Goldman Sachs on AI trading: Risks for "AI infrastructure" in the second half of the year, and "losers" in "AI applications" unlikely to recover in the short term.
```
As AI capital expenditures surge and valuations become increasingly expensive, Goldman Sachs reminds the market: The real risk often emerges at the moment "growth starts to slow down."
On February 24, Goldman Sachs Global Investment Research stated in its strategy report "The Broadening and Narrowing of the AI Trade" that recent AI-related market volatility has risen significantly, with two forces pulling in opposite directions: on one side, tech giants' capital expenditures continue to surpass expectations; on the other, investor concerns about "AI disrupting traditional industry profit pools" are rapidly escalating.
Driven by strong capital expenditure guidance, storage chip stocks have jumped an average of 55% year-to-date. Conversely, due to panic around the "AI disruption" narrative, software stocks have plunged 24%. The same "AI theme" has produced nearly opposite market trends at different stages.
Goldman divides this volatile AI trade into four stages, and the stock performance in each has already become starkly different:
Phase 1 (Computing Power Leaders, e.g. Nvidia): Now facing scrutiny for "excessive profitability," these stocks recently show a disconnect where earnings expectations are significantly revised upward but the stock price stagnates.Phase 2 (AI Infrastructure, e.g. memory, equipment, servers, etc.): Driven by robust capex guidance from tech giants, these stocks have continued to soar recently, with memory stocks up 55% year-to-date.Phase 3 (AI Application Enablement, e.g. software services): Amid heightened market fears that their traditional business models will be disrupted by AI, these have suffered panic-driven selloffs, with software stocks plummeting 24% YTD.Phase 4 (AI-driven productivity gains in non-tech industries): Due to unclear financial returns, stock prices here have been moving sideways.
In the face of this extreme divergence, the report shows that whether a "infrastructure winner" surging now or an "application loser" plunging, both currently harbor their own risks.
Capital Expenditure Growth Nearing a Peak, "AI Infrastructure" Faces Valuation Risk
The first thing the market must digest is another upward revision in capital expenditure expectations.
According to Goldman’s compiled consensus expectations, tech giants’ AI capex will reach $667 billion by 2026. This number is $127 billion higher than at the start of Q4 earnings season, a 62% YoY increase.
The other side of these capex upgrades is that free cash flow is being squeezed.
The report stresses: "Super cloud vendor capex is set to exceed 90% of operating cash flow this year, a level even higher than during the Internet bubble." In more detailed estimates, Goldman points out that by 2026, capex is projected to be 92% of tech giants’ operating cash flows.
To fill this huge funding gap, giants are forced to significantly cut shareholder returns. In 2025, total share buybacks will be cut by 15%; the ratio of cash flow used for buybacks has plummeted from 43% at the start of 2023 to 16% now. Oracle and Google are also increasingly tapping the bond market.
Goldman expects that there is still room for further capex increases this year. Since Oracle and Microsoft’s fiscal year ends in May/June, the upcoming Q2 earnings season could again catalyze higher spending expectations.
But Goldman warns: the core risk isn’t in the "absolute value," but in the "growth rate."
"We expect the consensus estimates for super-cloud capex spending have moderate room for further upside, yet we still anticipate capex growth rates will peak later this year."
This deceleration will become the "Achilles’ heel" of AI infrastructure stocks.
Second-half Risk for "AI Infrastructure": Slowing Spending Growth and "Over-Earning" Trap
Goldman stresses: "Once capex growth slows, some AI infrastructure stocks’ revenue growth and valuations will become extremely fragile."
The logic is straightforward: infrastructure chain orders, revenues, and profits are highly sensitive to capex growth rates. When the market shifts from "accelerating every quarter" to "still growing, but no longer accelerating," the most vulnerable part of valuations is the "growth premium."
Goldman straightforwardly notes, many AI infrastructure segments have experienced significant valuation multiple expansion over recent years, but history shows investors typically assign lower multiples to companies with slowing growth.
This is also the core meaning of the report's "valuation squeeze": even if profits are still rising, as soon as the market frets about growth sustainability, multiple contraction can offset the stock price support from earnings upgrades.
Among the sub-sectors listed in the report, manufacturing equipment, servers and networking, wafer foundry and IDMs, and power/utilities all carry valuations higher than their past five-year averages.
Goldman believes that the latest "bottleneck" within infrastructure is in the memory segment.
The report notes that major memory stocks (Micron, Western Digital, SK Hynix, Samsung) have risen an average of about 145% since the start of Q4 2025 and 55% year-to-date. The strong demand and price increases, which have improved profitability, explain most of this rally.
They also point out that the average forward P/E for memory stocks is about 12 times, lower than the broader market and their own five-year average—so, on the surface, not "expensive."
But Goldman immediately uses Nvidia as a warning: when the market worries that a company is "over-earning," the stock no longer follows earnings upgrades upward.
From the end of 2022 to mid-2023, Nvidia’s stock price and earnings grew 12-fold in sync, with its valuation multiple largely unchanged. But recently, the logic has changed.
Goldman points out: "Over the past five months, although Nvidia's forward earnings expectations were raised by 37%, the stock price has been basically flat."
Goldman summarizes this as the market psychology of "over-earning": when a company’s performance is excessively strong at the cycle’s peak, this tends to spark concerns about more competition and sustainability of demand, resulting in "stronger earnings, but multiple contraction."
On the trading side, this means: Even if infrastructure results continue strong in the short run, investors will start scrutinizing the 'second derivative' of growth and whether multiples can expand further.
Tech Giants to Continue Short-term Divergence: Focus Is Not on Capex, But "Returns"
Goldman judges that short-term performance divergence among tech giants will continue.
Because in the first half of 2026, as quarterly capex growth stabilizes, the market’s attention will turn to "whether AI investments are actually delivering returns."
The report provides a clear comparison: tech giants’ free cash flow yield is about 1%, near all-time lows, while other S&P 500 companies average about 4%.
As free cash flow weakens and conversion rates fall, capital will naturally seek alternatives. Goldman plainly states, "Investors are increasingly allocating funds elsewhere."
AI Application Layer: A "Thin Line" Divides Winners and Losers
If the key issue for infrastructure is "how fast capex can grow," the key for the application layer is "who will be disrupted, and who can capture incremental revenue."
Goldman judges that the trend of AI trading spreading to the application layer is a natural progression: after the infrastructure is built, value creation shifts from "selling picks and shovels" to "transforming business models" and recouping initial investment by reshaping profit pools.
But this makes market results far more "micro-level." Goldman stresses that, going forward, more company-specific judgments will be needed, such as competitive position, entry barriers, and pricing power.
The report summarizes the core uncertainty at the application layer:
"With the ultimate competitive landscape still uncertain, the dividing line between a company being seen as an AI revenue ‘winner’ or facing disruption concerns is very thin."
A direct result is that investors are not assigning high valuations to many listed companies for "incremental AI-driven revenue."
Goldman says, "Contrary to our expectations, investors are pricing in almost no upside for incremental AI revenues at public companies; instead, the most attention is given to AI applications in private companies."
The report lists progress at several private firms: Anthropic launched its Claude Cowork tool (including legal, HR, and business service plugins); Insurify launched a price-comparison app within ChatGPT; Altruist created tools for personalized tax strategies for wealth management clients.
These types of cases reinforce a worry in public markets: Even if AI brings new demand, the incremental revenue may not go to public companies.
Why "Losers" Will Have a Hard Time Recovering Short-term: Disruption Concerns Are Hard to Falsify with Short-term Results
The application layer's flip side is the impact that disruption narratives have on valuations.
Goldman notes that recent weeks’ market focus has been on "AI disruption risk."
The report says software stocks have fallen about 23% in the past six weeks, and "despite resilient short-term profits, investors are increasingly skeptical about the long-term growth prospects for the industry."
Goldman gives a very clear judgment here: "It will be hard to disprove concerns about AI disruption in the short term."
They further point out: For companies labeled by the market as "potentially disrupted by AI," stabilizing the stock price depends on first stabilizing profits—but "this disruption uncertainty is unlikely to be resolved in the short term."
Goldman spells out what it would take for "application losers" to recover: "Investors will need either multiple quarters of evidence demonstrating business resilience or for these stocks to see substantial valuation downgrades relative to the market before large-scale re-engagement."
This is the dilemma facing software and other sectors now: even if short-term earnings are alright, the market is trading on whether the "long-term profit pool will be redistributed."
Goldman's Two Metrics for "Quantifying" Disruption Risk: Human Labor Exposure and Asset Intensity
As for observing "which companies are more easily disrupted," Goldman provides two factors (noting that there are regulatory barriers, market power, and other factors too).
First, exposure of the workforce to AI automation.
Goldman notes that recent concerns about white-collar job replacement are rising.
They worked with economists to estimate each company's wage expenses exposed to AI automation and considered the ratio of "labor cost/income."
Goldman reminds that this is a "double-edged sword": AI can both improve efficiency and eliminate jobs.
But in trading terms, over the past six months, sectors with lower "exposure" have been rewarded, while those with higher exposure have been penalized.
Second, tangible asset intensity.
Goldman uses "(assets - cash - intangibles)/revenue" to measure asset intensity and constructed sector-neutral, equal-weight baskets.
They observed that companies with "heavier" assets have recently outperformed "lighter" asset firms, to an extent exceeding what macro factors can usually explain.
Similarly, goods-producing companies have outperformed service-oriented firms.
For investors, these two clues do not suggest "heavier assets are always better," but that the market is using them as "proxy indicators for moats/barriers to entry" to counter application layer uncertainty.
Three Major Catalysts: Goldman Bets the "Turning Point" Comes in Second Half of 2026
Goldman believes that for tech giants to regain market leadership, three catalysts are needed.
Their base case is: these catalysts are "more likely to emerge in the second half of 2026."
First, AI-related revenues must accelerate. As earnings season reactions show, as long as revenue growth beats expectations (e.g. Meta rallying 10%), investors regain confidence in AI investment returns.
Second, "visibility" on free cash flow (FCF) bottoming, as capex growth slows. Goldman believes that once a bottom in cash flow is clear, investors may start pricing these firms on profitability rather than cash flows, reducing volatility in valuations.
Goldman explains: "Slower capex growth will give investors hope that free cash flow can rebound from the bottom. This will prompt investors to price these giants based on earnings again." Currently, tech giants' forward P/E of 24x is only at the 14th percentile of the past decade, making valuations highly attractive.
Finally, fading macro tailwinds. Goldman economists predict the cyclical acceleration in the U.S. economy will peak by mid-year and fall back in the second half. When the macroeconomic boon fades, capital will inevitably return to the tech giants with high long-term certainty.
~~~~~~~~~~~~~~~~~~~~~~~~
The above content is from Chasing Wind Trading Desk.
For more detailed interpretations, including real-time analysis and frontline research, please join [Chasing Wind Trading Desk Annual Membership]
Risk Warning and DisclaimerThe market involves risk, and investment should be cautious. This article does not constitute individual investment advice, nor does it consider the special investment objectives, financial situation, or needs of any particular user. Users should consider whether any opinions, perspectives, or conclusions in this article fit their particular circumstances. Investments are made at your own risk. ```