platoseedplatoseed
Distributed Compute for AI
We build distributed compute clusters with the cheapest CPUs and GPUs across Hyperscalers and Neoclouds for AI. Our mission is to bring frontier-grade infrastructure to everyone. We're starting by building large scale high performance computing (HPC) clusters for quantitative trading firms to run parallel simulation workloads such as backtesting. Our technology generalizes to critical AI workloads such as post-training with reinforcement learning, fine-tuning, long-horizon agents with high tool use, and batch inference.
ZibraLabs provides distributed compute infrastructure for AI workloads, leveraging low-cost CPUs and GPUs across hyperscalers and neoclouds. They offer a scalable engine built for massively parallel compute, from backtesting to RL and large-scale inference, backed by YC. The team emphasizes experience with Ray and large-scale systems.
ZibraLabs builds distributed compute clusters that span across multiple providers and regions, using inexpensive CPUs and GPUs to run massively parallel AI workloads. It supports backtesting, large parameter sweeps, post-training and reinforcement learning, multi-modal data processing, batch and high-volume inference, and long-horizon agent workloads. The platform targets clusters from 100 to 50,000 nodes, handling 6,400,000 tasks in flight with 50 ms dispatch overhead and supports spot instances across providers.
Who it’s for: Organizations needing large-scale AI compute and simulation workloads, including backtesting, RL, multi-modal data processing, and high-volume inference.
Backed by Y Combinator; references to Ray and open-source compute platform involvement; governance of large-scale production workloads hints at traction and ongoing development
Former tech-lead of Ray. Previously building databases at LinkedIn. I've built systems that run on hundreds of thousands of machines and serve over a billion people.
After a career making software that serves over a billion people, with a background primarily in large databases, I'm now building HPC clusters at Zibra Labs The past two years I've been focused on what the next generation of AI data center software looks like — and working to shape that future. I've been fortunate to do much of this in open source, which I'm deeply passionate about. Talk to me about databases, Mahjong, guitars, or weird horror movies. I'm currently building Zibra Labs.
HPC Clusters for Parallel Workloads on Spot-instances Across Hyperscalers and Neoclouds.
Zibra Labs builds HPC clusters for quantitative trading firms to run parallel backtesting simulations at scale. It handles 100 to 50,000 nodes with sub-50ms task dispatch overhead and spot instances across hyperscalers.
▲ 7
Formerly “TBD. We don't have a name yet” · why startups rename →
AI for AI Infrastructure
The Fastest Multimodal Inference OS
