Amazon Tech Radar: Expanding custom AI silicon and agentic platforms to drive enterprise revenue

Amazon is aggressively scaling its custom generative AI infrastructure and agentic capabilities to dominate enterprise cloud workloads and drive long-term revenue growth. CEO Andy Jassy confirmed Amazon is heavily investing in custom silicon, noting their new Trainium2 chips are "30% to 40% more price-performant than comparable GPUs" to lower the massive costs of AI inference for enterprises. To manage deployment risks and accelerate adoption, AWS launched Bedrock AgentCore, providing the industry's first secure, scalable runtime environment to unblock enterprise AI agent deployments. On the retail side, Amazon's AI shopping assistant, Rufus, is directly driving revenue, making users "60% more likely to complete a purchase". Under the radar, Amazon is undergoing major business transformations that require heavy backend technology support.

Its rapid expansion of "Amazon Now" 30-minute deliveries and same-day perishable groceries logically requires massive upgrades to real-time, ultra-low-latency inventory visibility systems and dynamic cold-chain routing algorithms.

Additionally, the imminent commercial rollout of its Amazon Leo (Project Kuiper) satellite network implies a need for complex new edge-to-AWS network integrations and large-scale telemetry data ingestion pipelines.

All sourced directly from Amazon's Q2, Q3, and Q4 2025 earnings calls.