Building the Cloud for AI Agents | AWS CEO Matt Garman

AWS remains the primary engine for modern enterprise and startup innovation, currently navigating a massive shift toward agentic workflows. To support this, the company is prioritizing infrastructure that enables agents to operate with high performance, security, and granular permissions, moving beyond traditional human-centric interfaces. With a $220 billion capital expenditure plan for 2026, AWS is aggressively scaling GPU capacity and power procurement to meet insatiable demand while maintaining a diversified ecosystem. Proprietary hardware like Graviton and Tranium chips now serve as the backbone for both training and inference, offering significant cost-performance advantages. As enterprises transition from proof-of-concepts to production, the focus is shifting toward safe, autonomous agent deployment, robust evaluation systems, and maintaining data sovereignty within VPCs, ensuring that AI-driven development remains both scalable and secure.