AWS Drops Major AI Updates: FinOps Agent, Faster Chips, and a Surprise Model Withdrawal

AWS made several significant announcements at this week's New York Summit, spanning cost management, compute performance, and AI model availability. Here's what tech leaders need to know.

**FinOps Agent hits preview.** AWS's new AI agent can answer cost questions, flag optimization opportunities, investigate spending anomalies, and even open Jira tickets or post findings to Slack automatically. For engineering teams managing cloud spend, this tool promises to automate much of the grunt work around cost governance.

**New Graviton5 instances go live.** The M9g and M9gd instances, powered by AWS's fifth-generation Graviton processors, deliver up to 25% better compute performance than the previous generation. They also introduce the Nitro Isolation Engine, which AWS says is the first formally verified cloud hypervisor - a notable security feature for enterprises with strict compliance requirements.

**Gemma 4 arrives on Bedrock.** Google DeepMind's latest open models are now available across three variants, supporting multimodal input, function calling, and 35+ languages. This gives teams more flexibility in choosing model size vs. latency tradeoffs.

**Claude Fable 5: here and gone.** Anthropic launched Claude Fable 5 on Bedrock June 9, but by June 12, AWS revoked access at Anthropic's request to comply with a US export control directive. Teams that had integrated the model will need to pivot to alternatives like Opus 4.8, which remains unaffected.

**Other notable updates:** AWS CLI v1 enters maintenance mode (migrate to v2); a new Workload Credentials Provider lets apps outside AWS get short-term credentials without long-term keys; and Kiro launched a Pro Max tier for high-volume developer use.

For deeper analysis on how frontier teams are achieving 4.5x deployment velocity using AI-native development practices, read Swami Sivasubramanian's detailed post on the AWS blog.

Source: AWS