By Stuart Kerr, Technology Correspondent
Published: 05/08/2025
Last Updated: 05/08/2025
Contact: [email protected] | Twitter: @LiveAIWire
Author Bio: About Stuart Kerr
The AI infrastructure arms race is no longer in the future—it’s a war fought now. In 2025, Google, Amazon, and Meta are collectively pouring $240 billion+ into foundations that won’t just power models—they’ll power market domination. From GPU farms to green data centres, from custom silicon to cross-border strategy, each company is taking a distinct path.
Google/Alphabet recently raised its annual infrastructure budget to approximately $85 billion, a $10 billion increase from earlier forecasts and a historic high for the company—signals analysts ascribe to its escalated focus on AI-bound cloud and AI‑Mode Search infrastructure (AI infrastructure arms race: Google, Amazon & Meta).
Amazon is planned to spend over $100 billion in 2025, with mid-year quarterly figures already showing CapEx expenditures in the $55 billion+ range, pressuring AWS margins into the low 30s for the first time since 2023 (Amazon Web Services profits squeezed).
Meta will invest between $66 and $72 billion this year, doubling its prior capital forecast to fuel GPU-rich agentic compute models and cloud infrastructure support (Meta to spend up to $72B).
Together, these if-spent line up near $240 billion capex. The broader tech sector—including Microsoft—is expected to bring total AI infrastructure investment above $300 billion in 2025 (The trillion-dollar AI arms race is here).
Google is extending its AI investment into existing Cloud and Java-dominated verticals—most notably through AI Mode in Search, which relies on server-side inference delivered in milliseconds. This indicates Google isn’t just building data centres—it’s optimizing the entire path from user query to GPU acceleration (Google AI Mode Search Canvas Live).
Amazon (AWS) continues doubling down on scale: the goal isn’t just to build for AI volume, but to crush cost-per-inference. Yet its margins are tightening, with Q2 2025 operating margin slipping to 32.9% as CapEx and operating expenses surged 23% year-on-year.
Meta is building “agentic AI pipelines” at hyperscale. While budgets remain in the high-60s, the firm has spoken publicly of 1.3 million GPUs under deployment by year-end, aligned with its ambition to lead decentralized multimodal intelligence.
Our earlier coverage explained the internal logic behind these figures:
AI infrastructure arms race: Google, Amazon & Meta surveyed the entire capex landscape and its broader implications.
For geopolitical deliberations on supply chains, Nvidia chip exports: US–China strategy remains key context.
Coverage of latency-driven features like Google’s AI Mode in Search Canvas ties product design directly to infrastructure decisions.
Two public research reports frame the wider stakes:
AI Arms Race 2.0: From Deregulation to Industrial outlines how AI infrastructure is now a front in U.S.–China industrial rivalry.
The AI-Driven Data Center Construction Boom 2025 highlights hyperscaler capex and global land-energy bottlenecks.
These dynamics underscore a simple truth: power is no longer hidden, and capacity is leverage.
The scale and structure of these investments reveal more than ambition—they reveal strategy. If Google leans into platform efficiency, Amazon pushes for global dominance by scale, and Meta invests in autonomy-first ecosystems, then the war is not about compute. It’s about the kind of AI world they’re building.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire. He writes about AI infrastructure, ethics, and global tech policy. Read more