Apple Silicon costs more than OpenRouter
William Angel's blog post compares the total cost of ownership for running LLMs locally on Apple Silicon versus using OpenRouter's API. Factoring in electricity costs (e.g., $0.30/kWh in California) and hardware amortization over three years, a Mac Studio with M2 Ultra costs about $0.50 per million tokens for a 70B model, while OpenRouter charges $0.15–$0.30 for similar models. The analysis assumes 24/7 operation and includes idle power draw. For light usage, local inference can be cheaper, but for heavy workloads, cloud APIs win on cost. The post also notes that Apple Silicon's efficiency advantage is offset by high upfront hardware costs and electricity rates.
Developers may overestimate cost savings of local LLM inference versus cloud APIs.