Microsoft’s AI infrastructure race hit an unexpected pothole when the company reported quarterly capital expenditure and finance leases of $45.2 billion—up roughly 30% year on year, but below analysts’ $50.7 billion expectation.¹

The immediate market reading was blunt. CoreWeave and Nebius each fell nearly 5%, while IREN dropped about 3%, as investors treated Microsoft’s lighter spend as a warning that the companies supplying overflow computing capacity could face slower growth.¹ That anxiety was already amplified by higher rates and tighter lending conditions for neoclouds, which need enormous external financing to build data centres.

Microsoft’s explanation, however, was not a retreat from AI. Chief Financial Officer Amy Hood said delays in third-party GPU capacity would push some availability into the first half of 2027.² The company reduced its 2026 capex forecast to $170 billion from $175 billion, but expects spending to accelerate as the delayed capacity arrives.³

The operating numbers strengthen that argument. Azure revenue climbed 48% year on year, and Microsoft said growth exceeded capex because of “timing differences between receipt of goods and payment and a lower mix of finance leases.”¹ Satya Nadella likewise argued that Microsoft can shift its “fungible fleet” toward the highest-value uses while bringing capacity online faster.Satya Nadella (satyanadella)⁴

Still, the episode exposed the divide between hyperscalers and their infrastructure partners. Microsoft can delay spending and redirect workloads; neoclouds must keep borrowing and building. Gartner has predicted that 80% of neocloud companies could be acquired or disappear by 2030, while Elon Musk has warned that debt markets have largely closed to companies without substantial revenue beyond AI data centres.¹

For now, the evidence points to a bottleneck rather than an AI demand collapse. But the market’s reaction shows that, in this boom, delivery schedules and balance sheets may matter almost as much as chips.