Are AI stocks in a bubble? Why this isn’t a dot-com redux

Strength in a handful of mega-cap tech stocks, plus widespread excitement about artificial intelligence broadly, has many investors asking: Is the market in a bubble?

Comparisons to the late-1990s dot-com era are easy to draw. Circular revenue streams, big spending on hardware, and rising valuations all have echoes of the dot-com era.

But we see key differences between today’s environment and the late 1990s:

• AI datacenters are building for real demand, today. Demand for AI compute is growing exponentially¹ with no observed slowdown.
• The AI datacenter build-out is primarily being funded by established tech giants with massive balance sheets and cash flow.
• Valuations, while elevated, are far below dot-com extremes.

Today’s tech valuations are not cheap, but we believe they’re far from bubble territory.

• At the peak of the dot.com bubble, the top four tech leaders of early 2000 (MSFT, CSCO, INTC, ORCL) traded near 70 times 2-year forward earnings.²
• Today, the average 2-year forward Price/Earnings (P/E) is about 26 times for the biggest AI datacenter spenders — Microsoft, Alphabet, Amazon and Meta — also known as the hyperscalers.³

Concentration risk is real, just as it was at the start of the AI rally in early 2023. The so-called Magnificent 7 stocks — Apple, Nvidia, Microsoft, Amazon, Tesla, Alphabet, and Meta Platforms — comprise about 35% of the market cap of the S&P 500.⁴

But concentration may not always equal a bubble. We feel it raises portfolio risk if a few names stumble, but it isn’t proof of mispricing when those names are producing the bulk of the earnings growth.

In 2001, the tech leaders of the dot-com era saw their annual net income drop by 65%, whereas median Wall Street consensus now expects the hyperscalers to grow earnings by 17% over the next year.⁵

Recent reports of deals where a supplier helps finance a customer who then spends money back with that supplier (either directly or via a partner) have raised concerns about the dot.com-era practice known as circular financing. Circular financing can be risky when it hides weak end‑demand. During the dot-com boom, equipment vendors boosted sales to mask a demand gap that later imploded.⁶

In our opinion, though, that’s not the case today. AI usage is already large and growing very quickly. OpenAI’s revenue is running around $13 billion a year⁷, and Anthropic is targeting $9B in 2025 run‑rate with a plan to more than double again in 2026.⁸

Notably, industries with big, long‑lived assets have tended to blend sales and financing. Boeing, for example, has long provided asset‑backed financing so airlines can take delivery of new planes.  Energy and heavy‑equipment vendors typically arrange leases or loans so customers can afford multi‑year projects, with the asset and future revenue as collateral.⁹

The same logic has been showing up in AI: an AI lab needs compute for years, a cloud provider needs to lock in supply, and a chip vendor wants reliable end‑demand. In our view, it’s a classic way to line up customers, financing, and capacity.

AI datacenter capacity is constrained, driving multi-year, prepaid commitments to lock in scarce AI chip and datacenter supply. The demand curve has no clear cap: every leap in capability has been creating appetite for more compute.¹⁰

Most AI-related capital investment to date comes from retained earnings and corporate cash, rather than debt.¹¹ The largest platforms have generated immense operating cash flow and retained substantial net cash. Many haven’t needed to lean heavily on debt to build datacenters.¹²

Yes, there’s debt in the system, mainly around newer, specialist AI clouds and specific large projects. But these look more like project-finance structures: lenders underwrite long-term customer contracts and the underlying hardware, rather than swinging at blue-sky forecasts.¹³

In our view if we see stress, it’s likely to be local, not systemic. Smaller providers that grow ahead of demand, or whose contracts don’t hold, could face refinancing pressure. But they represent a small slice of total capacity, and GPUs are highly redeployable assets if demand shifts.

In short, we believe the credit channels that could transmit a system-wide shock look far narrower today than in 1999 – 2001.

As AI models train, practice, and then think across multi-step tasks, they consume far more compute at every stage. That’s why the industry must build capacity ahead of revenue, and why financing structures that look “circular” are being employed. Meanwhile, debt has been low and at the edges rather than at the system’s core and valuations, while not cheap, are nowhere near dot-com extremes.

Put together, we don’t see recent financing loops or project-level debt as a reason to back away from the AI theme. The spending has been largely powered by profits, the risk looks local rather than systemic, and the demand story has been anchored in real, expanding usage.¹⁴

Still, our optimistic thesis could be challenged by any of the following scenarios:

• Power constraints. If AI builders can’t secure consistent, 24/7 power, then our constructive view may weaken. Interconnection queues and transmission bottlenecks in hubs like Northern Virginia and Texas already suggest power constraints could slow the AI datacenter build-out.¹⁵
• Rising idle capacity or sustained utilization slippage. Evidence that new datacenters aren’t filling, or that booked capacity is being canceled, would challenge the “insatiable” view.
• Capability plateaus. If new training cycles stop delivering meaningful capability gains, we believe the value case for stepping up compute would erode.

None of these are flashing red today, but they’re potential tripwires we’re watching. For the foreseeable future, we continue to believe the AI investment cycle is rooted in genuine, accelerating demand and sustained by the financial strength of today’s leading technology companies.

While vigilance is warranted as the landscape evolves, we feel that current evidence points to a growth story that is fundamentally distinct from past bubbles. Investors willing to remain disciplined and discerning may find the AI theme remains one of the most compelling opportunities of the decade.

Investors interested in AI’s potential may consider:

• The iShares A.I. Innovation and Tech Active ETF (BAI). This actively managed fund selects a concentrated portfolio of global AI and technology equities across all market capitalizations, chosen through bottom-up, research-driven fundamental investing.
• The iShares Future AI & Tech ETF (ARTY), which seeks to provide targeted exposure to the full value chain of companies at the forefront of AI innovation in areas including generative AI, AI data & infrastructure, AI software, and AI services.