On October 28, Microsoft and Alphabet reported third-quarter results on the same day, and both delivered a version of the same message: the heaviest spenders in the AI buildout are no longer simply raising their budgets every quarter without friction. Microsoft cut its full-year capital expenditure forecast to $170 billion from $175 billion, and chief financial officer Amy Hood said delivery delays from third-party GPU providers would push some available capacity into the first half of 2027. Microsoft did not name the providers, but it relies on companies such as CoreWeave, IREN and Nebius, known in the industry as "neoclouds," to supplement its own data centers. Azure revenue still grew 48%, and Cantor Fitzgerald analysts called the quarter "further validation of Microsoft's AI-led growth opportunity," raising their price target.
Alphabet's report, released the same day, pointed in a related direction. Rather than raising its 2026 capital expenditure guidance again, as it had done earlier in the year, the company held it at $195 billion to $205 billion. Google Cloud revenue more than doubled to $30.7 billion, beating analyst estimates, yet Alphabet missed on both revenue and earnings for the first time in nearly four years and posted negative free cash flow of more than $9 billion; shares fell more than 10%. Chief financial officer Anat Ashkenazi said outside computing capacity only came online at the very end of the quarter, and that capital expenditure is expected to increase significantly in 2027, with details to come on the fourth-quarter call. CNBC's account puts that annual figure on course to exceed $300 billion next year.
The companies behind the companies: why "neoclouds" matter to your contract
Two weeks earlier, on October 14, Gartner had already flagged the strain underneath these numbers. The research firm forecast that more than 80% of neocloud providers, the specialized firms that rent out GPU capacity to hyperscalers and enterprises, will stop operating independently by 2030, driven by cost pressure and a lack of differentiated business models, and warned that more than $3.5 trillion in AI infrastructure spending could be exposed to impairment by then. Gartner did not say how many of those firms will fail outright rather than be acquired or merged, but the direction points the same way Microsoft just confirmed from the buyer's side: the GPU supply chain underneath the AI boom runs partly through companies whose own survival as independent businesses is now in question.
Why it matters even if your company never buys a GPU directly
Most companies never sign a contract with a neocloud provider; they buy AI capacity bundled into a cloud or software subscription from Microsoft, Google, Amazon or a smaller vendor built on top of them. But the supply strain Microsoft just described, and the consolidation risk Gartner just quantified, sit upstream of every one of those contracts. If the hyperscalers themselves are short on compute and the specialized suppliers behind them are not guaranteed to still exist as independent companies in four years, that risk eventually shows up somewhere in a customer's price, availability or service terms.
Four questions to ask before signing or renewing an AI cloud contract
- Ask who actually runs the hardware behind your contract. A reseller or software layer built on top of a smaller cloud provider inherits that provider's supply and survival risk, even if the invoice never mentions a GPU.
- Treat delivery dates as estimates, not commitments. Microsoft, with far more leverage over its suppliers than any single customer, still saw capacity pushed into 2027; build slack into project timelines that depend on new AI capacity.
- Check what the contract says if a provider is acquired or shuts down. Data portability, notice periods and service-continuity clauses matter more when Gartner is forecasting that most specialized providers in this market will not remain independent.
- Revisit pricing assumptions for 2027, not just this year. Both companies pointed to next year as the point where spending, and by extension cost pressure on customers, rises again.
On our side, when a client asks whether now is the moment to commit to a given AI vendor or capacity contract, this is the layer we check first: not the model's accuracy, but whether the infrastructure underneath it, and the company that owns it, will still be there on the terms promised.