Google Cloud debuts agentic HPC skills, AlphaEvolve for SC26

Google Cloud is heading into SC26 with an agentic HPC stack rather than a hardware-only pitch, releasing two reusable-skill repositories, packaging its Gemini-driven evolutionary optimizer AlphaEvolve into a Cloud Batch solution, and expanding its machine menu to eighth-generation TPUs and NVIDIA's Vera Rubin bare metal. The November 13 Google Cloud post by HPC product managers Megan Gawlik and Annie Ma-Weaver frames the direction as packaging expertise, not shipping more cores.

The two repositories, Google Skills and DeepMind Science Skills, pair with Cluster Toolkit's declarative YAML blueprints to script workflows for OpenFOAM and Siemens STAR-CCM+ fluid dynamics, GROMACS and AlphaFold 3 life-sciences pipelines, and Devito or Open Porous Media subsurface runs with browser-based 3D visualization. Guardrails are explicit: agents operate inside IAM, Kueue quota and budget controls with full provenance in Cloud Logging, and a human has to approve deployment. "Agentic HPC is not about turning unguided models loose on a terminal," Gawlik and Ma-Weaver write.

It lands in a stretch where our agents tracker has logged 601 stories in 90 days, and one of the booth demos, Antigravity for HPC, uses the same brand as the multi-agent system that just formalized Weil and Hilbert's tenth.

AlphaEvolve on Cloud Batch is the second lever. "A single gcluster blueprint provisions the AlphaEvolve controller environment and uses Batch to fan out hundreds of candidate evaluations concurrently across single-VM sandboxes or multi-node MPI task groups," the post says, with Cloud RDMA, Managed Lustre, Spot VMs and the Dynamic Workload Scheduler available to tune cost against turnaround.

The hardware breadth is where Google is competing on menu. Eighth-generation TPUs split the line between TPU 8t for pre-training, reinforcement learning and mathematical modeling and simulation, and TPU 8i for low-latency reasoning and agentic inference. On GPUs, A4 and A4X ride NVIDIA Blackwell B200 and GB200 NVL72, and the newly available A5X bare metal instances sit on the Vera Rubin NVL72 platform, while G4 shapes carry RTX PRO 6000 Blackwell for CAE and CFD. H4D CPU instances combine fifth-generation AMD EPYC with Cloud RDMA, including 3 TB high-memory shapes, and Arm-based C4A runs on Google Axion. Managed Lustre is advertised at up to 10 TB/s of read throughput, GPUDirect Storage over RDMA, and an 80 PB Dynamic Tier mirrored to Cloud Storage.

Live demos of Antigravity for HPC and an HPC Prompt Playground run at booth 3418 in McCormick Place West through November 20. The authors' own framing: "combining the reasoning power of Gemini with the codified engineering of Cluster Toolkit skills, evolutionary optimization on Batch, and purpose-built supercomputing hardware... so your teams spend their time on discovery rather than infrastructure plumbing."