The latest technology earnings show substantial infrastructure investment, but the word capex does not describe a single payment event. A server bought for cash, a lease beginning today and a principal instalment on an older lease can all appear in the discussion. Treating those measures as interchangeable obscures when the funding burden arrives.
The distinction matters to the debate over the AI buildout. Spending can support future capacity while weakening current cash conversion, yet the headline alone establishes neither a profitable future nor a failed investment. The first analytical task is to align the dates, definitions and businesses behind the figures.
The same quarter appears under different fiscal labels
Microsoft’s fiscal first-quarter release covers the three months ended 30 September 2026. That aligns with the calendar third-quarter reports of Amazon, Alphabet and Meta, despite the different fiscal-year label. Comparisons against annual guidance or trailing-year cash flow would answer a different question.
Amazon’s filing reports $75.9 billion of cash capital expenditure for that quarter. Its cash-flow statement separately shows $77.259 billion of property and equipment purchases and $1.381 billion of proceeds from sales and incentives. The net is approximately the smaller figure. Gross purchases and a net cash measure therefore should not be silently substituted for each other.
Amazon also says the investment supports technology infrastructure, predominantly AWS growth, and fulfillment capacity. It is not a clean disclosure of spending exclusively on AI. Its October earnings release retained approximately $220 billion of expected cash capex for 2026. That annual outlook is a forecast, not money already spent during the September quarter.
A lease can enter the comparison before its cash leaves
On its earnings call, Microsoft explained that its capex presentation recognises the full principal value of finance leases when the lease begins. It separately identified $39.4 billion of cash paid for property and equipment. Recognition of an asset commitment and payment for equipment are different clocks, even when both support the same infrastructure strategy.
Meta’s release uses another relevant convention: its $34.91 billion quarterly capital-expenditure figure includes principal payments on finance leases. Those are cash payments associated with financing arrangements, rather than the entire value of every lease newly commencing that quarter. Its $135 billion–$145 billion full-year outlook uses that stated inclusion too.
The practical implication is that a comparison must specify whether it measures cash paid now, infrastructure acquired or obligations recognised for future payment. Adding lease principal to a figure that already includes it would double-count the same payment. Comparing a commencement measure with current instalments would mix timing. Neither error is repaired by placing every number under an AI-spending heading.
The cash-flow deficit has a defined perimeter
Alphabet’s reconciliation makes one current cash burden explicit. September-quarter operating cash flow of $42.290 billion less property and equipment purchases of $51.355 billion produced negative free cash flow of $9.065 billion. This is the company’s disclosed non-GAAP liquidity measure, derived from identified cash-flow lines.
It shows that those purchases exceeded operating cash generated during the quarter. It does not mean Alphabet reported an equivalent accounting loss, that every dollar of investment was an AI expense or that the company could not obtain financing. Purchases of financial investments and acquisitions occupy other lines in the statement and should not be smuggled into this particular subtraction.
The economic question then becomes whether productive assets can generate sufficient future cash to justify the resources committed. Current free cash flow is evidence about today’s funding requirement. Estimating an investment’s eventual return additionally requires utilisation, operating costs, replacement needs and the durability of revenue. A one-quarter deficit supplies none of those answers automatically.
Infrastructure spending must eventually meet useful work
Microsoft attributed part of the quarter’s investment timing to delayed third-party capacity deliveries. That illustrates why a lower or later reported outlay cannot, on its own, establish that the strategic spending programme has ended. A delivery shift can move expenditure between periods without demonstrating that demand or the final funding requirement has disappeared.
There is a credible positive case for heavy investment: capacity could serve demand that existing infrastructure cannot meet and improve the provider’s competitive position. The opposing case is that assets may arrive ahead of economical demand, or require replacement before the expected cash contribution is achieved. These are scenarios to test, not conclusions supplied by the size of a spending plan.
Evidence that would distinguish them includes consistent cash and lease disclosures, delivery milestones, utilisation and the conversion of customer activity into cash after operating and replacement costs. Until those observations mature, the clearest comparison keeps current payments separate from future obligations. The buildout is real; a sector-wide return on it remains something to demonstrate rather than infer from a capex headline.