AI infrastructure spending has reached the point where the size of the cheque is no longer the most interesting number. The next phase of the investment case is accounting and utilisation: how much depreciation arrives, how much free cash flow is absorbed, and how quickly new compute capacity produces revenue.
That distinction matters because a data centre can support economic activity for decades while servers and accelerators turn over much faster. Companies are therefore building assets with different useful lives inside the same capital programme.
The cash leaves before the revenue is fully visible
Microsoft said capital expenditures were $41 billion in its fiscal fourth quarter, with roughly two thirds directed to short-lived assets such as CPUs and GPUs and the remainder to longer-lived assets. Meta reported $31.08 billion of Q2 capex and expects $130 billion to $145 billion for 2026.
Those figures are not directly comparable measures of 'AI spending'. They are company capex disclosures that include different categories. What they have in common is timing: cash is committed before the full revenue associated with that capacity appears.
Suppliers are already seeing the demand
Nvidia reported $89 billion of data-centre revenue in its fiscal second quarter, more than double the prior year. Vertiv, which sells power and cooling infrastructure, raised full-year guidance after Q2 organic growth and strong pipeline demand.
That helps confirm the buildout is physical rather than rhetorical. Accelerators, power systems and cooling equipment are being ordered at scale. The remaining investor question is whether the customers buying those systems earn returns that justify repeating the cycle.
Depreciation will become more visible in earnings
Once infrastructure is placed into service, capital expenditure moves into the income statement gradually through depreciation. A company can therefore report strong operating growth while the economic cost of previous capex keeps rising.
This is not automatically negative. High utilisation can turn infrastructure into durable cash flow. The warning sign would be rising depreciation and financing needs without comparable growth in cloud, AI software or advertising revenue attributable to the extra capacity.
What to watch next
Investors should track capex together with depreciation, free cash flow, cloud backlog, data-centre utilisation and management commentary on supply constraints. The AI capex cycle remains powerful. The market is simply moving from asking who can spend to asking who can earn.