All topics

AI Infrastructure & Financing

Who pays for the build-out, how the bill reaches earnings, and what the disclosures can actually tell us.

The infrastructure behind AI has three different clocks. Cash is committed when equipment is ordered or financed. Accounting expense arrives as assets enter service and depreciate. Customer revenue arrives only if someone buys the resulting capacity. A convincing account of the build-out has to connect all three.

Our starting point is the financing chain: a supplier can participate in a customer’s funding as well as sell to it. That does not by itself establish circular revenue or bad demand. It does make the identity of the ultimate payer, the terms of the commitment, and the allocation of risk worth investigating.

The second question is measurement. Company-wide property and equipment purchases include activities beyond AI. Finance lease additions are different from this quarter’s lease payments. Guidance is an intention rather than spending already incurred. Our disclosure comparison keeps those boundaries visible instead of collapsing them into a league table.

The reading order below moves from financing to accounting, physical constraints, the allocation of grid costs, and competitive advantage. Our electricity analysis distinguishes payment for a connection from the cost of dependable generation and energy: assigning one bill does not settle the others. Use the depreciation scenario to isolate one accounting assumption, then return to the filings: a lower annual expense is not a lower purchase price, and neither tells us whether a new facility will earn its cost of capital.

Read in this order

Read the primary source

A capex number has a boundary

Source reading · September 20, 2026 · Microsoft FY2026 Q4 cash-flow statement

What the document reports

Microsoft’s cash-flow statement reports $35.802 billion of additions to property and equipment for the quarter ended June 30, 2026. This is a company-wide cash-flow item. Its fiscal-year total is $115.948 billion.

What follows

The cash measure provides a traceable starting point for assessing investment. It should be matched to the exact period and compared with earlier observations using the same definition. Financing arrangements and the timing of payment can separate cash expenditure from equipment becoming available.

What it does not establish

The line does not identify spending solely on AI, the amount of capacity put into service, or its eventual return. It also cannot be substituted for a management capex measure that includes finance lease additions.

Our reading

Start with the cash line, then reconcile the additional definitions rather than select whichever number produces the strongest headline. The disclosure comparison keeps these measurement boundaries beside the figures; the depreciation scenario isolates a different question: how capitalized cost reaches earnings.

Investigate the numbers

AI Infrastructure Spending

Read four companies’ infrastructure disclosures without confusing cash spending, leases, and forecasts.

Compute Depreciation Scenario

Compare the accounting cost of the same investment over two useful lives.

What changed

Follow the power bill beyond the connection

· Source

Who Pays for the Next Gigawatt? extends the dossier’s financing analysis to grid obligations. AEP Ohio’s minimum-demand framework addresses underused infrastructure, while PJM’s July and August filings propose ways to secure supply and manage large loads. We distinguish those proposals from effective tariffs and explain why a contribution to connection costs alone does not establish protection from every electricity-cost increase. The update adds this distinction to our analysis; it does not announce a new regulatory decision today.

Separate the spending measures

· Source

The dossier now separates company-wide cash property-and-equipment spending from lease payments and useful-life assumptions. This establishes the measurement baseline for subsequent reviews; it is not a new claim that all of those expenditures fund AI.