Microsoft and Amazon just proved the artificial intelligence boom has flipped from a demand story into a supply crisis, and the market loved it. Both cloud giants reported quarterly results that smashed expectations while delivering the same blunt message: customers want more AI compute than anyone can currently build, and the bottleneck is no longer software but silicon, power and data-centre space.
Azure growth hits a hardware ceiling
Microsoft said Azure revenue grew 39 percent year on year, pushing Microsoft Cloud past $51.5 billion for the first time. Management was explicit that the pace was limited not by customer appetite but by the availability of GPUs. Amazon echoed the constraint almost word for word. AWS revenue jumped 37 percent to $42.2 billion, its fastest expansion in more than four years, and the company warned that AI demand could exceed available cloud capacity through 2028. To keep up, Amazon lifted its 2026 capital expenditure plan to nearly $220 billion.
The market prices in the new reality
Investors responded by bidding Microsoft shares up more than 15 percent and Amazon 15.3 percent. Amazon also posted more than $62 billion in quarterly profit alongside 20 percent revenue growth. The moves suggest the street has stopped fretting about whether AI spending will ever pay off and started pricing in a multi-year infrastructure supercycle.
Analysts see the investment cycle turning
Ross Maxwell at VT Markets said the results show the AI investment cycle is beginning to deliver tangible financial returns, with strong revenue growth supporting the heavy capital outlay. Ponmudi R of Enrich Money argued the industry has entered a new phase where infrastructure spending is generating measurable commercial returns. Jefferies data backs the supply-side narrative: global data-centre demand is expected to reach 21.1 gigawatts in 2025 against only 8.9 gigawatts of operational capacity.
The next bottleneck is power and permitting
The implication is clear. The competitive frontier has shifted from model performance to the unglamorous work of securing land, grid connections and permitting for massive GPU clusters. Companies that cannot convert their capital budgets into live capacity fast enough will watch rivals capture the demand. For now, the constraint is the product.