Dr. Robert Castellano's Semiconductor Deep Dive Newsletter

Dr. Robert Castellano's Semiconductor Deep Dive Newsletter

Share this post

Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
The AI Cloud Arms Race Is Still Accelerating

The AI Cloud Arms Race Is Still Accelerating

Dr. Robert Castellano's avatar
Dr. Robert Castellano
Jun 03, 2025
∙ Paid
1

Share this post

Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
The AI Cloud Arms Race Is Still Accelerating
Share

Hyperscalers and the Role of AI Infrastructure

Recent headlines have suggested that the AI buildout might be slowing. Microsoft was reported to be canceling up to 2 gigawatts (GW) of data center leases—nearly one-third of its global capacity—and Amazon is said to be pausing some overseas expansion talks. These stories have triggered speculation that AI demand is peaking. But the actual spending data from Q1 2025 tells a very different story.

Hyperscalers—Amazon (AWS), Microsoft (Azure), Google (GCP), and Meta—are the biggest builders of cloud infrastructure in the world. Their massive data centers are the backbone of artificial intelligence workloads, enabling everything from search and social media to ChatGPT and recommendation engines. Capital expenditures (CapEx) from these companies fund the chips, networking gear, power, cooling, and software that make modern AI possible.

In Q1 2025, hyperscaler CapEx reached $92.0 billion—slightly below the previously reported $93.8 billion, but still up significantly year-over-year.

Full-year CapEx is now projected at $414 billion in 2025 and $432 billion in 2026, both figures revised upward from earlier forecasts. Simply put: the cloud giants are still spending aggressively.

Microsoft and Amazon Are Strategically Rebalancing

The changes at Microsoft and Amazon appear to be strategic reallocations—not signs of retreat. Microsoft’s data center cancellations are likely tied to OpenAI shifting some of its future infrastructure to Oracle. Amazon’s adjustments are thought to be related to geopolitical concerns and upcoming U.S. AI export controls. AWS clarified that the changes are part of “routine capital management” and confirmed its long-term AI infrastructure plans.

Meanwhile, Meta and Google are going in the other direction. Meta raised its 2025 CapEx forecast to as much as $72 billion, citing increased AI data center demand and rising hardware costs from tariffs.

Google reaffirmed its $75 billion CapEx plan, focused on servers and datacenter infrastructure for Google Cloud and DeepMind.

These moves show that the hyperscaler AI investment cycle is not reversing—it’s realigning.

Hyperscaler CapEx Growth from 2022 to 2027

Hyperscaler capital expenditures are the financial backbone of the AI infrastructure buildout. Chart 1 shows projected CapEx from the four major U.S. cloud providers—Microsoft, Meta, Google, and Amazon—over a six-year period. The most significant increase occurs in 2025, where total CapEx spikes from $250 billion in 2024 to $336 billion, driven by aggressive GPU cluster deployments and data center expansion plans. Spending moderates in 2026 and 2027 as buildouts mature, but investment levels remain historically high. These figures reaffirm that the AI infrastructure cycle is not peaking—just shifting from first-wave to second-wave scaling.

A graph of a company

Description automatically generated with medium confidence

Chart 1

Source: The Information Network (www.theinformationnet.com)

AI Data Center Node Cost Breakdown

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Dr. Robert Castellano
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share