NextFin News - The market is no longer rewarding hyperscalers for promise alone. Goldman Sachs chief global equity strategist Peter Oppenheimer says investors are punishing the group because the AI buildout has shifted from a story about expansion to a test of whether massive capital spending can translate into faster revenue, stronger margins and durable cash generation.
The pressure matters because hyperscalers are the funding engine for the AI infrastructure cycle. When those names trade as if every extra dollar of data-center and chip spending must be justified by near-term earnings, the whole market’s lens changes. The question is no longer whether the firms can spend. It is whether they can prove the spend is compounding value quickly enough to keep valuations intact.
That is the central tension behind the latest move in the AI trade. Goldman Sachs Research says the consensus estimate for hyperscaler capex in 2026 has climbed to $527 billion, up from $465 billion at the start of the third-quarter earnings season, and that the group spent $106 billion in capex in the third quarter, up 75% from a year earlier. A separate Goldman Sachs Research estimate says AI hyperscaler capex still equated to only 0.8% of GDP, well below the 1.5% of GDP or more seen at previous technology booms. In other words, the spending is enormous in corporate terms but still not at an economy-wide peak. That makes the selloff in the stocks look less like a capex ceiling and more like a valuation reset.
Why the Market Is Turning More Demanding
Oppenheimer’s point is not that the AI cycle has ended. It is that the market is changing the terms on which it will finance it. Goldman said hyperscalers are still expected to spend hundreds of billions of dollars on AI infrastructure in 2026, and the company’s own research notes that investors are becoming more selective about where they pay for that growth. The simple version of the trade has already changed: higher capex is no longer enough by itself. The spend has to produce visible top-line acceleration, or the market starts discounting the return profile.
That shift has already been visible in the way large platforms are treated. Alphabet said in June that it expected 2026 capex to reach $180 billion to $190 billion, about six times its 2022 level and double last year’s. Microsoft said its capital spending would keep rising as it expands cloud and AI infrastructure, while Meta has repeatedly lifted its spending plans alongside its AI buildout. Amazon has also kept reinforcing that it is investing heavily in AWS and related infrastructure, even as investors focus on the return on that spending. The pattern is the same across the cohort: each larger budget announcement now invites a second question about monetization, not just ambition.
That is why the move feels more structural than a simple weekly wobble. Hyperscaler capex is not reverting to the old normal; the old normal no longer fits the AI race. But the share prices can still swing cyclically as investors move between rewarding growth and punishing capital intensity. The capex supercycle may be structural. The pain in the stocks is still cyclical.
The distinction matters because structural and cyclical forces are doing different jobs. The structure is the permanent shift in how the largest tech companies allocate capital: they are now building AI infrastructure as a core operating layer, not as a side project. The cycle is the market’s willingness to pay for that buildout. When the market gets impatient, multiples compress before fundamentals fully catch up. That is the mechanism behind the punishment.
The Mechanism: From AI Spend to Multiple Compression
The transmission channel is straightforward but powerful. Heavy capex lifts revenue for chipmakers, networking firms, power suppliers and data-center builders. It can also lift top-line growth for the hyperscalers if demand keeps pace. But once spending outruns visible monetization, the same market that once paid up for optionality starts pricing a lower conversion rate from revenue to free cash flow. That is the mechanism behind the punishment.
Valuation is doing a lot of the work here. For years, the dominant argument was that platform scale justified elevated multiples because future margins would expand. Now the market is asking for a faster proof point. If capex rises faster than revenue or operating income, the denominator in valuation models worsens. That compresses multiples even before any fundamental deterioration shows up in reported earnings. The punishment is therefore partly mechanical: higher spending raises the capital base, which can make current returns look thinner until monetization catches up.
There is also a second-order effect. When hyperscalers spend aggressively, they do not only affect their own margins; they alter the whole earnings hierarchy across tech. Semiconductor suppliers and infrastructure beneficiaries may still benefit, but investors may rotate away from the mega-cap platforms that finance the buildout toward the companies selling the picks and shovels. That is one reason the market can punish the spenders even while endorsing the spend cycle itself. The AI trade is not disappearing; it is fragmenting.
"A capex supercycle is taking hold," Peter Oppenheimer said in a client note.
The strongest counter-thesis is that this is still an early-innings investment cycle, not an overreach. On that view, hyperscalers are front-loading infrastructure into a once-in-a-decade platform shift, and the market is underestimating how quickly AI features, cloud demand and enterprise usage can turn that capex into revenue. Goldman’s own research points to a broadening AI investment base and to continuing earnings gains at companies exposed to the buildout. That argument has force because technology history is full of periods when the biggest spending phase produced the biggest future profits. If the AI buildout ultimately drives a broad productivity step-up, today’s skepticism will look premature.
But the falsifying signal for the bullish version is also clear: if the hyperscalers keep lifting capex guidance without a corresponding step-up in revenue growth, operating margins or free-cash-flow conversion over the next one to two quarters, the market will keep tightening the penalty. That would suggest investors are no longer willing to give the group credit simply for building. They want proof that the build is already paying back.
There is a reason this matters beyond the megacaps themselves. Goldman’s research says the average stock in its infrastructure basket returned 44% year to date, while the consensus two-year forward earnings estimate for that group rose only 9%. That gap captures the second-order tension in the trade: markets are already rewarding the pick-and-shovel layer for near-term demand, but they are becoming less willing to underwrite open-ended spending at the platform level. The beneficiaries and the financiers are no longer priced the same way.
What Changes From Here
In the short term, the market is likely to remain sensitive to any capex revision, margin guide or large AI-related debt issuance. Every upward tweak to spending will be read through the same filter: is this an earnings acceleration story or an efficiency problem? That keeps the shares vulnerable to fast reversals whenever the revenue math fails to improve as quickly as the investment plan.
Over the medium term, the beneficiaries may be less the hyperscalers themselves than the broader AI supply chain. Chipmakers, memory makers, networking vendors, power equipment suppliers and data-center developers all benefit from the spending stream. The hyperscalers create the demand, but the market may prefer the businesses that monetize that demand without having to defend the return on the whole platform.
Over the longer term, the question is whether AI spending rewrites the earnings base of the largest tech companies or merely redistributes profits around the ecosystem. If monetization broadens into durable enterprise demand, ad tools, cloud services and subscription revenue, the current punishment will prove to have been a valuation reset inside a healthy structural cycle. If not, the market may be starting to treat hyperscaler capex the way it treats any other capital-intensive boom: impressive at first, then unforgiving when returns take too long to arrive.
The base case is a volatile but ongoing buildout: spending stays elevated, winners and losers rotate, and the market keeps demanding faster proof of payback. The upside case is that revenue and margin acceleration catch up quickly enough to justify the capex and stabilize multiples. The downside case is that spending keeps rising while monetization lags, which would deepen the multiple compression and shift capital toward the less glamorous suppliers of the AI stack.
What would change that view? A clear break in the spending-to-revenue ratio. If the next two earnings cycles show capex still climbing but revenue growth and operating leverage inflecting materially higher, then the punishment will have been a short, sharp valuation reset inside a larger bull market. If not, the market is likely to keep treating every new AI dollar as a question, not a promise.
The market is not saying the AI race is over. It is saying the entry fee has gone up. The hyperscalers still own the infrastructure game, but they no longer get to call themselves cheap just for showing up to play.
Explore more exclusive insights at nextfin.ai.
