The AI trade isn't a bubble - it's a supply problem

Trillions of dollars have been poured into artificial intelligence over the past several years, on chips, data centers, power facilities and even orbital compute. Spending on that scale on a new development in technology invites an obvious comparison to the 1990s, a heady time that didn’t end well for investors. So, are we in a new tech bubble?

We don't think so. The comparison to the dot-com era is understandable, but it breaks down on close inspection. In 1999 and 2000, companies built the infrastructure first and waited for demand to show up. Today, the opposite is true: demand is already here, at a scale the tech industry has never had to serve, and supply is scrambling to keep pace.

Supply and demand

During the internet boom, companies wired offices and built servers on the assumption that a mass consumer audience would eventually follow. The problem lay in that word “eventually.” By the end of 1999, only about a third of US adults were online, and worldwide, fewer than one in 20 people had internet access. The infrastructure was built for a market that, at the time, barely existed.

AI is running the opposite playbook. The major AI providers already have billions of active users. These aren't speculative sign-ups; they're daily habits, with usage climbing every time a new capability — coding agents, voice, video, agentic tasks — gives people another reason to open the app. Demand isn't waiting for the product to mature. It's outrunning it.

That's precisely why the infrastructure build-out looks so large. Hyperscalers are racing to bring data center capacity online fast enough to serve the demand that already exists today — before accounting for where usage is headed in the years ahead. Capital expenditures from the largest AI firms are expected to top $750bn this year alone, yet demand keeps moving faster than construction schedules allow.

Valuations are contained

The second dot-com parallel people reach for is valuation: stock charts for AI-exposed names have gone nearly vertical, so surely a bubble must be hiding underneath. Memory chipmakers sit at the center of the AI infrastructure boom, and their earnings have followed: profits have risen several hundred percent year over year in some cases. Yet many of these stocks trade at single-digit forward price-to-earnings multiples. These aren't story stocks running on narrative. They have revenue, margins, and profits growing at least as fast as their share prices — the opposite of what a bubble looks like.

IPOs are back, but the market itself is larger than ever

Another feature of the dot-com era was IPO issuance, as a flood of new companies rushed to go public. There has been a real pickup here too, with 2026 US IPO volume approaching levels last seen in 2021. But nominal dollars can be misleading. It’s better to view issuance relative to the size of the market that has to absorb it. Those concerned that the scale of IPO issuance is excessive should consider the scale of the US equity market. The S&P 500's total market capitalization has grown to more than $65trn – much larger than in 2000 – so it is able to absorb much larger dollar amounts of new issuance than were seen in the dot-com era.

Our view

None of this means the AI build-out is risk-free. Capital cycles like this one tend to overshoot in places — some data center capacity will likely get built ahead of eventual need, and some companies riding the AI wave today may not be standing in five years.

But the core argument for a bubble rests on a pattern that simply isn't present this time: infrastructure built years ahead of a demand base that doesn't yet exist, funded by companies with no earnings, at a scale the market can't absorb. What we see instead is demand that's already arrived — over a billion users and growing — with supply struggling to catch up to it, backed by companies posting real profits at valuations that are, in places, historically cheap. On that basis, we see a good fundamental setup for investing in AI-exposed companies over the next three to five years.