Why are cloud growth and cash-flow pressure occurring at the same time?
- In the third quarter of 2026, AWS revenue rose 39% year over year to 46 billion dollars, and its operating margin rose from 34.6% a year earlier to 38.7%; Google Cloud revenue rose 103% to 30.7 billion dollars, and its margin rose from 23.7% to 36.2%. Revenue from Azure and other cloud services grew 48%.
- Google Cloud revenue includes sales of Workspace and TPU systems, so its growth rate cannot be directly compared with those of the other two. Nomura believes improving cloud profits can coexist with negative free cash flow: equipment investment is paid for in cash in the current year, while revenue and depreciation emerge gradually over the years after assets are put into use.
- When investment grows faster than cash is recovered from existing projects, the free-cash-flow gap widens; when it bottoms out and how quickly it recovers depend on investment growth, pricing and utilization.
Why is it difficult for tech giants to slow investment?
- Nomura believes AI could erode existing businesses, instead forcing technology companies to invest to remain competitive. Reasoning models and agents consume more computing power per answer, and insufficient capacity limits large-scale sales of models.
- Orders and capacity present a contrast: as of the end of September 2026, Google Cloud had a backlog of 560.8 billion dollars, while Microsoft’s commercial remaining performance obligations stood at 677 billion dollars; Microsoft said third-quarter capital spending was below its 50 billion-dollar plan because of delays in third-party capacity. Nomura therefore considers capacity one of the constraints on growth.
- The bank believes compute infrastructure construction over the next three years will affect the competitive landscape, making a complete freeze on capital expenditure unlikely; continued investment is favorable for semiconductor demand.
How can financing ease cash pressure?
- Nomura expects tech giants to be more likely to continue investing while bringing in joint ventures, leasing or financing backed by future cash flows. Infrastructure for Meta’s Hyperion project is being built by a joint-venture entity. The lease arrangement reduces current construction cash outlays, but rent must be paid over the next 20 years.
- Another approach is to sell guaranteed future cash flows from enterprise customer contracts in advance: cash is available during the investment peak, at the cost of transferring future cash flows at a discount; Nomura also expects the related transactions to be likely to be recognized as liabilities.
- Financing also has a cost. The four major tech companies and Oracle have issued 326 billion dollars in bonds in 2026, roughly three times the amount in 2025. Nomura estimates that even with continued heavy investment, the four companies’ financial debt and operating leases relative to equity would peak at roughly the 42% level recorded at the end of 2022; this assessment depends on earnings growth and reduced shareholder returns to increase equity.
How could excess computing capacity affect memory chips?
- Nomura believes the main risk is that excess computing capacity drives revenue per GW of capacity below expectations. Because depreciation of buildings and equipment is a fixed cost, falling prices directly hit profits.
- The bank estimates that 50%–70% of memory manufacturers’ total sales are locked in under long-term agreements with minimum margin protections. But in a severe oversupply scenario, customers would cut purchases outside those agreements, and prices outside long-term agreements could stop rising and, in the worst case, fall below protected long-term contract prices; manufacturers could then adjust capital spending or reduce capacity utilization.
- Nomura believes signs of oversupply remain weak at present and is watching cloud revenue growth, GPU rental prices and backlog growth.