Big Tech Earnings Tested by $725 Billion AI Capex Splurge

During the final week of April 2026, five of the so-called Magnificent Seven reported earnings within a 72-hour window. Alphabet shares rose. Microsoft and Meta fell. Amazon traded sideways after an initial pop. The aggregate market capitalisation shift across those four names exceeded $300 billion in a single trading week, a magnitude rarely seen outside of crisis periods. The common thread was not revenue growth, which was broadly strong. It was capital expenditure. When Zacks reported that Big Tech's 2026 capex estimates had hit $725 billion, the number was no longer an analyst projection. It had become the earnings-season story.

The $725 billion figure consolidates planned spending across Microsoft, Amazon, Alphabet, and Meta, the four US hyperscalers driving the AI infrastructure buildout. For context, that sum exceeds the annual GDP of Switzerland. It represents a year-over-year increase of roughly 40 percent from the approximately $520 billion the same firms allocated in 2025. Much of it flows to NVIDIA GPUs, custom silicon, data-centre construction, and the energy infrastructure required to power racks that can draw 120 kilowatts apiece. The scale has no precedent in enterprise technology. Even the dot-com era's fibre-optic overbuild, which left millions of miles of dark fibre across North America, was a smaller capital commitment relative to the cash flows of the firms involved.

What changed in late April was not the size of the spending but the market's tolerance for it. Alphabet, which reported first-quarter 2026 results on April 29, saw Google Cloud revenue rise 63 percent year over year, outpacing both Microsoft Azure and Amazon Web Services on a growth-rate basis. The company's search advertising business, its core profit engine, also beat estimates. Investors bid Alphabet shares higher because the AI narrative had an identifiable revenue line item attached to it. The same could not be said for Microsoft, which disclosed a $190 billion multi-year capex commitment that Forbes contributor Peter Cohan characterised as a "capex shock" that overshadowed otherwise strong results.

Microsoft's AI revenue, reported as a $37 billion annualised run rate according to data cited by 24/7 Wall St in a June analysis of Bill Ackman's portfolio positioning, is a genuine number. But it is a forward-looking one, and the market in late April stopped giving forward-looking numbers the benefit of the doubt. The shift was subtle but consequential. Investors stopped asking "how big is the AI opportunity" and started asking "how much of it is already priced into the current multiple, and how much more capital will be consumed before the free-cash-flow line bends upward."

Meta's predicament is the most instructive case study in this new regime. The company guided capital expenditures of up to $145 billion for 2026, according to MSN's reporting on its earnings, nearly all of it directed toward AI data centres and the compute infrastructure to support its open-source Llama model family. Meta's core advertising business grew at a healthy clip. But the capex number, paired with a fresh round of layoffs, told a story that equity markets found difficult to price. Meta was reducing its workforce while simultaneously committing more capital to infrastructure than it had ever spent in a single year. The efficiency narrative that had supported the stock through 2024 and 2025 collided with an investment cycle that looked, from the outside, like a strategic gamble with an uncertain payoff timeline.

The layoff numbers in isolation are striking. American technology companies eliminated more than 142,000 jobs in the first five months of 2026, a 33 percent increase over the same period in 2025, Tech Times reported, citing data tracked through May. The same employers posted record revenues. The rationale, articulated across earnings calls, was uniform: operating-expense savings from headcount reductions were being redirected toward capital expenditure on AI infrastructure. This is not, as a matter of accounting, an even trade. A dollar saved on a software engineer's salary is a dollar that flows through the income statement once. A dollar spent on a GPU cluster becomes a depreciating asset on the balance sheet, amortised over four to six years, and it requires additional spending on power, cooling, and networking to function. The substitution changes the shape of the financial statements.

Amazon's numbers put the cash-flow dynamic in sharp relief. The company's free cash flow fell to roughly $1.2 billion, as tracked in a May analysis by 24/7 Wall St, down from levels that had historically run in the tens of billions. Amazon Web Services revenue growth accelerated, driven in part by multi-year AI compute deals with OpenAI and Anthropic. But the capital required to provision that compute consumed nearly all of the cash the business generated. When a company of Amazon's scale sees free cash flow approach zero, the market begins to ask whether capex is investment or whether it is simply the cost of staying in the game.

The Bottleneck the $725 Billion Number Obscures

The aggregate spending figure captures chips, buildings, and power. It does not capture the constraint that several industry analysts now identify as the real bottleneck: the supply of transformers, switchgear, and high-voltage electrical equipment needed to connect multi-gigawatt data-centre campuses to the grid. Forbes, syndicated through Yahoo Finance, reported that lead times for large power transformers have stretched beyond three years in some North American markets, and that the domestic manufacturing capacity for medium-voltage switchgear is fully booked through 2028. A data centre with GPUs installed but not powered is a stranded asset. The hyperscalers are competing not only with each other for electrical infrastructure but with electric-vehicle charging networks, grid modernisation projects, and industrial electrification. The capital is committed, but the physical assets it is meant to purchase may not arrive on schedule.

This is the capital-allocation question that the press releases do not address. When a company announces $145 billion in planned capex, the market assumes that the spending will occur roughly as scheduled and that the assets will be productive within a reasonable window. If transformer and switchgear lead times push data-centre commissioning dates from 2027 to 2029, the net present value of that capital changes materially. Depreciation begins when the asset is placed in service, not when the purchase order is signed. The gap between commitment and commissioning is where returns get diluted.

Goldman Sachs published a note in early June arguing that the AI spending cycle would be larger and longer than consensus expectations, according to a summary carried by MSN. The bank's reasoning rested on the observation that enterprise adoption of generative AI workloads had not yet reached an inflection point and that when it did, the incremental demand for inference compute would sustain hyperscaler utilisation rates at levels that justified the current buildout. It is a coherent argument, and it may prove correct. But it is an argument about demand in 2028, not demand in the current quarter. The equity market's time horizon, particularly during earnings season, is considerably shorter.

The macroeconomic overlay adds another dimension. The Federal Reserve's interest-rate decision in late April landed in the same week as the mega-cap earnings flood, a scheduling coincidence that BeInCrypto noted via Yahoo Finance created a compressed repricing window across both equities and risk assets. Higher-for-longer rate assumptions raise the discount rate applied to future cash flows, which disproportionately affects companies that are spending heavily today for returns that sit several years out. The AI capex cycle is, by its nature, long-duration. Rising real yields compress the present value of those distant returns. It is a mechanical relationship, and it operated with particular force during the late-April earnings window.

The market's differentiation between the hyperscalers is becoming more granular with each earnings cycle. Barchart's earnings preview for that same April week flagged five Mag 7 names reporting, noting that the dispersion of outcomes was widening. Alphabet's cloud acceleration and TPU monetisation strategy gave investors something concrete to model. Microsoft's Copilot attach rates and Azure AI workload growth were strong but came packaged with a capex trajectory that made discounted-cash-flow models harder to anchor. Meta's open-source strategy had no obvious revenue line item attached to it, and its advertising business, while robust, did not benefit from the same AI-driven pricing tailwind that investors could see in Alphabet's search segment.

NVIDIA remains the clearest beneficiary of the capex cycle regardless of which hyperscaler strategy prevails. The company's data-centre revenue continues to track aggregate hyperscaler capex with a correlation coefficient that analysts at several sell-side firms have placed above 0.9. But even NVIDIA is not immune to the cycle's downstream effects. If lead times for electrical infrastructure delay data-centre commissioning, GPU purchase orders may stretch across a longer delivery timeline, flattening the revenue recognition curve. The chip designer's own forecast commentary has begun to acknowledge this timing risk, albeit in the measured language of supply-chain management rather than the language of demand destruction.

The memory market tells a related story. Micron's stock surged 865 percent over the trailing twelve months through late May, as 24/7 Wall St reported, driven by high-bandwidth memory demand tied directly to GPU deployments. HBM supply is pre-sold through the end of 2026. The pricing power is extraordinary. But it also means that every incremental GPU cluster ordered by a hyperscaler carries a memory cost that is fixed and elevated. Gross margins on AI infrastructure accrue to the component suppliers before they accrue to the companies operating the infrastructure. That distribution of margin across the value chain is another element the aggregate $725 billion figure does not reveal.

What the Next Two Quarters Will Test

The earnings reports due in July and October 2026 will test a proposition that has become the central debate in tech investing: whether the AI infrastructure buildout represents a genuine capital-deployment cycle with measurable returns or a competitive arms race where the only clear winners are the arms dealers. The distinction matters because the two interpretations imply very different terminal valuations for the firms doing the spending. If the infrastructure produces differentiated AI services with pricing power, the capex is investment in future earnings streams. If the infrastructure produces commodity compute that any hyperscaler can provision, the capex is a cost of maintaining market position, and it will not generate excess returns above the cost of capital.

The next test will be whether the revenue lines that investors rewarded in Alphabet's first quarter begin to appear with similar clarity in Microsoft's and Amazon's second-quarter results. Microsoft's AI revenue run rate of $37 billion is a large number, but the composition matters. If it is concentrated in infrastructure-as-a-service compute rather than higher-margin platform and application layers, the margin structure of the AI business will look more like AWS circa 2016 than like the software franchises that gave Microsoft its current multiple. Analysts at several sell-side firms, including Goldman Sachs, have begun to disaggregate AI revenue into infrastructure and application segments in their models, a level of granularity that was not common even two quarters ago.

The capital-markets dimension of the story is also evolving. Alphabet's reported exploration of an $80 billion equity raise for AI infrastructure, covered by Memeburn in early June, suggests that even the best-capitalised firms are assessing whether organic cash flow can sustain the planned spending trajectory. If equity issuance becomes a funding mechanism for AI capex, existing shareholders will be diluted to finance assets that may take years to generate returns. That calculus changes the investment case for the entire sector. It also raises the question of whether debt markets, which have been accommodating, will continue to price hyperscaler paper at investment-grade spreads if free cash flow remains compressed.

The most important number to watch in the coming earnings cycle is not a revenue figure. It is the ratio of capital expenditure to operating cash flow, calculated on a trailing twelve-month basis and compared quarter over quarter. When that ratio exceeds one, a company is funding its investment activity with external capital or balance-sheet drawdowns rather than internally generated cash. That is sustainable for a period, particularly at low interest rates. It is less sustainable when rates are elevated and when the assets being acquired have uncertain productivity. The market, in its repricing of Microsoft and Meta in late April, signalled that it has begun to track this ratio closely. The July earnings season will show whether management teams have adjusted their guidance accordingly, or whether the spending trajectory remains locked in regardless of what the market is saying.

The AI capex cycle is not a bubble in the traditional sense. The capital is being deployed toward real assets with genuine utility. But a cycle does not need to be a bubble to produce misallocation. The fibre-optic overbuild of the late 1990s was not a fiction; the fibre was real, and it eventually found productive use. The capital that financed it, however, was largely written off by the equity and debt holders who provided it, while the assets were acquired at cents on the dollar by the firms that operate the internet backbone today. The question for the current cycle is not whether the data centres being built will be useful. It is whether the shareholders funding their construction will earn an adequate return on the capital they have committed. The earnings reports over the next six months will begin to provide an answer.