Secondary research and roadmap division

LuxiEnergy · AI Energy Systems

Reduce waste across the AI energy system.

LuxiEnergy organizes LuxiEdge research on AI compute, memory movement, workload behavior, and power delivery. It serves the mission of reducing the total cost and energy required for trustworthy computational results. It is not a second commercial product: LuxiQuant Risk State Engine is LuxiEdge’s only commercial offer. Each research result remains inside its measured boundary; lifecycle savings from fewer reruns and less revalidation have not yet been measured.

Four connected sources of waste

AI compute

Numerical execution and kernel research aimed at reducing unnecessary work while preserving the declared result contract.

Memory movement

TRADE, WNSM, Qwen, Lambert, microkernel, and related experiments examine movement, materialization, residency, and locality.

Historical AttentionTransformer V2 studies and limits

Workload behavior

LuxiPak and LuxiPhase cover request packing and execution scheduling. Their current maturity and evidence remain documented in Research and Capabilities.

Power delivery

LuxiSDG is preserved as historical software-defined-generator work. It is not presented as a currently available product.

Measured work and system history

Inference-energy evidence

Existing H100 prefill energy measurements, correctness experiments, benchmarks, and limitations remain available in Research and Evidence.

Inference research · Measured evidence

Energy-system concepts

LuxiPak, LuxiPhase, and LuxiSDG are organized here without converting historical or planned work into available-product claims.

System architecture · Maturity labels

Measurement boundary. GPU-board energy is not wall-plug or facility energy. Prefill is not complete inference. Research results do not establish a system-wide energy advantage.