Goldman Sachs has warned that the growth rate of AI capital expenditure is expected to slow down in the second half of the year. This turning point will directly threaten AI infrastructure stocks that heavily rely on capital expenditure, leading to a collapse in their valuation premium. For instance, NVIDIA is currently experiencing a phenomenon of ‘surging profits without a corresponding rise in stock price.’ On the other hand, software and other AI application-end losers, plagued by fears of ‘disruption,’ cannot dispel long-term market concerns with short-term earnings reports alone, making it extremely difficult for them to rebound in the short term.
Goldman Sachs reminded the market that the real risk often emerges at the moment when growth starts to slow down, as AI capital expenditures surge and valuations become increasingly expensive.
On February 24, Goldman Sachs Global Investment Research noted in its strategy report ‘The Broadening and Narrowing of AI Trades’ that recent volatility in AI trades has significantly increased, driven by two opposing forces: on one side, tech giants’ capital expenditures have continued to exceed expectations, while on the other, investors’ concerns about ‘AI disrupting traditional industry profit pools’ are rapidly escalating.
Driven by robust capital expenditure guidance, memory chip-related stocks have surged an average of 55% year-to-date, whereas software stocks have plummeted 24% amid panic over the ‘AI disruption theory.’ The same ‘AI theme’ has presented nearly opposite trends across different segments.
Goldman Sachs divides this highly volatile AI trading into four phases, with stock price movements now diverging significantly across these stages.
Phase One (computing leaders such as NVIDIA): Facing scrutiny over ‘excessive profitability,’ a recent disconnection has emerged where earnings expectations have been revised sharply upward but stock prices remain range-bound.
Phase Two (AI infrastructure, such as storage, equipment, and servers): Driven by strong capital expenditure guidance from tech giants, there has been a sustained surge recently, with memory stocks up 55% year-to-date.
Phase Three (AI-powered applications, such as software services): Due to extreme market concerns about their traditional business models being disrupted by AI, there has been a panic-driven sell-off recently, with software stocks plunging 24% year-to-date.
Phase Four (AI productivity enhancement in non-tech industries): With actual financial returns still unclear, stock prices have remained range-bound recently.
In light of this extreme divergence, the report shows that both the currently soaring ‘infrastructure winners’ and the plunging ‘application losers’ harbor latent risks.
With capital expenditure growth rates nearing their peak, the risk of a valuation correction for AI infrastructure is approaching.
The market must first digest the ‘higher gear’ in capital expenditure expectations.
According to consensus forecasts compiled by Goldman Sachs, AI-related capital expenditures by tech giants are expected to reach $667 billion in 2026. This figure is $127 billion higher than at the start of the Q4 earnings season, reflecting a year-over-year growth rate of 62%.
The flip side of the substantial upward revision in capital expenditures is the squeeze on free cash flow.
The report emphasized: ‘Capital expenditures by hyperscale cloud vendors are on track to exceed 90% of operating cash flow this year, a ratio even higher than during the dot-com bubble.’ More specifically, Goldman Sachs projects that capital expenditures will account for 92% of tech giants’ operating cash flow in 2026.
To bridge the massive funding gap, these giants have been forced to significantly reduce shareholder returns. In 2025, overall stock repurchases by major firms fell by 15%; the proportion of cash flow used for buybacks plummeted from 43% at the beginning of 2023 to the current 16%. Meanwhile, companies like Oracle and Google have increasingly turned to the bond market for financing.
Goldman Sachs anticipates that there is still room for an upward adjustment in the absolute level of capital expenditures within the year. Given that Oracle and Microsoft’s fiscal years end in May/June, the upcoming Q2 earnings season could act as a catalyst for another upward revision in expenditure expectations.
However, Goldman Sachs warns that the core risk lies not in the ‘absolute value,’ but in the ‘growth rate.’
We expect the consensus forecast for hyperscale cloud vendors’ capital expenditures to have modest upside potential, but we still anticipate that the growth rate of capital expenditures will peak later this year.
This slowdown in growth will become the ‘Achilles’ heel’ for AI infrastructure stocks.
The H2 risk for ‘AI infrastructure’: Slower expenditure growth and the ‘excessive profitability’ trap.
Goldman Sachs emphasized: ‘Once the growth rate of capital expenditure slows down, the revenue growth and valuations of some AI infrastructure stocks will appear extremely vulnerable.’
The logic is straightforward: orders, revenues, and profits in the infrastructure chain are often highly sensitive to changes in the growth rate of capital expenditure. When the market shifts from ‘accelerating every quarter’ to ‘still growing but no longer accelerating,’ the most vulnerable part of the valuation is often the ‘growth premium.’
Goldman Sachs stated outright that many sectors related to AI infrastructure have experienced significant expansion in valuation multiples over the past few years. Historical experience shows that investors typically assign lower valuation multiples to companies showing signs of ‘slowing growth.’
This is also the core meaning behind the term ‘multiple compression’ in the report’s theme: even if earnings continue to grow, once the market begins to worry about sustainability of that growth, multiple contraction may offset any stock price support brought by upward revisions in earnings.
Among the sub-sectors listed in the report, valuations in areas such as manufacturing equipment, servers and networking, foundries and IDMs, and power and utilities are generally higher than their five-year averages.
Goldman Sachs believes that the ‘latest bottleneck’ within the current infrastructure space is concentrated in the memory segment.
The report notes that major memory stocks (such as Micron, Western Digital, SK Hynix, and Samsung) have risen by an average of approximately 145% since the beginning of Q4 2025, with an average increase of about 55% year-to-date. Goldman Sachs believes that strong demand and price increases driving profit improvements can explain the majority of this rally.
They also pointed out that the average forward P/E ratio for memory stocks is around 12 times, which is not only lower than the broader market but also below their own five-year average, making them superficially appear not ‘expensive.’
However, Goldman Sachs immediately issued a warning using NVIDIA as an example: when the market begins to worry that a company is in a state of ‘over-earning,’ stock prices may no longer follow upward revisions in earnings.
From the end of 2022 to the middle of last year, NVIDIA’s stock price and earnings both grew 12-fold, with the valuation multiple remaining largely unchanged. However, in the most recent phase, the logic has shifted.
Goldman Sachs noted, ‘Over the past five months, despite NVIDIA’s forward earnings expectations having been significantly revised upward by 37%, its stock price has remained largely flat.’
Goldman Sachs encapsulated this phenomenon as a market psychology of ‘over-earning’: when companies perform too strongly at the peak of a cycle, it tends to trigger concerns over intensifying competition and the sustainability of demand, ultimately manifesting as ‘continued strong earnings but contracting valuations.’
In trading terms, this implies that even if short-term performance continues to materialize in the infrastructure chain, investors will begin scrutinizing more closely the ‘second derivative of growth’ and whether multiples can continue to expand.
Short-term divergence among tech giants: focus not on capital expenditure, but on ‘returns’
Goldman Sachs anticipates that return divergence among tech giants will persist in the short term.
This is because, by the first half of 2026, when quarterly capital expenditure growth stabilizes overall, market attention will shift to whether AI investments are yielding returns.
The report provides an illustrative comparison: free cash flow yield for tech giants is approximately 1%, at its historical low; for the rest of the S&P 500 companies, it is around 4%.
As free cash flow weakens and conversion rates decline, capital naturally seeks alternative options. Goldman Sachs stated directly, ‘Investors are increasingly reallocating funds elsewhere.’
AI application layer: A ‘very fine line’ separating companies into winners and losers
While the contradiction at the infrastructure layer lies in ‘how much faster capital expenditure can grow,’ the contradiction at the application layer is about ‘who will be disrupted and who can capture incremental revenue.’
Goldman Sachs argues that the diffusion of AI trading into the application layer is a natural path of technological development: after the infrastructure is in place, value creation will shift from ‘selling shovels’ to ‘transforming business models,’ and recoup early investments by reshaping profit pools.
However, this also makes equity market outcomes more ‘micro-focused.’ Goldman Sachs emphasized that future assessments will increasingly rely on company-level judgments, such as competitive positioning, entry barriers, and pricing power.
A single sentence in the report highlights the core uncertainty at the application layer:
“In a situation where the ultimate competitive landscape remains uncertain, the line between a company being perceived as an AI revenue ‘winner’ and facing fears of ‘being disrupted’ is very thin.”
One direct consequence is that investors are currently not assigning high valuations to many listed companies for ‘incremental revenue driven by AI.’
Goldman Sachs noted, “Contrary to our expectations, investors have hardly priced in the upside potential of AI-driven revenue growth for public companies; instead, AI applications from private companies have garnered the most attention.”
The report highlighted progress in several private company products: Anthropic launched Claude Cowork (featuring plugins for legal, human resources, and business services); Insurify introduced a price comparison app within ChatGPT; Altruist rolled out tools to create personalized tax strategies for wealth management clients.
Such examples reinforce a concern in public markets: even if AI generates new demand, the incremental revenue may not accrue to listed companies.
Why ‘losers’ struggle to rebound in the short term: disruption fears are difficult to disprove with ‘short-term performance.’
On the flip side of the application layer lies the valuation damage caused by disruptive narratives.
Goldman Sachs pointed out that the market’s focus over the past few weeks has been concentrated on the ‘risk of AI disruption.’
The report stated that software stocks have fallen by approximately 23% over the past six weeks, and ‘despite short-term earnings remaining resilient, investors are increasingly questioning the industry’s long-term growth prospects.’
Goldman Sachs provided a very clear assessment here: ‘Concerns about AI disruption are unlikely to be disproven in the short term.’
They further noted: for companies already labeled by the market as ‘potentially disrupted by AI,’ the key to stabilizing stock prices lies in first stabilizing earnings; however, ‘this uncertainty of disruption is unlikely to be resolved in the short term.’
Goldman Sachs specified the conditions under which ‘losers in the application layer will struggle to recover in the short term’: ‘Investors either need evidence of business resilience over multiple quarters or a more significant valuation decline relative to the broader market before re-engaging on a large scale.’
This is also the dilemma facing sectors such as software at present: short-term earnings reports may still be acceptable, but the market is trading on whether ‘long-term profit pools will be redistributed.’
Goldman Sachs used two criteria to ‘quantify’ the risk of disruption: labor exposure and asset intensity.
In terms of how to assess ‘who is more likely to be disrupted,’ Goldman Sachs provided two vectors (while emphasizing there are other dimensions such as regulatory barriers and market power).
First, the exposure of the workforce to AI automation.
Goldman Sachs noted that concerns about the replacement of white-collar jobs have recently risen.
They collaborated with economists to estimate the proportion of wage expenditures exposed to AI automation across companies and observed this in conjunction with the ‘labor cost/revenue’ ratio.
Goldman Sachs cautioned that this indicator is a ‘double-edged sword’: AI could enhance efficiency but also replace jobs.
However, at the trading level, the market has rewarded industries with ‘lower exposure’ and penalized those with ‘higher exposure’ over the past six months.
The second factor is tangible asset intensity.
Goldman Sachs measures asset intensity using ‘(assets – cash – intangible assets) / revenue’ and constructs industry-neutral, equally-weighted baskets.
They observed that companies with heavier assets have significantly outperformed those with lighter assets recently, to an extent ‘beyond what the macro environment typically explains.’
Similarly, goods-producing companies have outperformed service-oriented companies.
For investors, these two indicators do not imply that ‘the heavier the assets, the better,’ but rather that the market is using them as ‘proxy metrics for moats/entry barriers’ to counter uncertainty at the application layer.
Three Major Catalysts: Goldman Sachs Bets on the ‘Turning Point’ Occurring in the Second Half of 2026
Goldman Sachs believes that three catalysts are needed for tech giants to regain market leadership.
Their baseline assessment is that these catalysts are ‘more likely to emerge in the second half of 2026.’
First, AI-related revenue must accelerate. Market reactions during earnings season have already demonstrated that as long as revenue growth exceeds expectations (such as Meta’s 10% surge), investors will regain confidence in the return on AI investments.
Second, there must be ‘visibility’ of a bottoming out in free cash flow (FCF) due to a slowdown in capital expenditure growth. Goldman Sachs believes that once signs of a cash flow bottom appear, the market may shift to pricing based on profitability rather than cash flow, thereby reducing valuation volatility.
Goldman Sachs explained, ‘A slowdown in capital expenditure growth will give investors hope for a rebound in free cash flow. This will prompt investors to re-price these tech giants based on profitability.’ Currently, these giants’ forward price-to-earnings ratio of 24 times is only at the 14th percentile over the past decade, making their valuations highly attractive.
Finally, the fading of macro tailwinds. Economists at Goldman Sachs predict that the cyclical acceleration of the U.S. economy will peak by mid-year and retreat in the second half. When macroeconomic benefits diminish, funds will inevitably flow back to these tech giants with extremely high long-term certainty.
Editor/Jayden
