AWS ships Lambda MicroVMs and Graviton5-powered EC2 C9g instances

AWS rolled out two significant infrastructure primitives. Lambda MicroVMs bring isolated, stateful execution with near-instant launch to the serverless platform, and new EC2 C9g and C9gd instances powered by Graviton5 deliver 25% better compute performance than the prior Graviton4 generation. Alongside the hardware, AWS expanded Bedrock model availability to include Claude Fable 5, Gemma 4, and OpenAI's GPT-5.4/5.5 with enterprise safeguards.

The Lambda MicroVMs primitive is aimed squarely at the emerging need for secure, isolated sandboxes to run AI-agent code and untrusted workloads — a use case that has exploded as agents increasingly execute arbitrary code. Per-second billing positions it as a cost-efficient alternative to spinning up full containers or VMs for short-lived, isolated tasks.

Graviton5 continues AWS's custom-silicon strategy to reduce dependence on x86 and improve price-performance for compute-heavy workloads, including inference. The 25% generational gain is meaningful for cost-sensitive enterprises weighing where to run models — and dovetails with the broader week's theme of businesses scrutinizing token and compute bills.

Developer reaction on Dev.to was strongly positive, calling MicroVMs a 'game-changer for isolated sandboxes' and praising per-second billing as a real cost win, though limited beta access generated FOMO. The combination of cheaper compute and expanded Bedrock model choice reinforces AWS's neutral-platform pitch: run any frontier model, isolate any workload, on AWS silicon. Watch general-availability timing for MicroVMs and whether Graviton5 inference benchmarks hold up against NVIDIA-accelerated alternatives.