Is the AI Supercycle Still Underpriced by the Market?

The past two weeks ended on Friday, February 13 have seen investors turn against sectors seen as in danger of the artificial-intelligence disruption. But, as Martin Baccardax of Barron’s notes, investor concerns that the biggest players in the AI investment boom had committed too much capital was also a driver of the selloff.

Interestingly, it doesn’t appear that the argument against the AI trade is that the technology is fake. What is clear is that the market feels that the bill has become too large to justify.

It is easy to understand this skepticism, especially considering how long the AI trade has kept markets in a seemingly unending bull trend. The truth is that the AI supercycle has matured from an abstract “future-of-work” narrative into a cash-flow reality. It is a reality where the biggest companies on earth are pouring unprecedented sums into servers, data centers, and networking gear. Alphabet, for example, just told investors it expects 2026 capital expenditure of $175 billion to $185 billion, which is nearly double its 2025 level, explicitly to expand AI computing capacity and relieve “compute constraints.”

But the market’s anxiety can also be a signal. To be sure, the AI buildout is gobbling up lots of liquidity. And when investors fixate on this cost side, they can miss what’s being built, and how long it can keep compounding once it becomes infrastructure.

So, is the AI supercycle still underpriced? The task ahead in this week’s newsletter is to try to answer this question.

For starters, the AI supercycle is underpriced, but in parts. And this is not because artificial-intelligence stocks are “cheap,” but because the market may still be underestimating the duration of this capex cycle and the breadth of businesses it will reshape.

The first clue is that Big Tech is behaving like it’s in a wartime economy. In fact, Barron’s recently framed this as a spending spree that could reach roughly $650 billion this year after more than $400 billion in 2025. And this is only across four tech giants. Plainly stated, we are no longer debating whether the hyperscalers will invest because they are, and they’re doing it at a scale that changes what their financial statements look like.

And that matters because markets have muscle memory. For a decade or thereabout, Big Tech was rewarded for being asset-light. They boasted high margins, swelling free cash flow, and reliable buybacks. Therefore, a capex shock disrupts that rhythm, even if it lays the foundation for the next era.

Expert observers observe that a lot of public-market pricing still treats artificial intelligence like a product cycle. This is an issue because what is happening is an infrastructure cycle. And this is where the underpricing case starts.

Take, for instance, Alphabet’s guidance. The company acknowledges that it expects capacity constraints to persist, but, at the same time, it describes an aggressive ramp to expand data centers and AI compute capacity. Alphabet executives wouldn’t speak like this if their buildout were a one-time refresh.

The second clue is demand diffusion. A year ago, many enterprises were still testing artificial-intelligence products. Now, they are describing AI agents as widespread inside large companies.

According to a Wall Street Journal report from February 12, AI agents are “now widely deployed in large enterprises.” The Journal cited real-world deployments from a number of firms, including Principal Financial Group, Sonic Automotive, and Bank of New York Mellon. That is a meaningful shift because now, the artificial-intelligence story is no longer about model releases and will become more about workflow redesign.

And if that diffusion continues, it’s believable that the market is still underpricing the second-order effects. That includes the companies that sell power equipment, cooling, networking, memory, and the software plumbing that makes enterprise deployment safe and repeatable.

If there was one thing that came out clear as day from the recent earnings releases by hyperscalers, it is that cash flow and electricity are the two hard constraints of the current AI buildout. Which explains why the market isn’t crazy to hesitate.

On cash flow, Barron’s argues the capex surge could limit buybacks and dividends. And that the spending is set to show up as rising depreciation expense over several years. This is an issue because even if revenue growth is solid, the path from “spend” to “returns” can be messy and uneven, especially if multiple hyperscalers overbuild at the same time.

The electricity constraint is much more physical. According to Goldman Sachs Research, global data center power demand could rise 50% by 2027 and by as much as 165% by 2030. This is compared to 2023 levels. And incremental data center supply has been constrained by utility transmission limits, permitting delays, and supply chain bottlenecks, said Goldman. The bank estimates that about $720 billion of grid spending through 2030 may be needed.

This is the part of the AI story that complicates the “underpriced vs. overpriced” debate. Simply put, even if demand is there, power and permitting can delay timelines, shift costs, and create regional winners and losers.

The short answer is that it depends.

From the perspective of infrastructure enablers, the markets appear to be pricing the near-term pain more aggressively than the long-duration nature of the buildout and the broadening adoption of AI agents inside enterprises. And in this sense, the supercycle is underpriced.

But this underpricing is uneven. In fact, some believe that the AI trade will fragment as investors differentiate between companies with clear monetization paths and those merely burning cash.