Here's a number that should make you rethink the "AI bubble" narrative: Hyperscaler capital expenditure started 2026 at $515 billion. Six months later, it's now $740 billion. And 2027 is already penciled in at $889 billion.
That's $225 billion in upward revisions in half a year. To put that in perspective, AI infrastructure spending is now 2.7% of U.S. GDP. In 2019, that number was 0.3%. This is not incremental investment. This is a structural shift in where capital is flowing.
And here's the part that's really messing with the bubble thesis: Earnings are actually following the spending.
In the dot-com bubble, that didn't happen. Companies poured billions into infrastructure, and profits lagged by a mile. The spending was speculative. The returns were hypothetical. That's how bubbles work.
But tech earnings estimates for 2027 are up nearly 20% year-to-date. The hyperscalers aren't just spending on AI infrastructure—they're making money from it. Microsoft, Amazon, Google, and Meta are reporting actual revenue growth tied to AI products and services. The spending isn't outpacing the returns. At least not yet.
That changes the equation entirely. If you're spending $740 billion and earnings are rising to match, that's not a bubble. That's a boom.
The other thing nobody's talking about: The capex doesn't stay in Silicon Valley. It flows directly to wherever the hardware gets built. Chip manufacturing, data center construction, energy infrastructure—all of that is happening globally, and the equity returns in those markets are jaw-dropping this year.
Semiconductor pricing, which has been deflationary for decades, is spiking. If oversupply hits, that's your canary in the coal mine. But right now, demand is outstripping supply, and the companies building the picks and shovels for the AI gold rush are printing money.
So is this a bubble or not?
Honestly, I don't know. The scale of spending is unprecedented. The speed of the revisions is alarming. And any time something moves this fast, you should be skeptical.
But here's what I do know: This is not how bubbles typically behave. Bubbles are characterized by spending without earnings, hype without substance, and capital flowing to projects that will never generate returns. The dot-com bubble was full of companies with no business models. The housing bubble was full of loans that couldn't be repaid.
AI spending right now is going to companies that are already profitable and building infrastructure that's already generating revenue. That doesn't mean it's risk-free. It doesn't mean valuations are reasonable. But it does mean the comparison to 1999 is lazy.
The real question is what happens next. If earnings keep pace with spending, this is sustainable. If they don't, we're looking at a massive correction. The signal to watch is semiconductor pricing and utilization rates. If data centers start sitting empty or chip prices collapse, that's when the bubble thesis becomes real.
Until then, the narrative that "AI is just hype" is getting harder to defend. The spending is real. The earnings are real. And the revisions keep going up.
Maybe that makes it scarier, not safer. When something is working this well, it's hard to know when to get nervous.
