Google Cloud Launches AlphaEvolve and Agentic HPC Skills

Google Cloud officially launched AlphaEvolve on Cloud Batch and released two agentic skill repositories for high-performance computing workflows on November 13, 2026, positioning software automation ahead of raw hardware ahead of the SC26 supercomputing conference. The announcement packages Google's Gemini-driven evolutionary optimizer into a production-ready Cloud Batch solution, pairs it with validated workflow templates for OpenFOAM, GROMACS, AlphaFold 3, and Devito, and expands the machine catalog to include eighth-generation TPUs and NVIDIA Blackwell bare-metal instances.
HPC product managers Megan Gawlik and Annie Ma-Weaver frame the release as packaging expertise rather than shipping more cores, writing that the goal is to let research teams focus on discovery instead of infrastructure plumbing. The launch combines reasoning automation, declarative orchestration, and a broadened hardware menu under a unified stack.
Agentic Skill Repositories for HPC Workflows
Google Cloud introduced two skill repositories, Google Skills and DeepMind Science Skills, that work with Cluster Toolkit's declarative YAML blueprints to automate workflow configuration. The repositories provide validated templates for fluid dynamics simulations in OpenFOAM and Siemens STAR-CCM+, life-sciences pipelines in GROMACS and AlphaFold 3, and subsurface modeling in Devito and Open Porous Media with browser-based 3D visualization.
Guardrails are explicit: agents operate inside IAM, Kueue quota, and budget controls with full provenance in Cloud Logging, and human approval is required before deployment. Gawlik and Ma-Weaver state that agentic HPC is not about turning unguided models loose on a terminal. The repositories let agents configure runs on Cluster Toolkit blueprints while maintaining security and cost boundaries. The release arrives as agentic systems continue to proliferate across enterprise infrastructure, with over 601 agent-related stories logged in the past 90 days.
AlphaEvolve on Cloud Batch
AlphaEvolve, the Gemini-driven evolutionary optimizer, officially released on November 13, 2026, as a generally available Cloud Batch solution. A single gcluster blueprint provisions the AlphaEvolve controller environment and uses Cloud Batch to fan out hundreds of candidate evaluations concurrently across single-VM sandboxes or multi-node MPI task groups. The solution supports Cloud RDMA, Managed Lustre, Spot VMs, and the Dynamic Workload Scheduler to balance cost against turnaround time.
The architecture distributes optimization runs across RDMA-fabric and Managed Lustre, enabling large-scale parameter searches without manual cluster provisioning. Researchers can tune workloads to leverage spot pricing for cost-sensitive experiments or priority scheduling for deadline-driven campaigns. The integration with Cloud Batch means AlphaEvolve inherits the same IAM and quota controls applied to the agentic skill repositories.
Eighth-Generation TPUs and Expanded Hardware Menu
Google Cloud debuted eighth-generation TPUs split into two lines: TPU 8t for pre-training, reinforcement learning, and mathematical modeling, and TPU 8i for low-latency reasoning and agentic inference. The release also brings NVIDIA Blackwell A5X bare-metal instances on the Vera Rubin NVL72 platform, A4 and A4X shapes on B200 and GB200 NVL72, and G4 instances with RTX PRO 6000 Blackwell for CAE and CFD workloads.
On the CPU side, H4D instances combine fifth-generation AMD EPYC processors with Cloud RDMA, including 3 TB high-memory shapes, while Arm-based C4A runs on Google Axion. Managed Lustre advertises up to 10 TB/s of read throughput, GPUDirect Storage over RDMA, and an 80 PB Dynamic Tier mirrored to Cloud Storage. The hardware breadth positions Google Cloud to compete on menu diversity, offering specialized shapes for pre-training, inference, simulation, and life sciences in a single orchestration framework.
Release Date and SC26 Presence
Officially released on November 13, 2026, the AlphaEvolve and agentic HPC stack will be demonstrated live at SC26 in McCormick Place West through November 20, 2026. Booth 3418 features demos of Antigravity for HPC and an HPC Prompt Playground. Antigravity for HPC uses the same brand as the multi-agent system that formalized Weil and Hilbert's tenth problem, signaling Google's intent to position multi-agent reasoning as a core HPC capability.
What This Means
Google Cloud's SC26 announcement shifts the narrative from hardware specifications to workflow automation and agentic orchestration. By packaging AlphaEvolve, two skill repositories, and a broadened machine catalog under Cluster Toolkit, Google is targeting researchers who want declarative, auditable, and cost-controlled automation rather than manual scripting. The explicit guardrails around IAM, logging, and human approval signal that Google is addressing enterprise concerns about uncontrolled agent behavior while still delivering the productivity gains of agentic systems. The release sets a template for how cloud providers can bundle reasoning models, validated workflows, and specialized hardware into a single production-ready HPC stack.
on Emergent today





