If the AI trade has already peaked, history will probably remember this stretch as a hybrid bubble. Part earnings-driven, part valuation-driven. Think echoes of the late 1990s, but with stronger balance sheets and much bigger companies. The catch is that true bubbles rarely deflate quietly. They usually need more fuel first.
Right now, the market does not look euphoric enough to call time. If anything, it still looks cautious in key places.
Valuations say excitement, not euphoria
One popular yardstick, the so-called Fed model, compares the S&P 500’s earnings yield to Treasury yields. Today, that spread sits around 0.35 percent in nominal terms and about 2.6 percent in real terms. During the peak of the dot-com bubble, those figures were deeply negative, signaling extreme risk appetite.
That gap matters. Lower readings mean investors are far more willing to buy risky assets over government bonds. We are simply not there yet.
Bank of America’s equity derivatives team recently argued that the AI cycle may still resemble 1996 more than 1999. In their view, the odds of avoiding an asset bubble in AI are low, but the core of the trade in the S&P 500, Nasdaq, and the Magnificent Seven remains far from true bubble territory. Translation: this thing may still have room to run into 2026.
Big Tech looks expensive, but not dot-com expensive
None of the hyperscalers is anywhere near the valuation extremes seen during the dot-com era. Cisco traded at more than 130 times forward earnings at its peak. By comparison, Amazon $AMZN ( ▲ 0.16% ) , Meta $META ( ▼ 0.75% ) , Alphabet $GOOGL ( ▼ 0.18% ) , and Microsoft $MSFT ( ▼ 0.14% ) look restrained on traditional forward price-to-earnings metrics.
That said, another lens tells a different story. On price to estimated free cash flow, most of these companies hit new highs in 2025. The reason is capex. Massive AI investment weighs on free cash flow today through depreciation and upfront spending, even if earnings look reasonable.
This creates a strange split. Earnings-based valuations suggest skepticism. Cash-flow-based valuations suggest investors are already paying up for future dominance.
Capex is the real stress test
The unresolved question is whether all this spending turns into durable returns. Ryan Cummings of the Stanford Institute for Economic Policymaking points out that AI still represents a relatively small share of sales and earnings for Big Tech, while AI-related capital spending is enormous by comparison.
That imbalance makes sense early in a technological cycle. Companies are building capacity before monetization fully shows up. But it also explains why valuations have not completely detached from reality. The market still wants proof.
If confidence were absolute, forward multiples would already be far higher. More doubt keeps prices grounded. Less doubt pushes stocks higher. No doubt at all is how bubbles really form.
As Michael Purves of Tallbacken Capital Advisors put it, Big Tech may look attractive on growth-adjusted metrics, but free cash flow yields are hardly cheap. Until investors get clarity on whether AI capex delivers strong returns on invested capital, that question will continue to hang over the market.
If this truly becomes an AI bubble, history suggests it probably will not end quietly. It may need more inflation, more demand pressure, and more belief before it finally breaks.