ReleaseBytes
gcp Google Cloud Blog ·

Google Cloud: New AI, Serverless, and Infrastructure Updates

blogaiinfraawsazuregcpgapreviewengineerretailmediagovernmentgcp-bigquerygcp-cloud-rungcp-gkegcp-cloud-storagegcp-pubsubgcp-vertex-aigcp-compute-enginegcp-firestore
feature announcement

Google Cloud has released several updates including new AI agent capabilities, enhanced serverless options, and infrastructure improvements. These changes aim to help organizations build, govern, and scale AI applications more effectively and efficiently. Key updates include sandboxed code execution in Cloud Run, advanced AI governance tools, and performance optimizations for AI workloads on GKE. These features are largely available now or in public preview, impacting developers, architects, and ML engineers working with Google Cloud services.

  • →Cloud Run sandboxes offer secure, isolated execution for dynamic code
  • →Service health enables automated cross-region failover for Cloud Run
  • →Apigee enhances AI agent governance and scalability
  • →Google Cloud releases report on infrastructure needs for agentic AI
  • →TPU model loading accelerated on GKE with Run:ai Model Streamer
Features (2) ›
  • Cloud Run sandboxes offer secure, isolated execution for dynamic code

    Cloud Run sandboxes, now in public preview, provide lightweight, isolated environments for running dynamically generated code, such as LLM-generated scripts or headless browser tasks, directly within existing Cloud Run service instances.

  • Service health enables automated cross-region failover for Cloud Run

    Service health is now Generally Available on Cloud Run, automating cross-region failover for business-critical applications. This feature uses readiness probes for instance-level health checks and can be configured with global or cross-region Application Load Balancers.

Enhancements (3) ›
  • Apigee enhances AI agent governance and scalability

    Apigee is being positioned as an intelligent AI Gateway to govern, secure, and scale AI agent architectures. New solutions and community tech talks focus on automating AI governance with YAML patterns and building AI portals for autonomous agents.

  • TPU model loading accelerated on GKE with Run:ai Model Streamer

    The Run:ai Model Streamer now natively supports TPUs with Google Cloud Storage in TPU vLLM 0.18.0. This integration speeds up inference pipelines on GKE by streaming tensors, reportedly over 2x faster for large models and halving peak host memory usage.

  • New OpenTelemetry-based AI Telemetry Collector agent for TPUs

    Google Cloud's AI Telemetry Collector agent standardizes TPU monitoring using OpenTelemetry. It provides operational metrics for enterprise ML workloads, routing telemetry to Google Cloud Monitoring or Prometheus, and is available on optimized Ubuntu images or via Docker.

Notes (1) ›
  • Google Cloud releases report on infrastructure needs for agentic AI

    A new report indicates that 83% of organizations require infrastructure upgrades to support agentic AI workloads. The report highlights the shift towards fluid compute and unified architectures to manage inference bottlenecks and scaling costs.

Read the original announcement →

https://cloud.google.com/blog/topics/inside-google-cloud/whats-new-google-cloud/

Related releases