The right question is: how much energy did that answer cost?
Energy per correct, traceable, auditable answer is the metric that matters for enterprise AI.
FRONTIER LLM
Surface
100s of GPU cores
1 vCPU
100s of GB VRAM
1GB RAM
~300–700W per GPU
~10–15W
3–25% hallucination rate
Architectural traceability
Hyperscaler data centre
Your hardware