Key Takeaways:
The $200 billion-plus annual AI infrastructure build by Microsoft, Amazon, Alphabet and Meta has created a market dynamic Goldman Sachs traders compare to a rubber band — stretched further with every data center announcement, with the only question being when it snaps.
"The AI market has become a rubber band — the question is how much further it can stretch," Rich Privorotsky, head of EMEA Equity Flow Intermediation at Goldman Sachs, said in a note published this week. Privorotsky pointed to the first signs of strain: frontier model capabilities are diffusing rapidly while inference costs collapse, threatening the economic moat that hyperscalers have relied on to justify their spending.
The four largest cloud operators are investing at a pace that exceeds their combined operating cash flow, according to the analysis co-authored by Mark Wilson, Goldman's head of EMEA Equity Hedge Fund Sales. Apollo Global Management economist Torsten Slok, citing the Epoch capability index, noted that open-source models now trail frontier closed-source models by roughly four months — down from a year-plus gap in early 2024. That compression is fueling price wars among API and model providers, with at least two competitive providers already slashing per-token costs.
Privorotsky said the market's current pricing logic initially puzzled him. If cheaper intelligence expands the total addressable market and drives exponential enterprise adoption, hardware stocks should be leading the rally. Instead, the S&P 500 has advanced even as semiconductor and infrastructure names pulled back, suggesting investors are betting that value is migrating up the stack. "The market may be expressing a different judgment — that value is moving to platform companies that own customer relationships, distribution and workflows," he said. The S&P 500 excluding AI-related weights has broken out to new highs, with sector rotation absorbing the full burden of index performance.
The divergence carries high stakes for the second-quarter earnings season, which Goldman's trading desk called the "near-term critical test" for the AI investment return narrative. Momentum-driven strategies have already taken losses from recent positioning unwinds, but the longer-term rotation from hardware to tollgate platforms — companies that control access to AI capabilities rather than the compute itself — is directionally correct, according to the analysts.
For investors, the calculus is shifting. Hardware suppliers face multiple compression risk even before earnings downgrades materialize, as compute capacity commoditizes with supply expansion. Platform companies — those embedding AI into existing software, cloud and distribution channels — stand to capture the economic surplus as inference costs fall and adoption scales. If any single hyperscaler signals a capex pullback in the coming earnings reports, the analysts warned, it could trigger a cascading reassessment across the entire AI value chain.
This article is for informational purposes only and does not constitute investment advice.
