AWS Summit NY debuts Lambda MicroVMs, EC2 G7 Blackwell instances, and Bedrock Managed Knowledge Bases

AWS used its New York Summit to ship a dense slate of serverless and agent infrastructure. Headlining were Lambda MicroVMs, a new serverless primitive offering VM-level isolation with rapid launch/resume and up to 8-hour state preservation, and EC2 G7 instances powered by NVIDIA RTX PRO 4500 Blackwell GPUs, which AWS says deliver up to 4.6x inference performance over the prior G6 generation.

On the agent side, Amazon Bedrock AgentCore reached general availability, letting teams build and run production-grade AI agents through configuration rather than extensive code. AWS also introduced a fully Managed Knowledge Base for Bedrock that streamlines enterprise RAG pipelines with native data connectors, smart parsing for multi-format data, and an Agentic Retriever for complex queries, integrating with AgentCore Gateway. Additional launches included Continuum for vulnerability detection and S3 annotations for agents.

Strategically, AWS is positioning itself as the neutral execution layer for the agent era — emphasizing config-over-code, managed RAG, and isolation primitives rather than competing head-on at the frontier-model layer (where it relies heavily on Anthropic's Claude, a relationship now under pricing strain). The MicroVM bet targets the messy reality of agents that need persistent, isolated, long-running sessions.

Reaction was enthusiastic but cost-cautious: developers welcomed the infra but flagged pricing opacity on the MicroVM isolation model as a blocker to adoption planning. Swami Sivasubramanian, VP of AWS Agentic AI, framed the momentum as proof that 'agentic AI works for organizations at every scale.' What to watch: whether MicroVM pricing is competitive enough to pull workloads off containers, and how the Managed Knowledge Base stacks up against dedicated RAG vendors.