Microsoft’s intelligent cloud grew 32%, while its property, plant and equipment rose from $109 billion to $313 billion. Which number matters more to shareholders?
Growth used to feel like proof. Today, it is only the beginning of the analysis.
AI demand is visible across cloud revenue, infrastructure spending, contracted backlog, and customer commitments. The investment question is whether the companies financing this buildout can earn durable returns after depreciation, leases, purchase commitments, financing costs, competition, and potentially falling compute prices.
Growth Is No Longer Proof of Value
Amazon spent $170 billion over the last 12 months. Meta plans to spend between $130 billion and $145 billion. These figures demonstrate capital commitment, not value creation.
Aswath Damodaran says marginal returns on invested capital have fallen sharply at Meta, Alphabet, and Microsoft. Unless earnings become commensurate with the tens of billions invested, these companies may become more capital intensive businesses earning lower returns on invested capital.
That is the central tension in AI investing: strong cloud growth can coexist with deteriorating economics.
Revenue may arrive before the full burden of depreciation, leases, purchase commitments, or financing costs becomes visible. Poor returns are not inevitable, but revenue growth alone is no longer enough to justify valuing hyperscalers as the lean, high return businesses investors previously owned.
A falling stock price does not automatically prove that AI demand is weakening. Likewise, rising revenue does not automatically prove that AI infrastructure is earning an attractive return.
Nebius Proves Demand, Not Returns
Nebius generated $399 million in Q1 2026 revenue, up 684% year over year, with contracted backlog approaching $50 billion. Meta committed up to $27 billion over five years, including $12 billion for dedicated Vera Rubin capacity and a $15 billion backstop for capacity Nebius cannot sell elsewhere.
Those commitments are powerful evidence of demand. The backstop may also reduce near term risk around selling that capacity.
But backlog is not profit, and contracted revenue is not an attractive return on invested capital. Nebius still has to build and operate the infrastructure, then convert that scale into attractive margins.
Customer concentration adds another layer. Meta’s commitment provides visibility, but it also ties part of the economics to one major customer and to the terms under which capacity gets financed, used, and ultimately repriced.
The $15 billion backstop is especially revealing. It protects Nebius against capacity it cannot sell elsewhere, but its existence suggests utilization risk is important enough to require contractual protection.
The Winners Will Be Chosen by Incremental Returns
The next phase of the AI investment case will be decided through economics, not announcements.
Investors need to distinguish revenue growth from the incremental returns on the capital producing it. That means watching how quickly the infrastructure depreciates, who finances it, how much capacity remains productive, and what happens if customers eventually demand the same capability for 10% of today’s cost.
A nearly $50 billion backlog may become the foundation for substantial earnings. Microsoft’s 32% intelligent cloud growth may justify the expansion in property, plant and equipment. The current evidence does not resolve either question.
AI demand can be real without every provider creating durable shareholder value. The winners will be those that earn attractive returns on capacity, not those announcing the largest spending plans or backlogs.