Local AI: What's the Power Bill?

What does running a local LLM actually add to your electricity bill? We compare the RTX 5090, power-limited GPUs, older multi-GPU workstations, AMD Strix Halo, and Apple Silicon.

Using an illustrative electricity rate of $0.15/kWh, we break down monthly running costs, 24/7 operation, and electricity cost per million tokens. We also explore how inference speed changes efficiency—and how local electricity costs compare with cloud inference once hardware ownership is included.

Power figures are illustrative estimates, not controlled benchmarks. Actual costs depend on your hardware, model, inference software, workload, and electricity rate. Monthly calculations use 30 days; annual calculations use 365 days.

#LocalLLM #LocalAI #LLM #AIHardware #RTX5090 #AppleSilicon #StrixHalo #GPUPerformance #PowerConsumption #ElectricityCosts #AIInference #SelfHostedAI

00:00 The overlooked cost of local AI
00:58 RTX 5090 power and monthly electricity costs
01:54 Power limiting: a smaller electricity bill
02:42 Two GTX 1080s: memory and running costs
03:15 Four RTX 2080s: more GPUs, more power
04:29 AMD Strix Halo and unified memory
05:16 Apple Silicon electricity costs
06:06 What happens when systems run 24/7?
07:11 Electricity cost per million tokens
08:32 Local electricity versus cloud inference
08:58 Hardware ownership and the local-versus-cloud decision
09:36 Privacy, flexibility, and control
10:35 Your electricity rate and measuring your own setup