Goldman Sachs: AI Trade Narrative Shifts from Hardware to Application Layer
nashnova research
Two Goldman Sachs strategists say the AI investment thesis is migrating from the chip arms race toward inference economics and personal agents, with semiconductor valuations compressing and capital rotating into application-layer companies — the year-end base case is still bullish, but the narrative's staying power hinges on how fast agents reach commercial reality.
Why are semiconductor valuations under pressure?
The market has reached early consensus on 2027 capex, providing roughly 12 to 15 months of visibility.
Beyond 2027, the roadmap is unclear. This means → investors cannot tell whether today's elevated earnings are sustainable, so multiples compress.
In plain terms = the market will pay for profits it can see, but it hesitates when the next chapter is still blank — that is the core reason semis have pulled back.
Where is the trading focus migrating?
Goldman TMT specialist Pete Callahan sees capital shifting from chips to the application layer — specifically cybersecurity, data infrastructure, and other tools that help enterprises and consumers deploy AI.
He stresses this is "both/and," not "either/or" — semis remain within striking distance of their highs, and the market is making room for both directions.
This reflects a broader signal: the market has not abandoned hardware, but after digesting hardware's near-term certainty, it is hunting for the next incremental story one layer up.
Why are personal agents seen as the next breakout?
Callahan's read: coding AI is no longer a secret; the market is "very hungry for the next story."
Personal agents — AI assistants that autonomously execute tasks for users — and the consumer supply-chain ripple effects they trigger will be the key watch item in Q3 earnings season.
This means → if large companies show concrete evidence of agent-driven cost savings or new revenue during earnings, the AI narrative gets extended; if they cannot, markets will push back.
How narrow has market breadth become?
Goldman chief U.S. equity strategist Ben Schneider says the median S&P 500 stock sits more than 15% below its high, and the firm's preferred breadth gauge is at its narrowest since the dot-com bubble.
History suggests breadth gaps close through catch-up rallies, not sell-offs. In plain terms = most of the time it is the laggards that rise, not the leaders that fall.
Schneider's base case: combining strong earnings, light positioning, and reasonable valuations, the market moves higher by year-end.
What offsets the post-2027 earnings slowdown?
Current consensus expects earnings growth of roughly 30%, but that pace will decelerate into 2027 — driven by macro headwinds and a natural slowdown in AI capex growth itself.
This means → if semiconductor companies can extend visibility out to 2028, it would directly ease the single biggest overhang on valuations today.
The speed at which personal agents reach commercial reality is what Schneider calls the critical validation checkpoint for whether the entire AI narrative can endure.
市场有风险，内容仅供研究参考，不构成投资建议。