Google Cloud Demonstrates AI Agents for Supercomputing Workloads
The workflows use Cluster Toolkit for cluster setup and Cloud Batch to distribute hundreds of AlphaEvolve code evaluations. The announcement reports no independent benchmarks or customer results showing that the tools…

Google Cloud is demonstrating AI workflows for high-performance computing at SC26 in Chicago, pairing cluster setup skills with distributed algorithm testing for researchers managing complex workloads.
One set of reusable skills works with Google Cloud’s Cluster Toolkit to prepare computing environments. Before deployment, the workflow checks compute and storage quotas, tests capacity in the selected region and chooses a validated cluster blueprint. It then waits for a person to approve deployment.
The templates have undergone automated end-to-end testing and are available through the Google Skills and DeepMind Science Skills repositories. Related skills guide researchers through workloads such as fluid dynamics and life-sciences simulations.
High-performance computing can involve tightly coordinated simulations running across multiple nodes, as well as AI workloads. The setup workflow checks resources and prepares a cluster before deployment, while a separate workflow handles repeated algorithm tests.
That second workflow connects Google DeepMind’s AlphaEvolve system with Cloud Batch. AlphaEvolve iteratively modifies and benchmarks code. A cluster blueprint can provision a controller, and Cloud Batch can spread hundreds of candidate evaluations across single-VM sandboxes or multi-node MPI task groups.
The demonstrations cover both cluster preparation and code evaluation. SC26 runs through Friday, Nov. 20, at McCormick Place in Chicago.