The $600B Capex Update: Who’s Winning the Infrastructure Race at Q1 End
CEO BRIEFING • MARCH 24, 2026
This essay is available to the entire AI Strategies for CEOs community. If you find it valuable, please consider sharing it with peers facing similar decisions
Hyperscaler AI spending now exceeds $600 billion annually. However, the allocation of this investment is changing, with immediate implications for enterprise pricing.
The Signal
As Q1 2026 ends, hyperscaler AI capital expenditures are accelerating. Combined spending by Microsoft, Google, Amazon, and Meta now exceeds $600 billion annually, a 36% increase year over year. NVIDIA’s data center revenue continues to reach new highs. This infrastructure buildout represents the largest capital deployment in technology history.
However, the composition of this spending is changing in ways that directly impact enterprise AI economics. Three trends are particularly important.
The Strategic Read
The training-to-inference shift. Over the past three years, most AI infrastructure spending focused on training compute, which requires large GPU clusters for building foundation models. This is changing. Inference compute, which supports running models in production for real-time queries and agent workflows, is now the fastest-growing segment. This shift is significant because enterprises primarily pay for inference. As inference infrastructure expands, the cost per API call decreases, making enterprise AI operating costs substantially lower.
Edge deployment is gaining traction. Hyperscalers are now deploying AI inference capabilities at the edge, closer to where data is generated and latency-sensitive applications operate. AWS’s sovereign cloud zones, Google’s distributed cloud, and Microsoft’s Azure Edge are designed to support enterprise AI workloads that require minimal latency. For companies using agents in manufacturing, logistics, or customer-facing operations, edge AI enables faster data processing and immediate responses, reducing latency bottlenecks and supporting a more efficient, reliable operational architecture.
The value chain is evolving. In 2024 and early 2025, Nvidia was the main beneficiary of capital expenditures. While this remains the case, downstream beneficiaries now include energy providers, networking companies, cooling technology firms, and construction companies. For CEOs, the AI cost curve is now influenced by energy, real estate, and infrastructure supply chains, not just GPU pricing.
The companies investing $600 billion in infrastructure are betting on sustained enterprise demand. For enterprises, this means AI costs are set to decline. Plan accordingly.
Revisit your AI cost assumptions before Q2 planning. If your financial models are based on Q3 2025 API pricing, they are almost certainly too conservative on cost reduction. Request updated pricing from your primary AI vendors for Q2 and model the impact of a twenty to thirty percent reduction in inference costs on your agent deployment ROI.
Next, assess whether edge deployment should alter your architecture for latency-sensitive agent workflows. If your agents manage real-time customer interactions, operational decisions, or manufacturing processes, edge inference may improve performance and reduce costs.
The capex boom can be a significant advantage, but only if your AI strategy is positioned to capitalize on it.
• • •
Thursday’s Deep Dive: Part 4 of “The Agentic Enterprise” examines governance and risk, including the board’s new AI oversight mandate and the five agent-specific risks most companies are overlooking.
This briefing is part of AI Strategies for CEOs. Subscribe to receive every Tuesday’s CEO Briefing and Thursday’s Strategic Deep Dive.