Has the AI Boom Caused a Bubble in U.S. Stock Markets?

We’ve looked at the AI boom from a few angles already — from how hedge funds are actually using GenAI (and what that means for manager selection), to how crowding in the “Magnificent Seven” became its own risk factor.

This piece tackles the question many allocators are now asking more bluntly: Has the AI theme inflated a bubble in U.S. stocks?

Our answer is "probably yes" — but also "yes, it can keep going" (and possibly overheat) into 2026.

That “two-handed” view isn’t a cop-out. It’s the practical stance you need when (1) valuations look stretched, (2) fundamentals are unusually real (cash-flow-rich mega caps funding capex), and (3) macro/fiscal tailwinds can keep the music playing longer than valuation purists expect.

What makes this feel like a bubble (even if the fundamentals are real)

Bubbles aren’t defined by “no fundamentals.” They’re defined by price becoming increasingly dependent on narrative, positioning, and marginal liquidity — with a shrinking set of outcomes that justify today’s valuations. Valuation extremes are showing up in the cross-section, not just index multiples

U.S. stock market valuations are indeed extended in the Tech sector, but they are above historical averages across almost all sectors:

Chart 1: S&P 500 forward P/E per sector; source: MFS "Valuation: Why it Still Matters..."

One clean way to see “AI-bubble behavior” is to ignore the index headline P/E and look at how much of the market has migrated into extreme valuation buckets.

Chart 2: Percentage of S&P 500 stocks with a Price/Sales ratio > 10; source: GMO, "Artificially Inflated"

The chart above shows the percentage of stocks trading above 10x sales in the S&P 500, (market-cap-weighted). It has reached an all-time high.  This is a tell that the index is being driven by a narrow set of very expensive winners.

Even if you’re not “an AI investor,” you may already be one via index concentration:

Chart 3: Top 5 holdings of S&P 500 as % of total market capitalization; source: Invesco, "Beyond the AI bubble narrative"

A simple way to describe the current setup: the index is increasingly a bet on a handful of mega caps. That’s fine when they’re beating numbers. It’s dangerous when (not if) the cycle turns, regulation bites, or capex payoffs get questioned.

This concentration (together with above-average valuations of the other sectors) has also impacted the whole market's valuation, implying negative (real or nominal) returns for the next 7-12 years for U.S. large caps:

Chart 4: Expected 7-year real returns for various asset classes, including U.S. large caps; source: GMO, 4Q-2025 letter

Given the current S&P 500 valuations using a metric similar to the Buffet one (market cap / GDP) and assuming no continuation of extraordinary fiscal stimulus, the 12-year nominal expected returns of the index look similarly dismal:

Chart 5: Expected 12-year S&P 500 nominal returns; source: Hussman Strategic Advisors, January-2026

“Not dot-com crazy”

Still, it’s comforting to say “this isn’t 2000.” And on some simple measures, it’s not.

Chart 6: Forward U.S. Tech sector P/E is only half of its 2000 peak; source: Invesco, "Beyond the AI bubble narrative"

The above chart shows that the U.S. Tech sector P/E is only half as "insane" as during the dot-com bubble, but on the other hand it remains elevated at 33x and above average.

Similar may be said about market activity: the grind up has been steady with no euphoria to match the late 1990s exuberance:

Chart 7: Rolling 12-month returns of the Nasdaq-100 index show no dot.com euphoria levels; source: Invesco, "Beyond the AI bubble narrative"

So, if the current U.S. market is a bubble, then it is a rational and well-behaved one.

Why the bubble may not burst soon

If this were purely “speculation with no cash flows,” you’d expect a cleaner, sooner unwind. But today’s market has a set of shock absorbers.

1) The capex boom is huge — yet still “plausible” as a share of GDP

One of the best “sanity checks” is scaling AI investment by the size of the economy. While absolute AI/data-center capex is enormous (5 hyperscaler firms are expected to spend roughly half a trillion dollars in 2026 - and $1tr in 2028), AI spending as a share of GDP is estimated around ~1%–1.5%, which is not unprecedented compared with past investment waves.

That doesn’t mean “no bubble.” It means the real-economy footprint is already meaningful — and that matters for durability.

2) Mega-cap tech is funding capex from cash flow, not necessarily fragile balance sheets

Another difference versus 2000: many of today’s leaders are high quality businesses with real profitability.

Relative to the dot.com era, today’s mega-cap tech firms have stronger balance sheets and positive cash flows, and thus much of the spending is financed through cash flow rather than debt.

That’s a genuine stabilizer: it lowers the probability of a sudden “financing stop.” Though, large capital spending, if financed by a large debt increase, may be the ultimate channel to burst the bubble.

3) AI is already feeding into measurable macro channels

If AI-driven infrastructure is large enough to matter, it will show up in macro models — even if imperfectly.

Chart 8: AI and data centers: a boon to GDP growth; source: PGIM Global-Risk-Report, Oct-2025

The chart above maps potential GDP-growth contribution channels (construction, equipment, power, productivity spillovers) and shows the current importance of the AI theme for GDP growth.

You don’t need to accept any single forecast to take the point: AI capex is broadening into the real economy, which can keep earnings and sentiment supported.

4) Constraints can support the trade (until they don’t)

Bubbles often burst when supply overwhelms demand. But in parts of the AI stack, the nearer-term story is shortage, not glut.

Chart 9: Estimated data center supply shortfall; source: PGIM Global-Risk-Report, Oct-2025

This chart projects a widening gap between existing/pipeline supply and demand under base and “AI upside” scenarios. If physical capacity is scarce, pricing power can persist longer than skeptics expect.

And that scarcity spills into rising energy costs. AI demand, data centers, a resurgence in domestic factories, an increase in EV ownership, and other power-hungry endeavors are driving rapid capital expenditures to build out energy infrastructure. Electricity consumption from data centers alone is spurring significant investment from hyperscalers, data center operators, and asset managers.

Chart 10: U.S. electricity prices have soared; source: PGIM Global-Risk-Report, Oct-2025

Rising power costs are not great for the economy (and may be another trigger to burst the AI bubble) — but they’re consistent with the notion that AI buildout is colliding with real-world constraints. That collision can keep the investment cycle alive (more grid capex, more incentives, more urgency).

5) Fiscal tailwinds can extend late-cycle dynamics

Investors' “bubble timing” often fails because policy can change the runway.

The U.S. economy faces tailwinds ranging from fading trade-war uncertainty to dollar depreciation, lower oil prices, and event-driven boosts — plus tax-related demand/capex support, the One, Big, Beautiful Bill Act provisions and its tax/credit/deduction changes.

Whether you love or hate the policy mix, the market takeaway is simple: late-cycle stimulus can keep nominal growth and earnings supported longer than “valuation mean reversion” models assume.

This is where investors get trapped. If you’re right about “bubble,” you’re tempted to jump to “collapse.” But between those two sits a long stretch of:

• strong earnings beats from the leaders,

• ongoing capex headlines,

• policy tailwinds,

• and index concentration that forces passive flows into the same names.

What to do with this (without pretending you can time it)

If you accept three premises…

1. U.S. large-cap equity is increasingly concentrated and richly valued,

2. AI-related expectations are high enough to create asymmetric disappointment risk,

3. Fiscal/macro tailwinds can keep the cycle running,

…then the right response is usually not “all-in / all-out.” It’s portfolio architecture.

There are many other areas to invest - international value or small caps, Japanese stocks, driven by corporate reforms, even U.S. value stocks. That’s also exactly where hedge funds (used properly) earn their keep (compare How Much Alternatives Does Your SAA Really Want?):

• If concentration risk is your enemy, you want strategies that monetize dispersion and reduce single-factor dependence.

• If the path risk is a sudden drawdown, you want crisis convexity (or at least strategies that historically behave better when correlations spike).

• If the cycle persists but gets choppier, you want alpha engines that don’t require a market crash to work.

Bottom line

Yes, the AI boom has bubble characteristics in U.S. equities: valuation pockets are extreme, narrative saturation is visible, and index concentration amplifies fragility.
But the setup is not a carbon copy of 2000. Cash-flow-funded capex, macro spillovers, real-world infrastructure constraints / demand, and fiscal tailwinds can plausibly keep the cycle alive for the foreseeable future.

That combination — “bubble-like, but durable” — is exactly why the best response is often diversification and intelligent hedging, not binary timing. And it’s why hedge funds, selected for true diversification (not disguised equity beta), can be a practical risk mitigator in an AI-driven macro market regime.