When an AI company's spending is described as a 'trillion-dollar buildout,' the number under discussion is almost always capital expenditures — capex for short. In SEC filings, capex is not a marketing term; it is the cash a registrant pays to acquire or construct long-lived physical assets it expects to use for more than a year: land, buildings, datacenters, networking equipment, and the servers (including the graphics processing units, or GPUs) that run AI workloads. Because these assets are used over many years rather than consumed at once, their cost is capitalized on the balance sheet and depreciated over time, instead of being expensed all at once like a salary or a marketing campaign.

The most reliable place to find the figure is the consolidated statement of cash flows, in the section headed 'cash flows from investing activities.' There it usually appears as a line such as 'additions to property and equipment' or 'purchases related to property and equipment,' shown as a cash outflow in parentheses. In NVIDIA's annual report on Form 10-K for the fiscal year ended January 26, 2025, the investing-activities section reports 'purchases related to property and equipment and intangible assets' of $3,236 million, against $1,069 million the prior year. That single jump — more than tripling year over year — is the kind of movement that capex-watchers track across the AI buildout, and it traces to a specific, audited line in a filing rather than to a press release.

We will continue to invest in capital expenditures to support growth in our cloud offerings and our investments in AI infrastructure and training.— Microsoft Corporation, Form 10-K (fiscal year ended June 30, 2025), source

That sentence, drawn from Microsoft's most recent annual report, sits in the part of the filing the Securities and Exchange Commission requires management to write about its own numbers: the Management's Discussion and Analysis of Financial Condition and Results of Operations, commonly called MD&A. The disclosure obligation comes from Regulation S-K Item 303 (codified at 17 CFR 229.303). Item 303 directs a registrant to discuss its liquidity, its capital resources, and any known material commitments for capital expenditures — including the general purpose of those commitments and the expected source of funds to meet them.

Why the cash flow line and the MD&A line are different jobs

The two locations answer two different questions, and conflating them is a common error. The statement of cash flows reports capex that has already happened: cash that left the company during the period to buy or build assets. MD&A, by contrast, is forward-looking — it is where management explains what capital commitments exist going forward and how the company intends to fund them. Item 303's stated objective is to give investors 'material information relevant to an assessment of the financial condition and results of operations of the registrant including an evaluation of the amounts and certainty of cash flows from operations and from outside sources.' For an AI company financing a datacenter buildout, the certainty of future cash flows — operating cash, debt, leases — is exactly the question the rule is built to surface.

A second nuance matters for AI infrastructure specifically: not every dollar of computing capacity shows up as a clean capex line. Companies increasingly secure datacenter space and servers through finance leases and long-dated purchase commitments rather than outright purchases. Microsoft's 10-K, for example, separates its contractual obligations into construction commitments, operating and finance leases, and purchase commitments — each a distinct category that funds capacity in a different accounting form. Reading 'how much is this company spending on AI infrastructure' therefore requires reading past the single capex line into the leases and commitments disclosures, because the economic commitment can be larger than the cash-purchase figure alone.

What the rule does — and does not — require

Item 303 requires management to discuss material capital expenditures and commitments; it does not require a company to publish a precise multi-year capex forecast, and many issuers describe their plans in qualitative terms ('additions to property and equipment will continue') alongside whatever quantitative guidance they choose to provide. The rule also requires that MD&A focus on material events and uncertainties 'known to management that are reasonably likely to cause reported financial information not to be necessarily indicative of future operating results' — language that, read against an AI buildout, covers the risk that committed spend outruns the revenue it is meant to support.

It also helps to understand why capex is capitalized rather than expensed, because that accounting choice is what makes the AI buildout look the way it does on the financial statements. When a company buys a server or builds a datacenter, GAAP treats the cost as the acquisition of an asset with a useful life of several years, so the cash leaves immediately (and appears in investing activities) but the cost hits the income statement only gradually, as depreciation, over the asset's life. The consequence for an AI company spending heavily on short-lived computing hardware is that today's enormous cash outflow becomes tomorrow's rising depreciation expense — a cost that lands on the income statement for years after the cash is gone. That timing gap between when the cash is spent and when the cost is recognized is precisely the kind of matter MD&A is meant to illuminate, which is why capex disclosure and the depreciation that follows it are read together rather than in isolation.

For a reader trying to understand an AI company's capital intensity, the practical method follows the documents in order. Start with the investing-activities section of the statement of cash flows for the historical capex figure. Move to the MD&A liquidity and capital-resources discussion for management's forward characterization under Item 303. Then check the commitments and contingencies note and the contractual-obligations table for leases and purchase commitments that fund capacity outside the purchase line. Each of those is an audited or rule-mandated disclosure with a fixed location in the filing — which is why capex, despite being the headline number of the AI era, is one of the more disciplined figures to source: it does not live in a slide deck, it lives at 17 CFR 229.303 and on a line of the cash flow statement.