Problem

Zibra Labs addresses the cost, scale, and performance bottlenecks that arise when AI and quantitative-research teams need to run millions of largely independent computations. Its initial and clearest use case is quantitative trading: firms need to run massive parallel simulations and backtests to evaluate strategies, but conventional infrastructure is not designed for clusters spanning tens of thousands of mixed CPU and GPU nodes, nor does it make economical use of volatile spot capacity. The resulting pain is slower research cycles, high cloud bills, and unreliable or inefficient compute utilization.

The company’s stated goal is to make testing trading strategies “fast, cheap, and reliable.” It starts with backtesting for quant firms, while targeting the same underlying problem in reinforcement-learning post-training, robotics simulation, molecular dynamics, and other computationally intensive workloads.

Product / Service

Zibra is building a distributed runtime and cloud-infrastructure layer that assembles compute clusters from the lowest-cost CPUs and GPUs available across hyperscalers and neocloud providers. The planned system targets clusters of 100 to 50,000 nodes and as many as 6.4 million parallel in-flight tasks, with less than 50 milliseconds of dispatch and scheduling overhead. It is designed to exploit spot instances across regions and providers, and can run in a customer’s cloud or on premises so sensitive trading research remains under the customer’s control.

The delivery model is infrastructure and APIs rather than a consumer-facing trading application: customers bring large workloads, while Zibra handles distributed task placement and execution across heterogeneous resources. The benefit is higher-throughput, lower-cost compute for backtesting and other parallel workloads, with a path to use the same engine for batch inference, multimodal processing, long-horizon agents, and reinforcement-learning pipelines.

Market

Zibra competes in distributed AI infrastructure, cloud HPC, and workload-orchestration software, with an initial vertical focus on quantitative trading. Its closest adjacent alternatives are general-purpose distributed-compute systems such as Ray and traditional HPC schedulers such as Slurm, as well as cloud-native batch infrastructure. Zibra’s differentiation claim is that existing approaches do not adequately support its target scale, cross-provider spot instances, low-latency dispatch, and high task throughput. QuantConnect and similar platforms are relevant at the backtesting-application layer, but Zibra is positioning itself as the underlying compute infrastructure rather than as a research terminal or trading platform.

The company is very early: Y Combinator lists it as an active Spring 2026 company founded in 2026 with a two-person team. Public funding databases report a $500,000 pre-seed round in May 2026, with Y Combinator and other investors listed, but the company has not publicly disclosed customer names or a revenue figure. One database labels its financing status as “Generating Revenue” while leaving current revenue blank, so the most defensible characterization is early commercial or possibly pre-revenue, with traction not yet demonstrated publicly at scale.