The strange thing about this debate is that neither side is arguing over different facts. The bubble camp and the non-bubble camp are looking at the exact same numbers. They just draw completely opposite conclusions from them. Rather than taking a side, this article sets out both arguments side by side like a courtroom proceeding. The final verdict is left to the reader.
Exhibit A: The Hard Data Shared by Both Sides
First, the undisputed numbers. Combined AI capital expenditure (CapEx) forecasts for 2026 across five US Big Tech giants (Amazon, Alphabet, Meta, Microsoft, and Oracle) stand at around $725 billion. Gartner estimates total global AI spending this year at $2.53 trillion. In its Financial Stability Report, the US Federal Reserve listed AI as one of the top systemic risks. Asset manager Man Group described the current debt-funded AI CapEx cycle as being “among the most aggressive in modern business history.”
Up to this point, neither prosecution nor defense has any objections. Where they diverge is the interpretation.
The Prosecution: “This Is a Bubble”
The prosecution rests its case on three pillars.
First, capital is moving in circles. This is the structure critics call “circular financing.” When an AI startup receives funding and immediately buys compute from a cloud provider, that spending is logged as the cloud provider’s “revenue.” That revenue pushes up valuations, and the higher valuations justify further AI investments. As money spins around within the same ecosystem, it looks like organic growth from the outside. The prosecution points out that this closely resembles the “vendor financing” of the dot-com era, when telecom and equipment companies lent money to customers to buy their own gear—inflating demand metrics right up until the collapse. Indeed, when reports surfaced that Nvidia was pursuing a $750 billion infrastructure deal, skeptics warned that contracts of this nature artificially inflate industry-wide demand and valuations.
Second, revenues are not keeping up with spending. One financial analysis calculates that to achieve just a 10% Return on Invested Capital (ROIC) on cumulative AI CapEx so far, companies would need to generate an additional $160 billion in new annual profits solely from AI. To hit the 20% ROIC implied by current earnings multiples, that requirement jumps to $320 billion—on top of existing business profits, strictly from AI. Critics argue that no realistic short-term scenario can deliver numbers like these.
Third, macroeconomic signals are missing. Despite hundreds of billions of dollars poured in since 2022, analyses show little clear evidence that AI has delivered a measurable boost to US GDP growth. The prosecution’s final argument is that this gap between unprecedented capital deployment and delayed macro performance historically coincides with late-stage bubble dynamics.
The Defense: “It’s Not a Bubble, or at Least Not That Simple”
The defense reads the exact same evidence through a different lens.
Rebuttal 1: There is real revenue this time. Even Fed Chair Jerome Powell noted that, unlike past bubbles, today’s AI companies generate actual revenue and produce measurable economic output. Morgan Stanley highlights that corporate cash flows are roughly triple their 1999 levels, providing a much thicker cushion to absorb shocks. The fundamental financial health is completely different from the unprofitable dot-com darlings.
Rebuttal 2: The demand curve hasn’t bent yet. Data center demand is projected to grow by over 19% annually through 2030, and Nvidia envisions global CapEx expanding from $600 billion to as much as $4 trillion. The defense’s logic is straightforward: if demand maintains this pace, today’s spending rush will prove to be premature, not reckless.
Rebuttal 3: The better analogy is the telecom boom, not the dot-com crash. The fiber-optic overinvestment of the 1990s led to a brutal short-term crash, but the infrastructure laid down during that era ultimately underpinned decades of internet expansion. Amazon encapsulates the lesson that picking the right technology isn’t enough on its own: Amazon’s stock plummeted 93% from its 1999 peak before it went on to become one of the world’s most valuable companies. A technology being real and its current price being right are two entirely different questions.
An Interesting Twist: Even Hardline Skeptics Don’t Call AI a Fake
Here lies the subtlety of this case. Even the fiercest bubble proponents do not claim that AI is a fraud. Even Sam Altman admits that “some people are going to lose a vast amount of money,” while remaining completely unshakeable about the future of the technology itself. In short, the real debate is not “Is AI real?” but rather “Have investors pushed this boom too far, and if so, what breaks first?”
The vulnerability lies in what Man Group points to as “reflexive demand.” In a circular structure, if a single major player slows investment, revenue across the entire ecosystem can collapse in tandem. What looks like a robust chain is actually a series of links relying on each other’s spending. As credit tightens and lenders begin demanding proof in revenue, the strength of this chain will be put to the test.
In Lieu of a Verdict
The honest conclusion is closer to “both are true.” AI is a real, transformative technology, and companies are deriving genuine revenue from real customers. At the same time, valuations have outrun cash flows, leverage is accumulating, and circular financing structures remain unresolved. These two statements do not contradict each other.
This is why asking “Is it a bubble or not?” might actually be the wrong question. A far more useful question is this: How much of the infrastructure being built today will survive the funding crunch before the demand curve catches up? The dot-com crash didn’t kill the internet; it simply purged those who bought into it too early and at too high a price. If history repeats itself in AI, that is likely where it will happen.
⚠️ The figures in this article are as of August 2026 and do not constitute investment advice for specific stocks or assets. CapEx and valuation metrics change quarterly, so readers should verify the latest data directly from original sources (Gartner, the Federal Reserve Financial Stability Report, corporate earnings releases, etc.) before making decisions.
Frequently Asked Questions
What numbers do both sides agree on?
Figures such as the combined 2026 AI CapEx forecast of around $725 billion across five US Big Tech giants, and Gartner’s estimate of global AI spending at approximately $2.53 trillion. The disagreement is not over the numbers themselves, but how those numbers are interpreted.
What does the circular financing critique mean?
It refers to a loop where an AI startup receives funding and immediately purchases cloud compute, which the provider then logs as revenue. Critics argue this circular movement of money inflates perceived end-user demand.
Do bubble proponents deny the validity of AI itself?
No. Even the most hardline skeptics do not claim AI is fake. Sam Altman himself acknowledges that some people will lose vast amounts of money while remaining confident in the technology’s future.
Leave a Reply