GENOMA Pods: L-Series Compute Capsules

Inside the L400

The L400 is the current L-series capsule. It stacks graphics cards inside a printed double-lattice structure, manages its own thermal load, and reads its environment at the rack level, so heat handling and sensing travel with the capsule instead of living somewhere else in the building.

• Printed double-lattice structure forming the body of the capsule
• Stacked graphics cards as the compute payload
• Self-managed thermal load, handled by the capsule itself
• Rack-level humidity and temperature sensing, measured where the hardware sits

Sensing at the rack level means conditions are read next to the boards, not inferred from a room average. The capsule knows its own temperature and humidity, which is what self-managed thermals require.

From One Rack to a Full Data Hall

Capsules link. A deployment starts at a single rack and grows by adding L400s until the same design fills a data hall. The unit of growth is the capsule, and thermal and sensing behavior scale with it because every capsule brings its own.

Assembly happens on site without heavy equipment. That keeps deployment flexible: compute can sit close to the systems it serves, which matters for local-first serving and for models grounded in the physical world, like robotic world models.

Built for Sovereign Serving

GENOMA trains and serves on hardware it owns. The capsule is the physical layer of that choice. It decides where models run and who controls the machines, and the L-series capsule is the answer to that constraint at the infrastructure layer.

The workloads are production lines. KALYPSO v1.1L, GENOMA's open-weight coding and agent-orchestration model, is public on Hugging Face under Apache-2.0, and v1.2L is the production serving line. GENOMA's own models train and serve on this hardware, and every model generation is checked by deterministic gates before it ships.