Microsoft and Google Trimmed Their AI Spending Forecasts. Here Is What That Actually Means.

Key points

• Microsoft cut its full-year capital spending forecast to $170 billion, down from $175 billion, after GPU delivery delays pushed available capacity into early 2027.
• Alphabet held its 2026 capital spending guidance at $195 billion to $205 billion, missing analyst estimates, and reported negative free cash flow of more than $9 billion in Q3.
• Neocloud companies, independent data centre operators that rent GPU capacity to large platforms, saw sharp share price drops after both reports, with CoreWeave and IREN among the fallers.
• Gartner analysts predicted last week that 80 percent of neocloud companies will be absorbed or shut down by 2030.
• Microsoft's Azure cloud unit still grew sales 48 percent in the quarter, and investors pushed the stock higher on that result.

The numbers that usually dominate AI earnings season are big, round figures that only go up. Last Wednesday brought something different.

Microsoft and Alphabet, Google's parent company, both reported capital expenditure figures, meaning the money they spend on data centres and chips, that fell short of what Wall Street analysts had pencilled in. Microsoft trimmed its full-year capex target to $170 billion from $175 billion. Alphabet kept guidance flat at $195 billion to $205 billion rather than raising it. Neither company is pulling back from AI infrastructure. But neither matched the relentless upward revisions the market has come to expect, and investors in the neocloud sector felt it immediately.

Neoclouds are independent data centre companies that rent out GPU, or graphics processing unit, capacity to large tech platforms when their own facilities can't keep up with demand. CoreWeave and IREN saw their share prices fall after the reports. So did Oracle, a major buyer of AI infrastructure.

Why did Microsoft spend less than expected?

Microsoft CFO Amy Hood told analysts the shortfall came from outside the company's control. Third-party GPU suppliers ran into delivery delays that will push some capacity back into the first half of 2027. Hood did not name the affected vendors, though Microsoft is known to work with CoreWeave, IREN, Nebius, and others, as reported by CNBC Tech.

She also said that once the delayed gear arrives, spending will accelerate, driven by both the backlog and rising component costs. This is a timing shift, not a retreat. On 12 October, we reported that Microsoft has been building its own Maia 300 AI chip to cut reliance on outside suppliers, which adds context to why external GPU delays sting differently now than they would have a year ago.

Should ordinary users and businesses worry?

For most people the practical concern is reliability, not balance sheets. Anthropic's Claude AI assistant has experienced outages this year due to climbing demand. Anthropic draws much of its computing capacity from Amazon, Google, and Microsoft, with additional supply from smaller providers including Lambda and Nscale. If neocloud capacity stays tight, services built on top of AI models could face more downtime or slower performance during peak periods.

For businesses evaluating AI tools, this is worth watching. Capacity crunches tend to hit smaller customers first.

The broader neocloud picture is genuinely difficult. These companies need billions of dollars in outside investment to build data centres, but most lack investment-grade credit ratings, the financial score that signals to lenders that a borrower is a safe bet. With interest rates rising, financing is getting harder to find. Nvidia-backed Australian neocloud Firmus withdrew its plans to go public earlier this month despite holding large usage commitments from Meta and OpenAI. Gartner analysts told CNBC last week that 80 percent of neocloud companies will either be absorbed by larger players or cease operations entirely by 2030.

It's a fragile supply chain to be depending on. AI2Day has tracked the neocloud sector since 24 July 2026, and the pattern across those stories is consistent: the companies that sit between the hyperscalers and end users carry the most risk and have the least cushion.

What should readers watch for?

If you use AI services at work or through a subscription, watch service status pages. Outages caused by capacity gaps often appear with little warning and are sometimes attributed vaguely to "high demand." If a business-critical tool relies on a single AI provider, asking that provider how they handle capacity shortfalls is a reasonable question, not a paranoid one.

Watch Amazon's earnings report, due Thursday after the close, for the final read on hyperscaler capex this quarter. Amazon's AWS cloud unit is the largest in the world, and its numbers will confirm whether the shortfall at Microsoft and Alphabet is a pattern or an outlier.

Hyperscaler spending as a group is still expected to top $1 trillion next year. Azure revenue growing 48 percent in one quarter shows the demand that justifies the bill. But the plumbing that carries all that compute is clearly under strain, and the companies doing that carrying are running out of room to absorb further delays.