Sunk Cost: Calculate payback time for local LLM hardware vs cloud

What it does

Sunk Cost calculates the payback period for buying local AI hardware versus paying for API access. Users input a machine (Mac mini, Mac Studio, DGX Spark, or Strix Halo), select an open model, and estimate daily token usage. The calculator then determines how long until hardware costs are recovered through avoided API spending.

The tool displays hardware cost, estimated power consumption, model compatibility, and inference speed for each configuration. It also shows a breakdown of electricity costs, API pricing assumptions, and monthly bill comparisons.

Who it is for

The target audience is founders and builders evaluating whether to invest in local hardware instead of cloud API services. It appeals to teams trying to decide between capital expenditure on equipment versus operational spending on API calls. The tool is also useful for individuals curious about the financial feasibility of running open-source models on their own machines.

Pricing

The site does not show prices.

How it stands out

Most comparison tools focus on performance or capability. Sunk Cost strips those out and centers on a single question: financial payback. It includes often-overlooked costs like electricity consumption and allows users to input custom assumptions about API price decline over time. The calculator lets builders adjust variables like their own generation speed and monthly API bills, making the model more grounded in actual usage patterns.

The tool ranks models by payback speed at different capability levels, helping users see which hardware-model combinations break even fastest. It also shows what models fit on which machines, preventing users from considering infeasible configurations.

What a founder should check

A founder building a rival should verify the accuracy of hardware specifications (power consumption, memory bandwidth, actual inference speeds) against real-world measurements. Many estimates are currently model-based rather than measured, which could skew payback calculations significantly.

Second, they should test whether API pricing assumptions hold up. Cloud providers frequently adjust pricing downward and offer new competitive models, which directly impacts payback periods. The tool assumes APIs will become cheaper over time, but the rate matters enormously.

Third, consider the stickiness of the decision itself. Payback analysis only works if users actually stick with local hardware for years. Switching costs, operational overhead, and the difficulty of managing distributed inference deserve deeper examination beyond raw token cost.

Thinking of building something like this?

Every launch here is a competitor to somebody's idea. If yours is close, check it against the market before you build: the Full Check names the rivals, the prices and the gaps.