The AI Bubble Isn't an Accident — It's a Strategy

Let's start with the scale. In 2023, the "big four" AI spenders — Amazon, Microsoft, Alphabet, and Meta — collectively spent roughly $150 billion on AI-related capital expenditures. By 2025, that number had jumped to $350 billion.

For 2026, the projected combined capex of these four companies is $650 billion to $725 billion. Amazon alone is spending $200 billion. Microsoft is at $190 billion. Alphabet is at $180-190 billion. Meta is at $125-145 billion. Combined, the four are now on track for $713 billion.

That's 77% growth year-over-year. It's roughly 1% of global GDP. It's larger than the GDP of most countries.

By any historical measure, these numbers look like a bubble in the making.

But that's only if you stop at the spending side.

The revenue engine that changes the math

The difference between today's AI spending and the dot-com era is this: the companies doing the spending are also the companies that own the infrastructure the spending is building.

When Microsoft spends billions on AI infrastructure, that infrastructure becomes Azure capacity. When Amazon spends, it becomes AWS capacity. When Google spends, it becomes Google Cloud capacity. Every dollar of AI capex is not just an expense — it's an asset that generates revenue.

The numbers are already visible. Microsoft's AI business is now running at an annualized revenue rate of $37 billion, up 123% year-over-year. Google Cloud revenue surged 63% year-over-year to $20 billion in Q1 2026. AWS posted its fastest growth rate in 15 quarters at 28%.

Microsoft's contracted backlog across Azure and its cloud business is now $627 billion. Amazon's cloud backlog is $464 billion. Google's is $460 billion. These are not speculative future revenues. They are signed contracts with enterprise customers who have already committed to paying for the cloud capacity these companies are building.

The math is straightforward: every dollar of AI capex generates a multiplier in future cloud revenue. The hyperscalers aren't just spending on AI. They're spending on the infrastructure that AI runs on — and they're the ones who will be paid when that infrastructure gets used.

The closed loop that Wall Street hasn't fully priced

There's a second layer to this strategy that makes it even more clever.

Amazon, Microsoft, and Google have created a closed financial loop: they invest billions in AI labs like OpenAI and Anthropic — and then require those labs to spend that money back on their cloud services.

Amazon expanded its Anthropic commitment to $13 billion, with Anthropic reportedly committing to spend at least $100 billion on AWS over ten years. Microsoft's total OpenAI investment has been estimated at over $100 billion when Azure infrastructure is counted — backed by a reported $250 billion Azure services commitment from OpenAI.

As one industry observer put it: "OpenAI and Anthropic are not rivals — they are financially bound tenants to the same hyperscaler landlords."

The result is a system where the hyperscalers' AI spending isn't really "spending." It's capital recycling. They invest in AI labs, the labs use the money to buy cloud services from the hyperscalers, and the hyperscalers report that as cloud revenue. The money never really leaves the ecosystem.

The Meta exception — and the Meta solution

Meta is the one company in the big four that doesn't have a cloud business to absorb its AI spending. It doesn't have a $627 billion cloud backlog. It doesn't have AI labs paying it back for compute. Every dollar Meta spends on AI infrastructure is a pure cost.

And Meta is spending aggressively — $125-145 billion in 2026. That's more than many countries spend on their entire military budgets.

So what does Meta do? It's launching Meta Compute — a cloud infrastructure business that will sell excess AI compute and model access to external customers.

The strategic logic is transparent. CEO Mark Zuckerberg had already telegraphed the move: "Almost every week, there are external companies reaching out to us, hoping we launch an API service, or asking if they can purchase our compute — even willing to pay above our procurement cost."

When the news broke, Meta's stock jumped 8.8%. Competitors like CoreWeave and Nebius plummeted 14-17%.

Why? Because Meta had just solved its "no cloud" problem. It turned its AI spending from a pure cost center into a potential revenue stream. Goldman Sachs noted that Meta is now being seen as the first hyperscaler to demonstrate financial discipline in the AI spending race.

The prisoner's dilemma that keeps the bubble inflated

The most revealing part of this entire dynamic is why the spending continues even when everyone knows it's excessive.

This is a classic prisoner's dilemma. The four hyperscalers all know that unchecked AI spending is unsustainable. But none of them can afford to stop first. If Microsoft cuts spending and Amazon doesn't, Amazon gets the competitive advantage. If Google slows down and Meta accelerates, Meta catches up.

As one analysis put it: "This is not a rational investment decision. This is a classic prisoner's dilemma. All four giants know this spending is unhealthy. But none dares to stop first."

The result is an arms race where every participant is locked in — not because they think the spending is justified, but because the cost of stopping is higher than the cost of continuing.

What the "bubble" actually is

So when analysts ask whether AI is a bubble, they're asking the wrong question.

The hyperscalers aren't accidentally inflating a bubble. They are strategically building an infrastructure that, once built, becomes the foundation of the next decade of computing — and they are structuring their spending so that it generates revenue before the infrastructure is even fully operational.

The cloud backlogs are real. The AI revenue is real. The contracted commitments are real.

The "bubble" isn't a mistake. It's a deliberate, calculated strategy to own the infrastructure that every other company will need to rent.

When Amazon spends $200 billion on AI infrastructure, it's not hoping to get that money back from AI products. It's ensuring that when your company needs to run an AI workload, the only place you can afford to run it is AWS.

That's not a bubble. That's a moat.