Nvidia beats estimates but AI trade faces $2 trillion valuation doubts

Wall Street veteran analyst David Woo says the AI trade is driven by fear of missing out rather than economics, warning that a narrative collapse would hit nearly every S&P 500 investor holding AI-exposed megacap stocks.

"The entire capital expenditure story is driven by fear, the fear of being left behind, not economic considerations," Woo said on a podcast. "Fear has no logic, no economic logic, and fear has no ceiling."

Woo, who previously shorted AI assets before tactically exiting, targets Anthropic's roughly $2 trillion valuation goal as built on a winner-take-all assumption he says does not hold. He also questioned Microsoft's decision to extend AI data center depreciation from 15 to 25 years, which produced zero depreciation last quarter, and estimated that stripping out Amazon and Google's Anthropic stake revaluation gains would cut tech earnings growth by about 30 percent, with another 30 percent shaved by tariff benefits.

The warnings arrive as Nvidia, the AI trade's bellwether, beat second-quarter estimates and raised guidance, with data center revenue up 117 percent year over year to $89 billion. Yet the muted initial reaction to the report — shares rose 4 percent only after CEO Jensen Huang said AI had reached an inflection point — shows how high expectations have climbed.

A Winner-Take-All Assumption Under Scrutiny

Woo argues the AI sector's valuations rest on a winner-take-all outcome that history does not support. "People think, oh, we're going to have the next Google or Apple," he said. "But I'm very sure that no matter how good Anthropic is, it will never dominate the way Google dominated search." He describes AI as a commodity that gets cannibalized rather than a business with a natural moat, and predicts even Google's search franchise becomes a casualty as traditional search looks weak next to AI.

Accounting Questions and the $730 Billion CapEx Wave

The bearish case extends to how tech giants account for their AI buildout. Woo called Microsoft's depreciation extension "completely unreasonable," noting that chips make up about 60 percent of data center costs yet have a useful life of two to three years. His concerns come as Microsoft and Meta Platforms, two of Nvidia's largest customers, reinforce expectations that Big Tech will spend more than $730 billion on AI infrastructure this year, up from $400 billion last year.

Nvidia said its maximum gross exposure under land, power and shell guarantee agreements totals $3.5 billion, a fraction of its quarterly revenue. But skeptics warn that circular deals — where AI goods producers finance other firms in the sector — could inflate demand and distort economic signals.

Why Shorting the AI Trade Is Hard

Despite his bearish view, Woo concedes the trade is difficult to short. He points to Nvidia's reported $125 billion guarantee to OpenAI and SpaceX's reported $250 billion commitment to build new data centers exclusively with Nvidia, alongside a political environment that can turn on a promotional machine. "How do you fight that?" he asked.

His exit is tactical rather than a change of view. Because nearly every S&P 500 investor holds AI-exposed assets, a narrative break would spread quickly across the entire U.S. equity market. "That would be a disaster," he said.

So What for Investors

Nvidia shares, up about 12 percent year to date, trail the Philadelphia SE Semiconductor index's more than 60 percent gain, even as the company has beaten analyst estimates for eight straight quarters. Options traders priced a 5.4 percent move in either direction ahead of the report, the quietest expectation since 2021. JPMorgan reiterated its overweight rating with a $280 price target, implying about 31 percent upside, while Woo's warnings suggest the risk is not this quarter's numbers but the assumptions underneath them.

This article is for informational purposes only and does not constitute investment advice.