The $0 Problem: Why Every Tool Says Your On-Prem Inference is Free

Current cost-tracking tools report $0 for on-prem LLM inference because they lack the ability to factor in hardware amortization and electricity. InferCost is an open-source Kubernetes operator that calculates true per-token inference costs by combining DCGM power metrics with declared hardware economics. Testing on a homelab with Qwen3-32B on 2x RTX 5060 Ti GPUs showed a cost of $0.41 per million tokens, enabling honest comparisons against cloud API pricing.