AI Capex Hits $715B Annualized as Big Tech Free Cash Flow Goes Negative

In the first quarter of 2026, Amazon, Alphabet, Microsoft, and Meta together reported capital expenditures of approximately $175 billion, a figure that annualizes to the $715 billion combined infrastructure spend the four hyperscalers have guided toward for the full year. The number, reported by Blockonomi after the Q1 earnings deluge, nearly doubles the prior year's total. It is larger than the annual GDP of all but roughly twenty countries. And it arrived in the same two-week earnings window during which the same four companies disclosed that their free cash flow, in aggregate, had turned sharply negative for the first time since the early pandemic era.

The tension now animating every quarterly earnings call is not whether artificial intelligence will be transformative. It is whether the infrastructure buildout required to serve it, chiefly data centers filled with NVIDIA GPUs and custom silicon, can generate a return on invested capital before the companies funding it exhaust their balance-sheet capacity. The Q1 2026 results, released between April 28 and May 1, gave investors partial answers: revenue from AI-related cloud services is growing, in some cases at triple-digit year-over-year rates, but capital spending is growing faster still. The gap between those two curves has become the central fact of the current tech earnings cycle.

That gap prompted the cognitive scientist and AI critic Gary Marcus to call the current spending wave "the greatest capital misallocation in history," a phrase he deployed in an interview published by Business Insider on April 30. Marcus, an emeritus professor at NYU who has spent years arguing that large language models are less capable than their promoters claim, pointed to the disconnect between the hundreds of billions being spent on training infrastructure and the relatively narrow set of tasks at which current AI systems genuinely excel. His phrasing was extreme, but it captured something real: across the four hyperscalers, depreciation schedules are lengthening, and the gap between capex and operating income, a rough proxy for the strain on current earnings, is wider than at any point in the last decade.

The Seeking Alpha analysis "The AI Arms Race: Running On Fumes And Borrowed Money," published May 7, documented the balance-sheet machinery now at work. The four companies, traditionally characterized by net cash positions and prodigious free cash flow generation, have become net debt issuers in aggregate. Short-term free cash flow compression is expected to persist through the remainder of 2026, the analysis noted, before a projected recovery toward 2028 as AI investments mature. The timeline matters: two years of negligible free cash flow is manageable for firms with the credit ratings of Microsoft and Alphabet. Three years begins to invite questions about capital allocation priorities that management teams have so far been able to defer.

Debt issuance across the group accelerated in the first quarter. Alphabet alone issued $25 billion in new notes across multiple maturities in January, its largest single debt offering on record. Microsoft's total debt outstanding crossed $100 billion for the first time. These are not distressed borrowers; Microsoft's five-year credit default swap spreads remain below 30 basis points. But the direction of travel is unmistakable. In prior capex cycles, including the 4G rollout of the early 2010s and the initial cloud buildout of 2016-2019, free cash flow dipped but never turned structurally negative across the entire cohort for consecutive quarters. The current cycle has broken that pattern.

A separate analysis published on Seeking Alpha, "It's All About The 2027 AI Capex, And It Doesn't Look Pretty," zeroed in on the trajectory beyond the current year. Its central projection: AI capex growth, running above 100 percent year-over-year in 2026, decelerates sharply to a range of 15 to 30 percent in 2027. The implication for equity multiples is direct. If the market is currently pricing these stocks on the assumption that the infrastructure build will continue compounding at its current rate, then a deceleration to even 25 percent growth represents a sharp repricing event. The analysis framed 2026 as the peak not only of spending but of the narrative that unlimited spending will be tolerated.

The counterargument, advanced by the companies themselves and by a number of sell-side analysts, is that revenue is catching up. Alphabet reported that its Google Cloud revenue grew 34 percent year-over-year in Q1, with AI-related services contributing an increasing share. Microsoft's Azure growth accelerated sequentially, and CFO Amy Hood told analysts that AI services alone contributed roughly 15 percentage points of Azure's growth rate. Meta, for its part, argued that its AI infrastructure is directly improving ad-targeting models in ways that are visible in cost-per-action metrics for advertisers. These are real revenue lines, not speculative futures. The question is one of proportion: whether $715 billion in annualized spending can be justified by revenue streams that, while growing fast, remain a fraction of the total addressable market that the spending implies.

The market has already begun discriminating. A Fortune report published during earnings week noted that only Google convinced investors the spending is paying off. Alphabet shares rose on its results; Microsoft and Meta shares declined, in Meta's case by more than 8 percent in the two sessions following its report. The differential was not about the absolute level of spending, which rose at all three firms. It was about the visible link between infrastructure investment and revenue line items. Google's search and cloud businesses showed accelerating growth that could be at least partly attributed to AI enhancements. Microsoft's Copilot revenue, while growing, remained difficult to disentangle from broader Office 365 subscription trends.

The Apple Exception

Apple, which reported in the same window, provided the most instructive contrast. The company's capital expenditures for the quarter were a fraction of the hyperscaler average, and its AI strategy remains focused on on-device processing and privacy-preserving cloud queries that do not require the same infrastructure intensity. Apple's gross margins expanded year-over-year. Its free cash flow remained robust. The divergence between Apple's capital allocation and that of the other four members of the trillion-dollar club is now so wide that it effectively divides Big Tech into two categories: companies building AI infrastructure for rent, and companies buying AI inference by the token. The market has noticed: Apple's price-to-free-cash-flow multiple has compressed far less than those of the hyperscalers over the trailing twelve months.

The macro context intensifies the pressure. Hedge fund manager Paul Tudor Jones, cited in a Seeking Alpha opinion piece published in early May, has argued that the sheer scale of Big Tech's capex is contributing to a broader liquidity squeeze in capital markets. The reasoning is straightforward: when four of the world's largest corporate borrowers simultaneously pivot from returning capital to shareholders toward absorbing it for infrastructure, the effect on the corporate bond market and on the availability of long-duration project financing is material. Yields on investment-grade corporate debt have risen not because credit quality has deteriorated but because supply has expanded faster than demand at the long end of the curve.

There is an argument, heard increasingly from buy-side portfolio managers at long-only funds, that the infrastructure being built will retain value even if AI enthusiasm moderates. Data centers with modern power and cooling specifications can be repurposed. GPU clusters can be redeployed for traditional high-performance computing workloads. The physical assets do not vanish if the AI revenue thesis weakens. This argument has merit, but it has limits. The premium these companies are paying for accelerated delivery, for scarce power-purchase agreements, and for proximity to fiber interconnects reflects the urgency of an arms race. If the urgency fades, so does the premium. The replacement-cost floor for a data center built on an 18-month timeline at peak construction pricing is lower than the book value at which it will be carried.

The margin structure of the AI business itself remains unresolved. Inference, the process of running trained models to generate outputs, is a lower-margin activity than training. As the industry shifts from building models to serving them, the revenue mix should shift toward inference, which means the unit economics of the hyperscalers' AI businesses will face structural margin compression over time. Amazon's AWS, the most mature of the cloud platforms, reported stable operating margins in Q1 despite the capex surge, suggesting that some of the AI-related revenue is flowing through at reasonable incremental margins. But the mix shift is only beginning. The models being trained in 2026 will generate inference revenue in 2027 and 2028. The margin profile of that revenue is a known unknown.

What to Watch For

Three data points in the Q2 2026 earnings cycle, expected in late July, will determine whether the capex narrative holds or fractures. First, the hyperscalers' Q2 revenue growth rates, particularly in cloud and AI-specific line items, need to show sequential acceleration to justify the Q1 spending levels. Flat sequential growth would not be enough. Second, forward capex guidance for the second half of 2026 needs to remain within the ranges established in the Q1 calls; any upward revision would be received harshly by a market already testing the limits of its patience. Third, and most telling, depreciation and amortization expenses will begin to reflect the Q1 infrastructure additions. If D&A rises faster than AI-related gross profit, the earnings impact of the capex cycle will become impossible to ignore in GAAP net income, which remains the metric on which executive compensation is ultimately based.

The 2026 Q1 earnings season did not produce a capex capitulation. It produced a capex acceleration, accompanied by the first real evidence that revenue is responding. The market's reaction, selective and increasingly discriminating, suggests that the burden of proof has shifted. For two years, the hyperscalers were given the benefit of the doubt: spend now, demonstrate returns later. In the quarter ahead, the word "later" will need a date attached to it. The companies that can provide one, and that can point to a specific line item on a specific financial statement that moves in the right direction when capex rises, will keep their multiples. Those that cannot will begin to look less like infrastructure investors and more like speculators, a distinction that matters deeply to the institutional capital that ultimately funds the build.