How much will running a local AI rig add to my UK/US electricity bill?
For typical daily chat and coding—roughly 2 hours of active generation per day plus normal desktop idle time—expect a single-GPU desktop to add £8 to £18 per month in the UK, or $5 to $12 per month in the US. On an Apple Silicon Mac, it stays under £3 / $2 per month.
A power supply badge like "850W PSU" only tells you the unit's maximum capacity, not what your system actually pulls from the wall. Your real electricity cost comes down to two numbers: what your rig sips while idling between prompts, and what it draws during short 5-to-15-second generation bursts.
Estimated Monthly Electricity Cost (UK vs. US)
The numbers below model a realistic daily schedule: 2 hours of active AI generation under heavy GPU load, 6 hours of normal desktop/idle time with a model sitting in VRAM, and 16 hours asleep or powered off (calculated at ~26.3p/kWh for the UK and ~$0.18/kWh for the US):
| Rig Setup | Wall Power (Idle / Active Load) | Monthly Energy | Est. UK Bill (~26p/kWh) | Est. US Bill (~$0.18/kWh) |
|---|---|---|---|---|
| Apple Mac Studio / MacBook Pro | 15 W / 130 W | ~10.5 kWh | ~£2.75 / mo | ~$1.90 / mo |
| Mid-Range PC (RTX 4060 Ti / 5070 Ti) | 45 W / 320 W | ~27 kWh | ~£7.10 / mo | ~$4.85 / mo |
| Enthusiast PC (RTX 3090 / 4090) | 70 W / 550 W | ~46 kWh | ~£12.10 / mo | ~$8.30 / mo |
| Dual-GPU Rig (Dual RTX 3090s) | 105 W / 880 W | ~72 kWh | ~£18.90 / mo | ~$13.00 / mo |
If you leave a high-end desktop powered on 24/7 instead of putting it to sleep overnight, tack on roughly £9–£14/mo in the UK or $6–$9/mo in the US just to cover those extra 16 hours of idle standby.
Three easy ways to slash your AI power bill
- Use Sleep Mode (or auto-unload models): An idle PC holding a model in VRAM still pulls
45W–105Wat the wall. Setting Windows or Linux to enter Sleep mode after 30 minutes of inactivity drops wall draw down to5W–10W—costing pennies a month. - Power-limit your GPU by 20–25%: Because LLM text generation is gated by memory bandwidth rather than peak GPU core clocks, capping an RTX 3090 or 4090 at 75% power trims wall draw by
~100Wwhile keeping 95%+ of your tokens-per-second speed. - Batch heavy jobs onto off-peak tariffs (UK & Time-of-Use plans): If you run overnight fine-tuning or massive image batches on a smart tariff—such as Octopus Agile or Go in the UK, or Time-of-Use plans in California and Texas—scheduling heavy workloads after midnight cuts your running cost by 50–70%.
Even in the UK—where residential electricity runs roughly 45% higher than the US average—running a capable local AI workstation still costs less each month than a single ChatGPT Plus or Claude Pro subscription.
Looking for verified hardware rigs?
Check our real-time £/GB VRAM index to find systems optimized for local model inference.