Earnings Season: Capacity Absorption and the Cash-Conversion Divergence

This review covers Microsoft, Alphabet, Meta, Amazon, Intel, Tesla and SK hynix. Vertiv and Arm remain not-determined because no verifiable official earnings release was available at cut-off; no conclusions are drawn for them. All financial figures return to official company disclosures.

Three cross-company signals: cloud revenue absorbing capacity, a persistent supply gap, and diverging cash conversion. Together they show that total capex is no longer the only variable; delivery speed and free cash flow now determine investment returns.

Company-level depth appears in the published Microsoft, SK hynix and capex columns; this review focuses on the seasonal read and does not repeat them.

Placing the three cloud growth rates side by side first answers an aggregate question: has expanding investment killed demand? It has not. The central observation of the season is that demand and supply are expanding together, and the gap has not been closed by capex.

Cloud revenue is absorbing new capacity

Azure grew 43%, Google Cloud 82% and AWS 37%, its fastest growth in 18 quarters. Demand has not visibly weakened as infrastructure spending expanded; new capacity is still being absorbed.

The revenue structures differ: Azure's 43% sits on the largest absolute base, Google Cloud's 82% comes from a lower base plus backlog conversion, and AWS's 37% is the only acceleration of the three. The bases differ, but they point to the same fact: billable capacity is not sitting idle.

Revenue absorption is not return realization. Cloud revenue growth precedes the depreciation and energy cost peak, so the real test is whether gross margin and free cash flow improve after capacity comes online.

The supply gap has not closed

Microsoft still says demand exceeds supply. Management guides Azure constant-currency growth of about 45% next quarter with capex above $50 billion; the quarter added 31 data centers and about 1GW, and time from GPU arrival to deployment in large regions shrank nearly 50% during the fiscal year. The company expects overall capacity to roughly double within two years.

Alphabet's cloud backlog reached $514 billion, up about $50 billion quarter over quarter, with more than half expected to recognize within 24 months; the company will use third-party capacity as a bridge in Q3 and expects slight margin pressure. Backlog is a management disclosure, not recognized revenue, but it explains the durability behind 82% growth.

Amazon management expects capacity constraints through 2026 and likely 2027. Full-year company-wide cash capex guidance rose from about $200 billion to about $220 billion, mainly on memory costs; that is a company-wide figure, not AWS capex.

Meta is maximizing 2026 and 2027 capacity and laying groundwork for 2028 and beyond. API, enterprise agents and selling compute remain potential monetization paths, not reported revenue.

Across all four, the binding constraint in guidance has moved from whether capital exists to when capacity arrives. Capital budgets are no longer scarce; power, chips and data halls delivered on time are.

Cash conversion is diverging

Alphabet's quarterly free cash flow was negative $5.855 billion: operating cash flow of $39.069 billion against $44.924 billion of property and equipment purchases. Amazon's trailing-twelve-month free cash flow turned to an outflow of $7.6 billion. Meta posted only $784 million of quarterly free cash flow and narrowed full-year capex to $130-145 billion.

The three cash numbers come from different accounting structures: Alphabet is a quarterly purchase peak, Amazon is a twelve-month rolling measure, and Meta's capex includes finance-lease principal. Direct comparison is meaningless; the common signal is that capital spending has outrun operating cash flow, with the gap covered by debt and finance leases.

Microsoft's lease mix provides a scale for observation: of $41 billion in quarterly capex, $5.6 billion was finance leases and $35.8 billion cash purchases of property and equipment. Rising lease share means future rent obligations enter the balance sheet, structurally compressing free cash flow rather than merely reflecting investment timing.

Whether capex converts quickly into billable capacity has become more important than the budget total. Depreciation, energy and lease costs lag, and current free cash flow already carries the leading edge of future depreciation pressure.

Upstream does not benefit uniformly

Intel reported Q2 revenue of $16.1 billion, up 25%, with DCAI revenue of $6.3 billion, up 59%. The company will increase equipment, cleanroom and substrate investment to support product and foundry growth; the revenue gain comes from the data-center CPU cycle, while foundry returns still need proof.

SK hynix reported Q2 revenue of 79.32 trillion KRW and operating profit of 60.54 trillion KRW, an operating margin of 76%. HBM4 began shipping in Q2 with an H2 ramp planned and HBM4E samples sent in H1; roughly ten key customers use long-term multi-year supply arrangements, M15X is accelerating, and the Yongin phase-1 cleanroom is expected early 2027.

Tesla deployed 13.5GWh of storage in Q2. AI data centers raise the strategic value of grid-scale storage for power quality, peaking and interconnection, but Tesla does not disclose how much serves data centers, so not all storage growth can be attributed to AI.

Vertiv and Arm are marked not-determined: no verifiable official Q2 release was available at cut-off, so no orders, backlog, liquid-cooling revenue, licensing or full-year guidance conclusions are drawn. Update requires a verifiable release, 10-Q or stable call transcript.

Upstream follows a common pattern: revenue arrives before capacity. Intel and SK hynix orders and shipments are already in profit, but cleanrooms, equipment, packaging and yield decide next year's supply, the same constraint chain as the cloud capacity gap.

What would break the call

The constructive view assumes capacity lands on schedule and gets absorbed. The counter-case has three legs: owned capacity arrives late, forcing Alphabet and Amazon to rely on third-party bridging longer; memory and equipment costs rise faster than revenue conversion, compressing gross margin; and cloud growth slows after supply expands, producing a 2027 combination of capacity surplus and a depreciation peak.

Free cash flow is the warning line. If Alphabet, Amazon and Meta stay negative for two more quarters while billable capacity growth fails to catch up, capex is not converting into revenue and the story shifts from growth to cash burn.

Next quarter's watch list: energization and go-live of new capacity, the GPU arrival-to-activation cycle, the finance-lease share, free cash flow per unit of billable capacity, and whether Vertiv and Arm official releases fill their not-determined gaps.

For Atlas, this review is not a price forecast. It turns whether AI infrastructure is delivering into a set of verifiable engineering questions, each with official observation points: capacity, power, chips and cash.