GPU servers in hotels and homes. Waste heat captured to deliver free hot water. Compute sold through OASIS — our managed inference platform. Heat-aware routing steers workloads to devices that need thermal output.
Hardware deployed inside existing buildings. Compute sold as managed inference. Heat-aware routing makes the per-token cost structurally cheaper than any other GPU cloud.
GPU racks deployed in hotels, multifamily, district plants. 80–90% waste heat captured into existing DHW loops. Closed dielectric loop · 64× B200 per rack · 100 kW. Behind the meter. Zero new grid load.
Managed LLM inference on commodity NVIDIA. Think Groq — but on B200s, with free cooling. OpenAI-compatible API. Open-weight (Qwen, DeepSeek, Llama, Mistral, Kimi). Auto-routing across the fleet.
Workloads steered to devices that need thermal output right now. Hot-water demand becomes the compute scheduler. We monetize energy that hyperscalers waste.
Live commercial deployment · production inference platform · government contract · residential utility partnership. Post-prototype. Entering deployment scale.
11 sites operational today — 1 commercial (Hyatt Reunion) + 10 residential Apollo units. Same architecture, same heat-aware routing, both segments served. Cheapest per-token by physics.
Inference jobs are dispatched not just by latency and price — but by thermal demand at the destination. Hotels heating water in the morning rush get the work. Homes pre-heating tanks at peak get the work. The rack pays for itself twice — compute revenue plus displaced gas.
One architecture. Both segments. Span/XFRA just launched the commercial-only version ($419M raised · $103M revenue) — the market validated the model. WATTER is already operating it, with heat capture, across both consumer and commercial. Validated, shipping, and broader.

1 GW of AI compute the old way: ~$60B and a 5–7 year utility queue. WATTER delivers it as 10,000 × 100 kW hotel sites riding mechanical rooms, electrical service, fiber, and hot-water demand that already exist. OPEX savings not even counted below.
LG hotel deployment: 64× B200 at $3.00/GPU-hr × 8,760 hours. Capacity sold through OASIS before energization.
MegaWATT Container ships pre-integrated. Same compute, same heat-aware routing — derisked, easier, cheaper at scale. Site-to-revenue compresses from 14 weeks (retrofit) to 14 days (drop-in). Logistics, not construction.
Commercial · government · residential · industrial. Four go-to-market channels operating in parallel.
Funds 10 LG hotel deployments ($16.8M forward ARR), the Apollo residential rollout, and the defense pipeline. Hyatt Reunion is live. OASIS ships in production. Every dollar accelerates already-operating revenue.