CALVERELL PERSPECTIVES

Reading AI demand through revenue, contracts and cash

· · 5 min read

Microsoft’s run rate and backlog show different aspects of demand. Neither measures the return on AI infrastructure, and Nvidia’s sales cannot supply the missing customer economics.

Microsoft said its AI business passed an annual revenue run rate of $37 billion, up 123% from a year earlier [2]. The release does not describe how that figure is built. That limits what it can support.

Three Microsoft figures, three different questions

A run rate extrapolates a recent measure. The excerpt does not say which period it draws on, how customer losses are treated, or whether it counts committed subscriptions or usage. Without those details it should not be read as recognised revenue for any completed period, and it is not cash received. Nor does the release say how the AI figure relates to Microsoft Cloud revenue of $54.5 billion for the quarter, up 29% [2]. These are separate categories, and neither should be added to, subtracted from or divided into the other.

Commercial remaining performance obligation rose 99% to $627 billion [2]. This is contracted revenue not yet recognised. Some of it may already have been paid for, so it cannot be described as a pile of future invoices. It also covers the whole commercial business, not only AI. What it supports is a narrow statement: signed commitments across Microsoft's commercial customers grew sharply. The excerpt does not give the AI share, when the revenue will be recognised, or the terms on which customers can cancel or reduce commitments.

Net income was $31.8 billion, up 23% on a GAAP basis and 20% on a non-GAAP basis, which excludes the impact of Microsoft's investments in OpenAI; the adjustment was $14 million this quarter and $583 million a year earlier [2]. Currency also moves the reported numbers. Revenue grew 18%, or 15% in constant currency, and GAAP net income grew 23%, or 20% in constant currency [2]. This bears on how to read any later slowdown. Lower net income would not by itself show weaker AI demand, because depreciation, other costs and currency all enter the result. The reverse holds too. Rising profit does not show that AI products are paying for themselves.

The segment lines are no more specific. Productivity and Business Processes revenue was $35.0 billion, up 17%, or 13% in constant currency. Within it, Microsoft 365 Commercial cloud revenue rose 19%, Microsoft 365 Consumer cloud revenue 33%, LinkedIn 12% and Dynamics 365 22% [2]. The disclosure excerpt does not attribute any of that growth to AI features, so these lines cannot be used to size an AI premium.

Supplier sales and what they leave open

Nvidia reported revenue of $81.6 billion, up 85% from a year ago, and Data Center revenue of $75.2 billion, up 92% [1]. Under its previous sub-markets, compute revenue was $60.4 billion, up 77%, and networking revenue was $14.8 billion, up 199% [1]. These figures show that buyers placed orders and that Nvidia recognised the revenue. They do not show that the buyers earn a return on the equipment. A supplier's sales and its customers' payback are separate questions, and the sources cited here address only the first.

The denominators also differ. Nvidia's new Hyperscale sub-market will include public clouds and the largest consumer internet companies, while ACIE covers AI clouds, industrial and enterprise buyers [1]. The excerpt provides no revenue for either group. Microsoft reports by its own categories. The populations overlap in part and are not identical, so Nvidia's Data Center revenue cannot be matched to Microsoft's AI run rate or to its Microsoft Cloud revenue.

Nvidia's $20.0 billion returned to shareholders in the quarter [1] concerns Nvidia's own distributions. No Nvidia cash-flow figures were supplied. Management's description of agentic AI as generating real value [1] is a statement from the company, and the excerpt does not provide customer-level measurement behind it.

Disclosures that would settle more

Several additions would turn inference into measurement:

Until then, these disclosures supports a limited reading: orders for hardware, contracts at Microsoft and a stated AI run rate all rose. It does not establish a single comparable measure of realised AI demand, isolate the AI share of commercial backlog, or show whether customers earn their money back.

At Microsoft’s next report, a completed-period AI revenue figure and a disclosed AI share of backlog would narrow the gap between commercial momentum and measurable demand. Customer retention, consumption and cash conversion would strengthen the case further. Without that bridge, the reported growth deserves weight, but attaching an investment return to it would go beyond what has been demonstrated.

Sources

[1] NVIDIA Corporation - NVIDIA Announces Financial Results for First Quarter Fiscal 2027

[2] FY26 Q3 - Press Releases - Investor Relations - Microsoft

Originally published at Ocean Capital Insights.