{
 "title": "AI Inference Energy Cost 2026: Pounds per Million Tokens",
 "url": "https://www.servnetuk.com/research/energy-cost-ai-inference-per-token-2026",
 "publisher": "Servnet Research",
 "license": "CC BY 4.0 — free to reuse with attribution",
 "collected": "2026-08-06",
 "records": [
  {
   "metric": "H100 PCIe total board power",
   "value": "350 W",
   "source": "NVIDIA"
  },
  {
   "metric": "H100 SXM total board power",
   "value": "700 W",
   "source": "NVIDIA"
  },
  {
   "metric": "H200 power envelope",
   "value": "700 W class",
   "source": "NVIDIA"
  },
  {
   "metric": "B200 total power",
   "value": "1,000 W",
   "source": "NVIDIA"
  },
  {
   "metric": "GB200 NVL72 inference speedup vs Hopper",
   "value": "30x",
   "source": "NVIDIA"
  },
  {
   "metric": "UK colo electricity benchmark",
   "value": "30 p/kWh",
   "source": "360Strategy"
  },
  {
   "metric": "GPT-4 Turbo input price",
   "value": "$10 per million input tokens",
   "source": "360Strategy"
  },
  {
   "metric": "GPT-4 Turbo output price",
   "value": "$30 per million output tokens",
   "source": "360Strategy"
  },
  {
   "metric": "Claude 4 Opus output price",
   "value": "£60 per million output tokens",
   "source": "360Strategy"
  },
  {
   "metric": "GPT-4-equivalent late-2022 price",
   "value": "$20 per million tokens",
   "source": "Introl"
  },
  {
   "metric": "GPT-4-equivalent early-2026 price",
   "value": "$0.40 per million tokens",
   "source": "Introl"
  },
  {
   "metric": "Budget API tier range",
   "value": "$0.06–$0.30 per million tokens",
   "source": "Mirantis"
  },
  {
   "metric": "Mid-tier API range",
   "value": "$0.55–$15 per million tokens",
   "source": "Mirantis"
  },
  {
   "metric": "Frontier API range",
   "value": "$15–$75 per million tokens",
   "source": "Mirantis"
  }
 ]
}