improved. "The fundamentals can be disconnected" from stock prices during bubbles. His key insight: bubbles fund both good and bad ideas indiscriminately, but "industrial bubbles" (unlike financial crises like 2008) leave useful infrastructure even when companies fail. The fiber optic companies went bankrupt, "but the fiber optic cable was still there and we got to use it."
This distinction matters enormously. A financial bubble, like the 2008 housing crisis, leaves nothing but wreckage and systemic damage. An industrial bubble might destroy individual companies and wipe out investors, but it can still advance human capability.
The Rational Bubble
Mohamed El-Erian, chief economic advisor at Allianz, introduces perhaps the most intellectually honest framework: the "rational bubble." "A lot of money is being thrown at different AI actors because the promise is so huge," he explains. "However, at the end of the day, there's going to be just a handful of winners, and therefore some investment will result in tears. But overall for the economy, this is good news. It promises significant productivity gains."
This isn't a contradiction. You can simultaneously believe that:
- Current valuations are unsustainable
- Most companies will fail
- Investors will lose enormous sums
- The technology is genuinely transformative
- Society benefits from the infrastructure built along the way
Bill Gates echoes this nuanced view. AI is "the biggest technical thing ever in my lifetime" with "extremely high" economic value. But "you have a frenzy, and there are a ton of these investments that will be dead ends." Some companies will "commit to data centers whose electricity is too expensive" or "buy a generation of chips" that become obsolete before capturing their value.
The Timing Trap
Even if you're absolutely convinced we're in a bubble, timing your exit is nearly impossible, and potentially disastrous. Cox shares a cautionary tale: In November 1998, a California investment manager declared, "This is a serious bubble. This will make the 1991 biotech bubble look like a picnic." He was completely right. He was also 16 months too early. If he'd pulled his money then, he would have missed the Nasdaq climbing from 2000 to 5000, a 150% gain. It took another full year after the peak for the index to fall back to his original exit point.
The Nasdaq had nine separate corrections of 10% or more in the four years leading up to the final crash. Each time, investors faced the same question: Is this it? Usually, the answer was no. Until suddenly, it was yes.
Paul Tudor Jones, the legendary investor, acknowledges this dilemma. He believes it "feels exactly like 1999" and is positioning his portfolio accordingly, for the last phase of a bubble that might run for months or years more. "The Nasdaq doubled between the first week of October 1999 and March 2000," he notes. Missing that final run would be financially devastating.
Goldman Sachs CEO David Solomon captures the tension: "I wouldn't be surprised if in the next 12 to 24 months we see a drawdown. But that shouldn't be surprising given the run we've had."
What Gelsinger sees on the Horizon
Despite admitting we're in a bubble, Gelsinger doesn't see it ending for several years. Why? Because "businesses are yet to really start materially benefiting from it. We have a long way to go.". More intriguing are his comments about breakthrough technologies on the horizon. He's working on systems like "snow cap" that promise "100x better in power performance"—potentially turning a gigawatt data center into a ten-megawatt facility while delivering the same AI capabilities. These "disruptive technologies start materializing in the latter part of the decade."
This timeframe is crucial. If genuine technological breakthroughs arrive in 3-5 years, dramatically reducing costs and expanding capabilities, current investments might prove prescient. If these breakthroughs are delayed or don't materialize, we're looking at a spectacular unraveling.
The Verdict: Yes, and...
After reviewing perspectives from leading investors, economists, and tech CEOs, a surprising consensus emerges:
- Yes, we're in a bubble, even the optimists admit this
- But it has years to run, because real adoption is still early
- Valuations are disconnected from fundamentals, this is obvious
- The technology is genuinely transformative, this is also obvious
- Most companies will fail, but a handful will reshape everything
- Investors will lose enormous sums, but society might benefit anyway
- Timing the exit is impossible, which makes this terrifying
The debate isn't really about whether we're in a bubble. We are, but bubbles aren't always bad. Sometimes they're how the future arrives, messy, expensive, and punctuated by spectacular failures that nonetheless leave behind the scaffolding of transformation.
The music is still playing. The question is whether you're brave enough to keep dancing, or wise enough to head for the exits. Just know this: whichever choice you make, you're going to question it.
