Distributed compute clusters combining cheap CPUs and GPUs across hyperscalers and neoclouds for AI and HPC workloads at scale.
Zibra Labs builds and manages distributed compute clusters that scale from 100 to 50,000 CPU and GPU nodes across hyperscalers and neoclouds, targeting AI and high-performance computing workloads. Initially focused on large-scale parallel simulation for quantitative trading firms (backtesting), the platform also supports post-training and reinforcement learning, multi-modal data processing, high-volume batch inference, and long-horizon agentic workloads. Scheduling overhead is under 50 ms and the system supports spot instances across regions and providers. Zibra Labs is a product of Zibra Labs.