Backed by Y Combinator

Frontier-grade AI Infrastructure for Everyone.

We build distributed compute clusters with the cheapest CPUs and GPUs across Hyperscalers and Neoclouds for AI.

Founders

We've built software for a combined 25+ years that serves billions of users and runs on hundreds of thousands of machines.

Every time you ask LinkedIn for a warm intro, you hit three databases we built: Venice, Liquid, and Espresso.

After LinkedIn, we were tech leads of Ray, the open-source compute platform used by xAI, Cursor, Bridgewater, and Two Sigma and many others. Ray has roughly 12 million weekly downloads and is a PyTorch Foundation project.

We started with backtesting, but the same engine runs any massively parallel compute workload:

Massively parallel simulation. Backtesting and other large parameter sweeps that fan out across tens of thousands of nodes.
Post-training and reinforcement learning. Large-scale RL pipelines with massive parallel rollout and reward computation.
Multi-modal data processing. High-throughput pipelines over text, images, audio, and structured data.
Inference. Batch and high-volume inference across heterogeneous, lowest-cost accelerators.
Long-horizon agents. Agentic workloads with high tool use that run for hours and coordinate across many calls.

We're targeting clusters that scale to:

100 to 50,000 nodes

CPU + GPU

6,400,000 tasks

Parallel, in flight

< 50 ms

Dispatch & scheduling overhead

Spot Instances

Across regions and providers

Reach out if you need to run large compute workloads and effectively utilize CPUs and GPUs.