If this really is an AI bubble, let’s see some more inflation

If the AI trade were to have already peaked, we’d probably retroactively refer to this stretch as a mix of an earnings bubble (like the period of over-earning based on some unsustainable credit trends that preceded the financial crisis) and a valuation bubble (like the dot-com boom, where rapidly expanding price-to-earnings ratios were the key driver of explosive gains).

But if we’re doing anything close to running back the late ’90s, well, this bubble is going to get some more air to inflate it.

The so-called “Fed model” used to value the S&P 500 by subtracting its expected earnings yield from bond yield currently sits at about 0.35% in nominal terms and 2.6% in real terms, versus lows of -2.75% and -0.4%, respectively, during the dot-com episode. Lower readings indicate a higher willingness to buy risky securities relative to risk-free US government obligations.

In a recent report, Bank of America equity derivatives strategists led by Benjamin Bowler wrote:

“The AI revolution represents another profound technological leap, one that we think is likely to also result in an asset bubble. Strikingly, the late 90s analogy suggests that 2025 is tracking 1996, an interesting parallel even if coincidental. Moreover, a unique tailwind this time is the amplification coming from government support and the perceived existential threat AI dominance presents to geopolitical power. In our view, the likelihood of avoiding a significant asset bubble in AI seems low given this backdrop.”

Using its in-house methodology for assessing whether assets are in a bubble, they judge that “​​the core of the AI trade in the S&P, Nasdaq and the Magnificent 7 stocks remains far from these levels,” which suggests “the AI trade may still have room to run into 2026.”

Indeed, none of the publicly traded hyperscalers has a forward price-to-earnings ratio anywhere near Cisco’s peak of over 130 during the dot-com bubble.

Amazon, Meta, Google, and Microsoft don’t really trade at ridiculously high multiples on this metric, relative to their own history or the S&P 500. But for all except Amazon, these stocks have set fresh valuation peaks based on price to estimated free cash flow in 2025, data from Bloomberg shows.

(Oracle seems to be its own kettle of fish, so we’ll leave that alone to its Sam Altman-filled sea of doubts and debt here).

The difference here is capex, which weighs on earnings over time via depreciation expenses but impacts how much money is going in and out of the door immediately.

Ryan Cummings, chief of staff at the Stanford Institute for Economic Policymaking, notes that AI isn’t anywhere close to the lion’s share of sales or earnings for these firms, and estimates that AI-centric sales are being far outstripped by AI capex.

That’s pretty reasonable, considering that we’re still arguably in the early stages of pushing this technological frontier and that a good chunk of this spending is dedicated to making AI models better — putting them in a position where they will be bigger drivers of financial performance going forward.

One way to square this circle between elevated, not crazy forward valuations based on one metric and sky-high ones based on another is to conclude that the lack of runaway forward price-to-earnings ratios suggests that the market does continue to have some skepticism about the long-term earnings power associated with all these capital outlays.

Less doubt would equal higher valuations and higher stock prices. No doubt and unbridled optimism about how much these first movers in AI will reap rewards for years if not decades to come… that’s how we really get a bubble.

“Big Tech has compelling valuation on a growth-adjusted P/E basis, but free cash flow yields are hardly attractive,” wrote Michael Purves, CEO of Tallbacken Capital Advisors. “Ultimately, this valuation methodology debate boils down to the question of whether this massive capex spend will generate compelling returns on invested capital (ROIC). While we won’t know the answer to this ROIC question for some time, we expect the mere existence of this critical question to hover over the markets for some time.”