On October 27, 2026, AWS raised Amazon Aurora serverless to a 384 ACU ceiling for PostgreSQL — 768 GiB of memory and 50 percent more capacity — while keeping scale-down to zero. For the bursty workloads of AI agents, this is the clearest signal yet that AWS wants the serverless database to be the default backplane, provided you treat the ACU ceiling as a real capacity decision.
On October 22, 2026, AWS made the SageMaker AI Training Session APIs generally available: you orchestrate forward_backward(), optim_step(), and sample() on stateful, serverless GPU sessions, while the service handles hardware, gradient synchronization, and isolation. If you do SFT or DPO, the high-level APIs still suffice; if you do custom RL, this removes the GPU-ops tax.
On 12 October 2026, AWS added a third form factor to Lambda — Web Functions — that runs an Express, Hono or Next.js server as-is and bills for the CPU it actually uses. Try it for I/O-heavy APIs and server-side rendering; keep Event Functions for pure event processing.
Cloudflare is absorbing Ryan Dahl’s Deno team and merging the workerd runtime with celld, Deno’s open-source implementation of Workers and Durable Objects released in August. If you want to run Worker-style apps on your own infrastructure, celld is now the path to watch.
On September 24, 2026, AWS announced an enhanced custom event bus in Amazon EventBridge: one centralized bus shared across every account in an AWS organization, with ordering guarantees, a Subscriber resource, and a new ingress/egress pricing model. For a multi-account platform still juggling cross-account rules, the workaround is officially over.
On September 9, 2026, AWS raised the Lambda function timeout to 90 minutes on Lambda Managed Instances and added Graviton5 instances, up to 25% faster than Graviton4. Teams that were working around the 15-minute cap or torn between serverless and EC2 now have a single bridge.
On 25 August 2026 AWS Lambda introduced managed runtimes in public preview, a first for the platform, starting with Node.js 26 and Python 3.15. This no-SLA testing channel exists to validate migration ahead of the end-of-support dates of current runtimes.
Announced at AWS Summit New York on August 7, 2026, AgentCore Runtime Instances bring persistent, stateful compute to Bedrock agents, removing the cold start penalty that plagued real-time deployments. If your AI agents take more than three seconds to respond, the bottleneck is your infrastructure — and AWS just fixed it.
AWS announced general availability of vector search in DynamoDB on July 30, 2026. If your operational data already lives in DynamoDB, you can drop your dedicated vector database — and the synchronization pipeline that comes with it.
