What Wall Street Really Thinks About the AI Trade Now

Investors across the US, Europe, and Asia remain overwhelmingly bullish on AI, but that consensus does not show how they are positioned. Alex Altman described the mood as an AI-bubble “denial phase.” Some semiconductor and memory stocks trade at about three times earnings, but low multiples for cyclical memory companies may reflect fears that earnings are near a peak, not confidence in rapid growth. The market’s central question is increasingly whether massive investment can produce enough revenue, rather than whether AI will continue to attract attention.

Financing looks most precarious for neoclouds, which build computing capacity to sell but often have little or no operating cash flow, unlike cash-generative hyperscalers such as Google and Amazon. Altman compared neoclouds’ role to subprime vehicles before the financial crisis, while stressing that the analogy is not direct. Financing strains have appeared in individual companies and credit markets: Cypher’s attempt to raise funds hurt its share price, CoreWeave bonds later sold off without new company news that week, and data-center bonds had been widening for months. A survey found about 60% of investors expected AI financing to remain a problem; borrowing is becoming selective, rates still matter, and AI competes with other sectors for capital. Investment-grade and high-yield net issuance was forecast at $1.1 trillion for the year, with 2026 issuance unusually high relative to two-year yields.

The bullish case depends on GPU demand continuing to rise and purchased or rented compute retaining value, but the returns must ultimately come from customers spending on AI. Ramp and OpenRouter data suggested token spending had begun to level off in the near term, although OpenRouter covers only a slice of users and may not capture frontier-model demand. Anthropic’s S-1 showed sharply rising top-line figures, but the gap between billions in earnings and valuations in the trillions remains contentious. Open-weight models have improved toward frontier models on benchmarks and cost much less to operate; one estimate put their market share as high as 80%, up from about 20%. Token costs have fallen while GPU rental rates have risen, and better hardware could shift more agent and model work onto phones.

Company announcements and results continue to move expectations. Oracle shares rose 12% as new guidance and investor-day information circulated; the company reportedly said it would limit additional fundraising to about $20 billion through May. Google’s indication that capex would be significantly higher helped revive AI enthusiasm, while many investors appeared more focused on hyperscaler spending and debt than on the Fed’s October decision. Alphabet shares fell 7.9% after revenue missed estimates, while Microsoft rose 6.4% after first-quarter cloud revenue beat estimates. Azure grew 48%, versus a Bloomberg-compiled estimate of 44.8%; Google Cloud revenue was reported at $30 billion, with 103% growth. Alphabet was also said to be processing 32 billion tokens per minute, up 40% quarter over quarter.

The scale of projected investment raises the stakes: forward capex figures discussed could imply spending of about 12% of GDP, though consensus estimates could change. About two-thirds of next year’s S&P earnings growth is expected to come from semiconductors, rising to roughly 75% from the broader AI ecosystem. Companies with strong legacy businesses and substantial cash generation can fund very large AI investments, helping explain why broad S&P exposure to AI can support the index even when individual stocks react sharply. Yet enterprise revenue may not support expected returns quickly enough. One speaker expected a familiar, roughly 100-IQ workflow to become nearly ubiquitous within 12–18 months, citing ChatGPT finding a child’s pickup time across scattered emails; the benefits are expected to differ by industry and company.

AI-related rotations can be severe even when the S&P 500 and standard volatility measures look calm, because much of the trading is index-agnostic. Multi-manager fund assets under management rose from about $300 billion in 2020 to more than $1 trillion today, while more capital and leverage have accumulated around a stable or shrinking pool of tradable securities. Factor neutralization may shift crowding rather than remove it, and funds have not settled on how to map or hedge an AI factor whose constituents and narrative keep changing. Shorting software was treated as part of the AI trade earlier in the year, but software later traded near 12-month highs; Altman said being short AI over the past three years was probably not lucrative. Investors are becoming more selective, and rates remain relevant at the margin, especially for leveraged businesses. Hyperscalers’ earnings power also depends on customers using AI to improve productivity, a benefit not yet visible in those customers’ earnings.