Today we are announcing a partnership with WATTER, the company turning water heaters into a distributed compute network. Our inference system and models are now available on the WATTER AI platform, so developers can build agentic workloads on infrastructure that heats homes and businesses while it reasons.
About WATTER
WATTER builds a thermal compute system that places GPUs directly on water heaters, in both residential and commercial settings. The GPU the heats the water as it works, and the water carries that heat away and keeps the hardware cool. It is a novel approach to a problem every data center fights: the same energy that powers inference becomes hot water for a home or business, instead of being lost as waste heat. Now waste heat from compute has value, and more pockets in our energy grid can be repurposed for valuable compute infrastructure.
Agent workloads, powered by your water heater
Agentic workloads are demanding in ways beyond chat and one hop queries. They run for long periods of time, hold large amounts of context across many steps, and need to generate tokens quickly so the agent can keep moving. Most inference stacks were not built for this, and both cost and latency climb fast as concurrency grows. In memory and compute constrained environments, capable agents impractical to run at all.
Subconscious is an inference system built specifically for this kind of work. Pairing it with WATTER's compute lets the two systems do more on the same hardware: more concurrent agentic tasks, longer context reasoning, and faster token throughput. The result is the ability to run real agentic workloads on WATTER's distributed fleet, at a scale, speed, and capability that's newly possible.
Try it today
Subconscious is live on the WATTER AI platform now. You can try it at oasis.watter.com, running our inference system and the TIM-Qwen3.6-27B model.
