A distributed AI compute network built into water heating. Every kilowatt-hour runs AI workloads first, then heats water for homes and buildings.
View the systemContactNearly every watt a processor draws ends up as heat. In a data center that heat is a cost. In a water tank it is the product.
of U.S. household site energy, 2020 [1]
of U.S. homes heat water with electricity, 2020 [1]
about 1.5% of world electricity [2]
projected, more than double [2]
Accelerated compute modules sit inside the water-heating system, where hot water is used.
Chip heat moves into the water and is stored as thermal energy until it's needed.
Jobs are scheduled by hot-water demand, tank temperature and grid signals.
Thousands of nodes work as one GPU cloud for flexible AI workloads.
Workloads → batch & edge inference · fine-tuning · rendering & simulation
Markets → homes · multifamily · hotels & gyms · hospitals & campuses
Assumes 95% heat recovery · water 4.186 kJ/kg·K
liters per day · US gal · illustrative