In September 2023, regrouping after failing to convince a board to invest in building out a serious illness management platform, I googled "local AI" and came across Ollama. Twenty minutes later I was chatting with a Llama model on my own machine. That sparked a hardware obsession: an NVIDIA GPU server, then more GPUs, then 96GB VRAM worth of CUDA horsepower. It worked, until it didn't. Too hot, too loud, too power-hungry to keep running. Sold the whole thing on eBay for more than I paid, and used the proceeds to buy a Mac Studio M4 Max 128GB in March 2025. Dozens of useful apps followed, bringing me most recently to MyOwnLifeOS. Foreseeing the tokenmaxxing era, I wanted to be set up in a way that I am never beholden to a cloud subscription. This is the cost story.
None of this is free of compromise. The honest version:
Speed and some accuracy. That's all I sacrifice. In exchange: no metered API, no rate limits, no subscription anxiety, no data leaving the machine. For my workload, that's an easy trade.
At 500M tokens/week · 2.17B tokens/month · 26B tokens/year
| Hardware (one-time)Mac Studio M4 Max 128GB 2TB, March 2025 | $5,000 |
| Electricity, annualTOU · active 5h/day off-peak $0.1633/kWh · idle 8W · no GPU load during peak | $54 |
| Ollama Cloud Pro, annual$20/mo voluntary · GPU-time billing · not a dependency | $240 |
| NetworkLocal inference · all clients on LAN · no internet dependency | $0 |
| Year 1 total | $5,294 |
| Year 2+ annual | $294 |