AWS puts stronger controls around agents as Trainium demand stays high

Key points

• AWS added a method for governing which tools an AI agent can reach through Amazon Bedrock AgentCore Gateway. Sources: S1
• Amazon said Trainium2 was fully subscribed and Project Rainier used about 500,000 Trainium2 chips. Sources: S2
• Bedrock was used by more than 100,000 companies, according to Amazon’s latest annual results. Sources: S2

Agent adoption needs access controls

AWS is treating agent governance as an infrastructure feature. The new guidance focuses on deciding which tools an agent may use and enforcing those rules at a shared gateway.

That matters because an agent can do more than produce text. It can call software, read data, or start a workflow, so a mistake can travel beyond the chat window.

Sources: S1

Custom silicon is becoming a core supply line

Amazon’s Trainium program is designed to lower the cost of training and running large models on AWS. Strong subscription levels suggest customers want another option beside standard GPU systems.

Project Rainier also shows that Anthropic and AWS are operating custom AI infrastructure at very large scale.

Sources: S2

Why it matters

AWS is connecting two parts of enterprise AI that are often discussed separately: affordable compute and safe access to business tools. Its position will depend on delivering both at scale.

Sources: S1, S2

Sources

1. Govern AI agent tool access with Amazon Bedrock AgentCore Gateway — Amazon Web Services · 2026-08-21
2. Amazon.com announces fourth-quarter results — Amazon · 2026-02-05