AI Capex Splits Big Tech Earnings, 2027 Diverges From 2026

Alphabet reported first-quarter 2026 revenue of $109.90 billion on April 30, and Google Cloud's accelerating growth rate was the number that moved the stock nearly 10% higher in after-hours trading. The top line was strong, but what the market bought was the signal embedded in the cloud segment: AI workloads were converting into booked revenue, not just consumed compute. That conversion, from infrastructure spend to recognizable, recurring top-line growth, became the axis on which the entire Big Tech earnings cycle turned.

Over the same 48-hour window, Microsoft and Meta both reported results that sent their shares lower, despite capex figures that, in any earlier quarter of this cycle, would have been read as bullish conviction signals. Microsoft committed to a capital-expenditure trajectory that Business Insider reported would push combined Big Tech AI spending as high as $725 billion this year, while Meta's annual infrastructure budget approached $145 billion. The market yawned at the spending and flinched at the margins.

The pattern was not subtle. In Forbes, Peter Cohan described it as the moment investors divided the hyperscalers into 'winners and losers of the AI arms race', not by who spent the most, but by who could demonstrate that the spending was producing something the income statement could see. Alphabet surged. Amazon, which reported AWS revenue growth of 28% alongside a $465 billion contracted backlog, saw Wall Street analysts rush out price-target hikes to as high as $350, according to 24/7 Wall St. Microsoft and Meta fell.

The capital-expenditure numbers themselves remain staggering by any historical yardstick. The four hyperscalers, Alphabet, Amazon, Microsoft, and Meta, are on track to invest well over half a trillion dollars in 2026 alone, the majority of it directed at AI infrastructure: data centers, custom silicon, networking fabric, and the energy supply to power it all. The trajectory has drawn comparisons to the railroad boom, the fibre-optic overbuild, and, less charitably, the 1990s telecom capex cycle that left balance sheets wrecked when demand failed to materialize on schedule.

Big Tech's AI spending is the greatest capital misallocation in history., Gary Marcus, AI researcher and author, in a statement to Business Insider

Gary Marcus, the cognitive scientist and persistent AI-industry sceptic, used the earnings deluge to sharpen that comparison into a declarative sentence. In an interview with Business Insider, Marcus called the collective spending 'the greatest capital misallocation in history' and warned that returns on AI capex remained far below what the outlays implied. His argument is the most pointed version of a thesis that has gathered force among sell-side analysts and buy-side portfolio managers: that the hyperscalers are building capacity for a demand curve that has been modelled on exponential extrapolations, not observed adoption.

The counter-argument arrived almost simultaneously, and from a figure with considerably more money at stake. Paul Tudor Jones, the billionaire hedge-fund manager, told CNBC in early May that the AI bull market had 'another year or two to run' and that the market remained in 'the middle innings' of an AI-driven rally. Jones was not dismissing the capex question; he was arguing that the cycle still had room to accommodate it before capacity met demand and margins came under genuine pressure.

Between those two poles sits the analytical question that will define the next four quarters: at what point does AI infrastructure spending stop looking like investment and start looking like overcapacity? Goldman Sachs weighed in with a note, covered by MSN and multiple other outlets, warning that capital expenditures by the largest cloud providers would consume roughly 90% of their combined free cash flow in 2026. That is a ratio that leaves almost no room for the buybacks and dividends that have, for a decade, been the underwriting mechanism for Big Tech's equity valuations.

Goldman's math is worth pausing over. If a company generates $100 billion in free cash flow and spends $90 billion of it on servers, networking gear, and real estate, the residual available for shareholder returns is $10 billion. That is a dramatic restructuring of the capital-return profile that made these stocks the core of institutional portfolios through the 2010s and early 2020s. HSBC, in a separate note published earlier in the quarter, observed that Big Tech firms would, for the first time in years, spend more on capex than on dividends and buybacks combined.

The strain is visible in headcount decisions, too. Gadget Review, via Yahoo Finance, reported that Meta's $145 billion AI investment was accompanied by 8,000 job cuts, a trade-off that makes the capital-allocation hierarchy explicit. Personnel costs are being reduced so that infrastructure costs can rise. This is the mirror image of the 2015-2021 era, when headcount growth was treated as a growth signal in its own right and cloud margins were expanding steadily.

Amazon's position in this debate is the most structurally interesting because it carries the heaviest capex load, approximately $200 billion in planned AI-related spending, but also the most tangible offset. The $465 billion AWS contracted backlog, disclosed in its Q1 filing and analysed by Seeking Alpha, represents revenue that is contractually committed but not yet recognized. It functions as a partial hedge against the demand-uncertainty critique. If a significant fraction of that backlog converts over the next 12 to 18 months, the capex-to-revenue ratio looks considerably less alarming than the raw spending figures suggest.

What the backlog does not answer is the margin question. Contracted revenue is not the same as high-margin revenue. AWS's 28% growth rate in Q1 was impressive, but the composition of that growth, how much of it comes from AI inference workloads versus traditional cloud services, determines the margin profile. AI inference is more compute-intensive and may carry different unit economics than the storage and database services that have historically driven AWS profitability. The same question applies to Google Cloud Platform and Microsoft Azure.

Alphabet's Q1 outperformance was partly a function of differentiation. Google Cloud's full-stack AI approach, spanning custom TPU silicon, the Vertex AI platform, and the Gemini model family, allowed it to capture workload migration from enterprises that wanted a single-vendor AI stack. In a Zacks note published May 8, analysts pointed to that integrated approach as the factor that insulated Alphabet's cloud margins from the pricing pressure that affected competitors. If the capex cycle is a race, Alphabet is currently winning not because it is spending the most but because it is converting spend to revenue at the fastest rate.

The 2027 Capex Cliff

The most consequential number in the post-earnings analysis may not be any single quarter's spending figure but the forward curve. Seeking Alpha published an analysis projecting that AI capex growth, which has been running above 100% year-over-year through 2025 and 2026, is set to decelerate sharply, to between 15% and 30%, in 2027. If that projection holds, it marks a cyclical peak in dollar terms within the next twelve months, after which the growth rate of new investment falls toward the rate of depreciation. Multiples that have priced in perpetual hyper-growth in AI infrastructure spending would need to be recalibrated.

This is the analytical hook that connects the capex debate to equity valuations. Big Tech stocks have, as a group, traded at elevated multiples throughout the AI buildout because the market has treated infrastructure spending as a growth proxy. If spending growth decelerates, the growth-proxy argument weakens, and the market begins to demand proof of return on the capital already deployed. The transition from 'capex is growth' to 'capex must produce earnings' is the moment when capital-allocation discipline replaces ambition as the metric that drives share prices.

The Seeking Alpha analysis flagged a second risk: that the deceleration in capex growth coincides with a wave of depreciation from the assets already built. Data centers commissioned in 2025 and 2026 will begin hitting depreciation schedules in 2027 and 2028, creating a headwind for reported earnings at precisely the moment when revenue growth from AI workloads needs to accelerate to justify the asset base. It is a timing mismatch that corporate finance textbooks warn about and that equity markets often misprice until it appears in the quarterly numbers.

The liquidity-squeeze argument, advanced by Paul Tudor Jones and elaborated in a Seeking Alpha analysis, makes a related but distinct point. It holds that the sheer volume of capex flowing from Big Tech balance sheets into physical infrastructure, chips, buildings, power contracts, is absorbing capital that would otherwise circulate through equity buybacks and the broader financial system. The effect, in this view, is to drain liquidity from the very markets that have supported Big Tech's own valuations, creating a reflexive tightening that could amplify any downturn.

Meta's situation illustrates the liquidity trade-off at the company level. The $145 billion annual AI spend, combined with 8,000 job cuts, represents a deliberate reallocation from labour to capital, a bet that AI infrastructure will eventually substitute for human labour at scale. Whether that bet pays off depends on variables, model capability progress, enterprise adoption curves, regulatory constraints on AI deployment, that are not within Meta's control. The company is spending as though the answer is certain, which is what makes the capex question a capital-allocation question rather than merely a technology question.

What Capital Allocation Reveals That Press Releases Do Not

Read across the four hyperscalers' Q1 filings and a pattern emerges that the earnings calls were careful not to emphasize. Free cash flow at Microsoft and Meta declined year-over-year despite revenue growth, because the step-up in capex outpaced the incremental operating cash flow. Alphabet and Amazon managed to hold free cash flow roughly flat or slightly positive, but only because their cloud and advertising segments generated enough operating leverage to absorb the infrastructure spend. The distribution of outcomes is widening, and the companies with the weakest operating leverage are the ones whose capex commitment is most aggressive relative to their free cash flow generation.

The question that follows is whether the capex cycle will be self-correcting, whether the companies that are overspending relative to their revenue growth will be forced to slow down, either by their boards or by the bond market. The answer is not obvious. Meta and Microsoft both have balance sheets strong enough to sustain elevated capex for several more quarters, and both have strategic reasons, Meta's bet on AI as a replacement for labour, Microsoft's bet on AI as the operating system for the enterprise, to continue spending through a period of uncertain returns. Capital discipline, in this cycle, is not a binding constraint. The binding constraint is whether the demand side of the equation accelerates before the patience of public-market investors runs out.

The sell-side is divided in a way that reflects genuine analytical uncertainty rather than the usual distribution of bullish and bearish priors. Some analysts, particularly those covering Amazon and Alphabet, point to the contracted backlogs and accelerating cloud revenue as evidence that the capex is being absorbed. Others, especially those who cover the semiconductor supply chain, warn that the lead times for data-center components mean the spending committed in 2026 will not produce revenue-generating capacity until late 2027 at the earliest, a lag that creates a window of vulnerability if the macroeconomic environment deteriorates.

The 2027 outlook, as Seeking Alpha framed it, is where the capex cycle and the earnings cycle converge into a single, uncomfortable question: can the hyperscalers sustain earnings growth at a rate that justifies current multiples when the year-over-year growth rate of their largest expenditure category is falling by 70 percentage points? The answer will not come from a single quarter's results. It will come from the slope of the revenue line across the next four quarters, and from the margin disclosures that tell investors whether AI revenue is high-quality revenue or merely high-volume revenue.

Watch the Q2 2026 earnings calls in July for three numbers: the sequential change in each hyperscaler's cloud-segment operating margin, the ratio of capex to operating cash flow, and any revision to the forward capex guidance for the full year. A company that raises capex guidance while cloud margins are contracting is signalling something very different from a company that holds capex steady while margins expand. The market learned to read that difference in the Q1 cycle. It will grade with far less patience in Q2.