Everyone says we’re in an AI bubble. CEOs say it. Economists say it. Retail investors say it on social media every day.
But while the public worries about overhyped chatbots and sky-high valuations, the biggest technology firms in the world keep signing bigger deals and spending faster.
What’s happening right now is an expensive arms race dressed up as inevitable progress.
It looks like a bubble because parts of it are a bubble. But it’s not the same species as 1999 or 2008, and treating it like that makes you miss the real risk, or upside.
What makes this “AI bubble” different
The numbers are still hard to digest. In the past year, OpenAI’s valuation has jumped from $90 billion to around $500 billion. Anthropic’s latest funding put it near $180 billion.
Nvidia has proposed investing up to $100 billion in OpenAI. Amazon agreed to a $38 billion cloud contract with OpenAI while still backing Anthropic. Microsoft, Google, Meta and Amazon together are spending more than $400 billion a year on data centers and chips.
The AI market looks inflated by any classic measure, but it’s being built on something unusual. Cash, instead of debt.
During the past two decades, tech giants quietly accumulated the largest corporate savings pile in history. Before 2017, US tax rules encouraged them to park profits in offshore accounts, mainly in Ireland. When the rules changed, they brought it home. By then, Apple, Microsoft, Alphabet and Meta had more cash than many countries.
For years they had nothing large enough to spend it on. Their markets were mature, profits stable, and regulators were watching. Then came AI. Suddenly there was a technology expensive enough and ambitious enough to absorb hundreds of billions. So they opened the vault.
Unlike the dot-com bubble, which ran on stock issuance and retail capital, or the housing bubble, which ran on debt, this one is financed by retained earnings.
This detail makes all the difference, because cash-funded bubbles do not crash the banking system when they burst. They just deflate, leaving a trail of overbuilt infrastructure and expensive lessons.
The circular money loop
The strangest part of this market is how money moves in circles. A lab like OpenAI raises billions from Microsoft or SoftBank. It then spends most of that on cloud contracts with Amazon, Microsoft, or Oracle.
Those companies in turn record the spending as revenue, pushing up their stock prices and valuations. Investors mark up their stakes, which justifies raising again at higher prices.
It is a closed loop of capital that makes everyone look richer and busier than they really are. Oracle’s stock jumped earlier this year after announcing a large data center deal with OpenAI, even though the money originated from another tech company’s investment in that same company.
Nvidia sells chips to Amazon and Oracle, invests back into OpenAI and Anthropic, and books profits from both ends.
On paper, this looks like a thriving, self-reinforcing ecosystem. In reality, it is the same money being passed between the same half-dozen firms. The result is that growth appears organic when it is partly manufactured.
Why smart money keeps buying
If everyone inside the system knows it looks circular, why keep playing?
Because for the companies involved, not participating is riskier than overspending. Microsoft cannot risk letting Amazon become the dominant cloud for AI.
Google cannot risk being the only firm without a flagship frontier model. Nvidia cannot risk slowing demand for its GPUs.
This is a market driven by fear, not greed. It is the fear of losing platform relevance.
In markets built on network effects, missing a generation of technology can lock a company out for a decade. That fear makes firms insensitive to price. They will overpay today to avoid being strategically trapped tomorrow.
From the outside it looks like madness. From the inside it is defensive survival. Each dollar spent is less an investment in profit than an insurance premium against obsolescence.
The deeper truth is that capital markets reward dominance stories more than disciplined accounting. A firm that convinces investors it will own the next technological layer enjoys a higher multiple, cheaper capital, and the freedom to keep spending.
That dynamic creates a reflexive loop where high valuation begets investment capacity, which reinforces dominance, which keeps the valuation high. And smart money understands this game.
Is this time different?
There’s another layer rarely mentioned, and that is government policy.
AI has been recast as national infrastructure. US, EU, China and Japan all want domestic AI capacity and are offering subsidies to make it happen.
The US Chips Act and related energy programs have effectively told tech firms that building data centers and fabs is patriotic work.
That creates a soft floor under the whole system. If demand slows, it is easy to imagine governments reframing unused AI facilities as “critical compute reserves” or research infrastructure.
The message to corporate boards is clear: the downside is limited. That safety net encourages even bolder spending.
When risk is partly socialized, rational managers take bigger swings. This is ultimately how modern industrial policy operates. It also explains why executives can publicly acknowledge that AI valuations are “a bit ahead of reality” while still committing to record capital expenditure.
What happens when the music stops
The danger feels more like a long plateau instead of a steep crash like the 2000’s or even 2022.
AI demand may take years to justify today’s construction boom. If adoption across industries slows or if open-source models become “good enough,” much of the new capacity could sit underused.
That won’t bankrupt Microsoft or Amazon, but it will compress returns and force write-downs.
Narrative is the biggest risk. Current stock prices assume that AI will add trillions in new value within a few years. If that revenue curve flattens, markets will rerate the sector sharply, not because the companies fail, but because the story does.
The second-tier firms will feel it first. Smaller labs without guaranteed cloud customers could face funding droughts as investors retreat to the giants.
The correction, when it comes, will look like fatigue, evident in cancelled projects, delayed data centers, massive layoffs.
The AI bubble, if that’s what we want to call it, is not a collective hallucination. It’s a rational response to a system where cash is abundant, interest rates are still manageable, and missing the next technological wave could mean permanent decline.
This circular flow of money between the giants does distort signals, but it also reflects a rare alignment of interests: everyone wants to own the future infrastructure of intelligence.
Whether that future arrives soon enough to justify $500 billion valuations is a separate question. For now, Big Tech is using old profits to buy optionality.
They are trading cash for relevance. In the process, they are building the most expensive safety blanket in corporate history.
The post How the ‘AI bubble’ really plays out appeared first on Invezz
