How Much To Run AI

How Much To Run AI is a specialist cost-planning platform for self-hosting open-weight AI models. Its positioning is clear: turn the vague question "what will it cost to run this model myself?" into a structured, comparable estimate covering VRAM, GPU hardware, electricity, staffing, depreciation and the hosted-API alternative. Target users include AI engineers, MLOps and platform teams, infrastructure architects, technical founders, CTOs, and procurement or finance staff who must justify GPU capex against metered API spend. Core features include dedicated calculators for text, image, video, speech and decision models; a browsable AI model catalog listing parameters, context window, weight format, license and repository; an API workload budget for uncached token usage; and side-by-side cloud, self-hosted and hybrid comparisons with break-even and depreciation analysis. Every result is transparently itemized: users see the VRAM formula, hardware-matching logic, electricity equation (power × hours × rate × PUE) and labour assumptions, and can adjust quantization, wall power, PUE, salaries, FTE share, system-cost multiplier, depreciation period and runtime headroom. Currency switching between USD and CNY is supported. The user experience is dashboard-like, with saved configurations, shareable configuration links and PNG/PDF report exports for stakeholder review. Technically it is a lightweight responsive web app with theme and language toggles, backed by curated public price references from AWS, Google Cloud and Microsoft Azure, NVIDIA specifications, Hugging Face model cards and channel quotes, each stamped with an update date and source link.