AWS re:Invent 2026 doesn’t happen until November 30–December 4, 2026 — so if you’ve seen a headline claiming AWS already unveiled a wave of new AI services this year, that’s getting ahead of the calendar. What did actually happen is re:Invent 2025 (December 1–4), and it was one of AWS’s biggest AI-focused years yet. Here’s what was real.
Bedrock AgentCore: the centerpiece announcement
AWS’s flagship AI-agent push at re:Invent 2025 was Bedrock AgentCore, expanded with new Policy, Evaluations, and Memory capabilities. The pitch is straightforward: as companies move from experimenting with AI agents to actually running them in production, they need guardrails (Policy), a way to measure whether an agent is performing well (Evaluations), and a way for agents to retain context across sessions (Memory) — three gaps that had been slowing enterprise agent adoption.
New models: Nova 2 and 18 open-weight additions
AWS introduced the Nova 2 model family, its next generation of in-house foundation models, alongside a notable move toward openness: 18 new open-weight models made available through AWS, including Mistral Large 3, Ministral 3, and Gemma 3. That’s a meaningful shift — AWS positioning itself as a place to access a wide range of both proprietary and open models rather than pushing customers toward one house model exclusively.
New compute: Trainium3 and Graviton5
On the infrastructure side, AWS announced Trainium3 UltraServers — its next-generation AI training silicon — alongside Graviton5, the latest generation of its custom Arm-based CPU line. Both are aimed at the same underlying problem every hyperscaler is chasing right now: enough compute, cheap enough, to keep up with AI training and inference demand without depending entirely on Nvidia GPU supply.
Three new “Frontier Agents”
AWS also introduced three purpose-built AI agents under a “Frontier Agents” banner:
- Kiro — an agentic coding tool
- AWS Security Agent — automated security monitoring and response
- AWS DevOps Agent — automated operations and infrastructure management
The pattern here is AWS shipping its own opinionated, ready-to-use agents on top of the more general-purpose Bedrock AgentCore platform — narrower tools built for specific jobs, rather than leaving every customer to build an agent from scratch.
S3 Vectors goes GA
Amazon S3 Vectors — native vector storage and search inside S3, aimed at powering AI retrieval and search applications — reached general availability, supporting up to 2 billion vectors per index. This matters for any company building retrieval-augmented generation (RAG) systems, since it removes the need to run a separate dedicated vector database for large-scale use cases.
AWS Transform, its automated legacy-application modernization tool, also got enhancements aimed at making that migration process more AI-driven.
What to watch for at re:Invent 2026
Early-bird registration for re:Invent 2026 opened with a deadline around August 25, and the event itself runs November 30–December 4, 2026 in Las Vegas. Based on the pace AWS set in 2025 — a major agent platform expansion, new proprietary and open models, new custom silicon, and purpose-built agents all in one week — expect this year’s event to build on those same four fronts rather than introduce an entirely new direction. We’ll cover the actual announcements once the event happens, rather than guessing at specifics now.
Frequently Asked Questions
Did AWS already announce 12 new AI services in 2026?
No confirmed announcement matches that framing — re:Invent 2026 hasn’t happened yet as of this article. The most recent confirmed AWS AI announcements are from re:Invent 2025.
What was the biggest AI announcement at re:Invent 2025?
Bedrock AgentCore’s expansion (Policy, Evaluations, Memory) was the centerpiece, aimed at helping enterprises move AI agents from pilot projects into production.
When is AWS re:Invent 2026?
November 30–December 4, 2026, in Las Vegas.
What is Amazon Nova 2?
AWS’s next-generation family of in-house foundation models, announced at re:Invent 2025 alongside 18 additional open-weight models from other providers made available through AWS.
What is S3 Vectors used for?
Native vector storage and search built into Amazon S3, designed for AI search and retrieval-augmented generation (RAG) applications, supporting up to 2 billion vectors per index as of its general availability announcement.
Bottom Line
re:Invent 2025 was a genuinely dense AI news week for AWS — a new agent platform, new models (both in-house and open), new silicon, and purpose-built agents for coding, security, and DevOps. re:Invent 2026 is still ahead of us; we’ll update this piece with what’s actually announced once the event happens in late November.
As of August 2026. re:Invent 2026 has not yet occurred — this article will be updated with confirmed 2026 announcements after the event.
