Capital Intensity Threatens Margins as AI Infrastructure Spending Surges

On April 30, Meta Platforms reported first-quarter 2026 revenue of $56.31 billion, a 33 percent year-over-year jump that beat Wall Street estimates by a comfortable margin. Net income came in at $26.8 billion. Then the company raised its full-year capital expenditure guidance to a range of $125 billion to $145 billion, a $10 billion increase at both ends from the figure management had given investors in January. The stock fell 9 percent in after-hours trading, as Zacks reported. The revenue beat and the earnings beat were not the story. The capex number was the story.

Meta is not alone. Within 48 hours of that earnings release, Microsoft confirmed roughly $190 billion in 2026 spending expectations, Amazon reported $44.2 billion in first-quarter capex alone, and Alphabet disclosed $35.67 billion in quarterly capital outlays with a cloud-services backlog exceeding $460 billion. The four firms, alongside Apple and a handful of others, now project combined 2026 capex of roughly $725 billion, up from approximately $670 billion before the April earnings cycle began, according to 24/7 Wall St. That figure exceeds the annual GDP of Switzerland.

What these headline numbers do not show, and what the sell-side research notes published in the days that followed mostly sidestepped, is the second-order question: what happens to margins when all this capital intensity begins to age. Every dollar spent on a server, a fabrication plant, a wind turbine, or a fibre-optic line becomes a depreciation charge that flows through the income statement for years. The bill for the current spending wave has not yet landed. When it does, it will arrive in a form that equity analysts have not priced with much precision.

The dynamic is not confined to Silicon Valley. A Financial Post report in February flagged capital intensity as a structural headwind for Canada's manufacturing sector, which had entered what the publication described as a deep recession. The piece noted that capital-intensive production models, while essential for global competitiveness in sectors such as automotive assembly, impose fixed-cost burdens that compress margins when demand softens. The same geometry applies to data centres, semiconductor fabs, and telecom networks. The dollars are larger in tech, but the arithmetic is universal.

To understand why depreciation matters so much right now, it helps to recall what the hyperscalers did between 2021 and 2024. Microsoft, Alphabet, and Amazon each extended the assumed useful lives of their server and networking equipment, typically from four years to six. The accounting change reduced annual depreciation expense by billions of dollars and boosted operating income without altering a single line of operational performance. In effect, management told investors that the hardware was lasting longer. The judgment may prove correct. But the extension also deferred a cost that must eventually be recognised.

The scale of the deferral is material. Alphabet disclosed in its 2024 annual filing that the server useful-life changes alone had reduced depreciation expense by approximately $3.9 billion that year. Across the four big spenders, the cumulative effect since 2021 likely exceeds $25 billion in depreciation that was not taken. Those are real dollars, representing real physical assets that are now several years older. When replacement cycles arrive, the capex required to refresh that installed base will land on top of the expansion capex already underway. The two streams will converge, and depreciation expense will rise even if the pace of new investment moderates.

The semiconductor supply chain is a concentrated lens through which to view this tension. Zacks reported in mid-May that Entegris, the Massachusetts-based materials and contamination-control supplier, is positioned at the intersection of rising materials intensity and the transition to 2-nanometre process nodes. Entegris management guided for a 2026 consumables rebound tied to higher wafer starts, alongside rising capex from chipmakers building advanced fabrication capacity. The 2nm ramp increases the volume of specialty chemicals, filters, and purification products required per wafer, which in turn raises Entegris's revenue per unit of industry output.

The first quarter was a solid start to the year as we continue to execute with focus and discipline against the constructive and improving semiconductor industry environment.David Reeder, CEO and President of Entegris, Q1 2026 earnings call

Reeder's language, delivered on the company's first-quarter 2026 earnings call, reflects an industry that sees demand materialising after a prolonged inventory correction. But the capital intensity of the 2nm transition is also a margin story for Entegris's customers. Leading-edge fabs now cost upwards of $30 billion apiece. The depreciation on those facilities, once they enter production, will be among the largest single line items on the foundries' income statements. TSMC, Samsung, and Intel are effectively pre-booking a multi-year margin headwind in exchange for future revenue growth that remains uncertain in both magnitude and timing.

Away from semiconductors, industrial equipment maker Astec Industries provides a parallel case. Seeking Alpha reported on May 6 that the company maintained its full-year 2026 adjusted EBITDA guidance of $170 million to $190 million, with capital expenditures projected between $40 million and $50 million, even as first-quarter margins came under pressure. Management held the line on the outlook, but the margin compression in Q1 illustrates the asymmetry that capital-intensive businesses face: the fixed-cost base does not flex downward when revenue softens. Depreciation is the largest and least-flexible component of that base.

Rogers Communications, reporting first-quarter 2026 results in April, offered a telecom variant of the same equation. The company posted continued year-over-year growth in total service revenue and adjusted EBITDA and upgraded its guidance for capital expenditures and free cash flow. Telecom is among the most capital-intensive sectors in the modern economy; Rogers's capex-to-revenue ratio runs well above 15 percent in a typical year. Every dollar of that spending eventually enters the depreciation schedule, where it weighs on reported margins for the useful life of the asset, which in telecom can stretch to 15 or 20 years for fibre and tower infrastructure.

The energy sector adds yet another layer to the capital-intensity story. ReNew Energy Global, the Indian renewable-power producer, reported fourth-quarter fiscal 2026 earnings of $0.02 per share on May 18, beating the consensus estimate of a $0.21 loss. The beat was driven by higher generation volumes and improved operational efficiency, but the company's balance sheet carries the depreciation burden of a large and growing portfolio of wind and solar assets. In renewable energy, depreciation is not an accounting abstraction. It is the mechanism by which the upfront capital cost of a turbine or a solar farm is allocated across a 25-year revenue stream, and any mismatch between the assumed useful life and actual asset performance flows directly to the bottom line.

What connects Meta's $145 billion capex plan, Entegris's 2nm materials opportunity, Astec's margin squeeze, Rogers's upgraded free cash flow guidance, and ReNew's depreciation-heavy asset base is a single economic force: the gap between when capital is deployed and when it is recovered. Capital intensity, measured as the stock of fixed assets relative to revenue or value added, has been rising across multiple sectors for the better part of a decade. The AI infrastructure buildout has accelerated the trend in technology. But the accounting treatment of that capital, and specifically the depreciation schedule that governs how it flows through the income statement, has not attracted nearly as much investor scrutiny as the top-line capex figures.

Part of the reason is that depreciation is a non-cash charge. It does not affect free cash flow directly, and the equity market has spent the past decade training itself to value companies on cash-flow metrics. But depreciation is not irrelevant. It reduces reported earnings, which affects price-to-earnings multiples, management compensation benchmarks, debt covenants, and the optics of profitability that shape analyst narratives. A company reporting 30 percent EBITDA margins can look very different on a net-income basis if depreciation consumes 12 percent of revenue rather than 6 percent. The gap between the two, for the hyperscalers, is widening.

The Depreciation Bill, Deferred

The accounting question that matters most for the 2027 and 2028 income statements is whether the hyperscalers will be forced to shorten their server useful-life assumptions. If AI workloads are driving faster hardware refresh cycles, as Nvidia's annual product cadence and the shift to liquid-cooled racks suggest, then servers purchased in 2024 and 2025 may need to be replaced sooner than the six-year depreciation schedules currently assume. A one-year reduction in the useful-life assumption across a $300 billion installed base of server equipment would add tens of billions of dollars in incremental annual depreciation expense. That would flow through the income statement as a direct reduction to operating income, with no corresponding cash outflow. The market would notice.

The semiconductor tooling sector faces a structurally similar question on a different timeline. The 2nm ramp that Entegris is betting on requires extreme ultraviolet lithography tools, atomic-layer deposition chambers, and advanced etching equipment that cost substantially more than their predecessors. The depreciation on those tools will be spread across the wafer output of the next five to seven years. If demand for 2nm chips materialises as the foundries project, the depreciation burden will be absorbed by high-margin revenue. If demand falls short, as it did during the 2023 inventory correction, the depreciation becomes a fixed cost on an underutilised asset base, and margins compress sharply.

The same logic holds for renewable energy, though the time scales are longer. ReNew's portfolio additions, which the company outlined on its fourth-quarter call, will add to the depreciation line for the next two decades. The revenue side depends on Indian power demand growth, regulatory tariff structures, and the absence of curtailment. Depreciation, by contrast, is fixed the moment the turbine begins spinning. The asymmetry is the defining financial characteristic of capital-intensive industries.

Investors have, for the moment, chosen to look past the depreciation question. The earnings beats delivered by Meta, Microsoft, Alphabet, and Amazon in April were strong enough that the sell-off in Meta shares appeared to be a reaction to the capex number itself rather than to any depreciation-related concern. But the two are linked. The $725 billion being deployed in 2026 will begin generating depreciation charges within months of each asset going into service. By the fourth quarter of 2027, the combined quarterly depreciation expense of the four largest hyperscalers will likely exceed $40 billion, up from roughly $25 billion in the first quarter of 2026. That is a $60 billion annual headwind to aggregate operating income before any change in useful-life assumptions. The math is straightforward.

What makes this moment different from previous capex cycles is the convergence of three forces: the absolute scale of the spending, the uncertainty around the useful life of AI-specific hardware, and the fact that the depreciation deferrals of 2021 through 2024 have compressed the timeline on which the bill comes due. A decade ago, when the cloud hyperscalers were building their first-generation data centre footprints, the capex was smaller, the hardware was replaced on a predictable four-year cycle, and depreciation expense was a manageable share of revenue. Today, the capex is an order of magnitude larger, the refresh cycle is unclear, and depreciation has been artificially suppressed by accounting changes that cannot be repeated.

There is no reason to expect a crisis. The hyperscalers generate enough free cash flow to absorb the depreciation, and their revenue growth rates remain well above the rate at which their fixed-asset bases are expanding. But the margin profile of the sector will shift, probably beginning in the second half of 2027, as the depreciation from the 2025 and 2026 capex vintages enters the income statement in full. The companies that have been most aggressive in extending useful lives, led by Alphabet and Microsoft, have the least room to absorb further accounting adjustments. When the depreciation bill arrives, it will arrive in cash terms that no accounting convention can soften. That is the checkpoint to watch.