Microsoft Cuts Capex Forecast to $170 Billion as AI Build-Out Hits Delivery Delays

By James Calloway

Microsoft lowered its capital expenditure forecast for the year to $170 billion from $175 billion on Wednesday, the first cut from a major hyperscaler after quarters of relentless AI spending escalation. Alphabet held its guidance at $195 billion to $205 billion but still missed analysts' estimates on both capex and results, sending its shares down about 10% in extended trading.

The numbers do not signal a slowdown in the AI build-out. They do expose a new point of fragility: the supply chain of outside computing capacity that Microsoft, Alphabet and their AI customers increasingly depend on is delivering late.

What caused the shortfall?

Microsoft CFO Amy Hood told investors on the earnings call that third-party providers of graphics processing units had experienced delays that will push some available capacity into the first half of 2027. She did not name the providers. Microsoft relies on CoreWeave, IREN and Nebius, among others, for rented GPU capacity.

Alphabet's finance chief, Anat Ashkenazi, told analysts that outside computing power "came online only at the very end of Q3." That timing shortfall contributed to a quarter in which Alphabet missed on earnings and revenue and reported negative free cash flow of more than $9 billion.

The context matters: for several quarters, hyperscalers have raised capex forecasts each earnings season as they race to build AI infrastructure for what they describe as insatiable demand. Wednesday's reports broke that pattern. Both companies spent less than Wall Street expected, not because demand softened, but because capacity did not arrive on schedule.

The takeaway in plain words: the bottleneck has shifted from willingness to spend to the ability of suppliers to actually deliver machines.

How did the market react?

The read-through hit the companies most exposed to hyperscaler rental contracts. Shares of neoclouds — smaller firms that rent out GPU-heavy data centers — CoreWeave, Nebius and IREN all dropped after the reports. Oracle, one of the biggest spenders on AI infrastructure, fell as well.

Alphabet's after-hours drop of roughly 10% was among the harshest reactions to its earnings as a public company, according to Bespoke's Paul Hickey.

Not everyone sold. Chris Ballard, managing director of Check Capital Management, wrote in an email to CNBC: "We are pleased with what we see and what we heard, and trust management's approach to their AI spending."

Microsoft's investors, by contrast, cheered. The company beat earnings and revenue estimates, driven by 48% sales growth in its Azure cloud business, and its stock rose despite the capex shortfall.

"We view the quarter as further validation of Microsoft's AI-led growth opportunity," Cantor analysts led by Thomas Blakey wrote in a Wednesday note to clients. They recommend buying the stock and raised their one-year price target to $630 from $608.

The takeaway: revenue momentum can excuse a spending miss, but only when the growth numbers are as strong as Azure's 48%.

Where do the other hyperscalers stand?

Investors will get the final read on hyperscaler capex for the quarter when Amazon reports results after the close Thursday.

Meta, in its earnings report last week, said it expects full-year capex of $135 billion to $145 billion, raising the bottom of the range from $130 billion. Like Alphabet, Meta shares sank after the company reported negative free cash flow.

As a group, hyperscaler capex is expected to top $1 trillion next year and, according to some estimates, continue rising from there. Hood said Wednesday that she sees Microsoft's capital spending accelerating in the first half of 2027, driven by higher component costs and the delayed deliveries. A cut today, in other words, may simply be a deferral into a steeper bill later — though Hood offered no guarantee that the delayed capacity itself will arrive on the new schedule.

The takeaway: aggregate spending forecasts still point up; the calendar of that spending is what shifted.

Why do the neocloud delays matter now?

Delays from emerging neocloud partners have hit Microsoft before. They matter more now because Anthropic, one of the largest consumers of AI compute, is preparing to go public. Anthropic depends on infrastructure from Amazon, Google and Microsoft for much of its computing needs, alongside capacity from SpaceX, Lambda and Nscale — which filed to go public in September but has not updated its prospectus in weeks.

Anthropic's Claude assistant has at times been unavailable this year due to climbing demand. Rival OpenAI, which has said an initial public offering in 2027 is likely, is also in constant need of more capacity.

The structural problem is financing. Neoclouds require billions of dollars in outside capital to fund their build-outs but lack investment-grade credit ratings. With interest rates rising and concern about their debt loads mounting, lenders are pulling out of the market.

The contraction has already begun. Last week, analysts at Gartner shared with CNBC the firm's prediction that by 2030, 80% of neocloud companies will either be absorbed or completely cease operations. Nvidia-backed neocloud Firmus, based in Australia, withdrew plans to go public earlier this month, despite boasting large usage commitments from Meta and OpenAI.

The takeaway: the sector supplying hyperscalers with flexible capacity is financed on thin ice, and its consolidation would concentrate risk further.

Is the AI build-out stalling?

None of the evidence suggests the build-out is over, or even close to plateauing. Alphabet kept guidance in a range that would make 2025 its biggest spending year ever. Meta raised its floor. Microsoft says spending accelerates in 2027. Hyperscaler capex is still expected to cross $1 trillion next year.

What changed this week is the composition of risk. The question is no longer only whether demand justifies the spending. It is whether the supply chain — GPU vendors, neocloud landlords and their increasingly reluctant lenders — can deliver the capacity on the timeline the forecasts assume. Hood's 2027 acceleration, Gartner's 80% failure prediction and Firmus's withdrawn IPO all point to the same squeeze: more money chasing a delivery pipeline that keeps slipping.