Fly Labs
Ask FlyBot

Sunk Cost – How long until a local LLM rig pays for itself?

I kept hearing "just buy a Mac and run models locally, it pays for itself" and wanted to check. Sunk Cost takes a machine, a model and how many tokens you use a day, and works out how long the hardware takes to pay back against renting the same model by the token.Obviously there are other reasons to buy your own hardware aside from just saving money on llms but this is just looking at it from a raw cost saving perspective.If you have any ideas on how I can make this more helpful lmk!

Hacker NewsToolAISource
0Sign in to voteCopy link

FL score

42

out of 100

Verdict

SKIP

medium confidence

Competition

No competitor data yet

Trend

No signal yet

A calculator that tells you when local LLM hardware breaks even against API costs.

The pain

People considering buying expensive hardware (Mac Studio, RTX 4090 rigs) hear conflicting claims about ROI. They lack a concrete way to compare: $3,500 upfront hardware cost vs. $0.10 per million tokens on Claude/GPT-4. The math is non-obvious because it depends on their actual usage patterns, which vary wildly.

The gap

No existing tool lets you input your hardware cost, model choice, daily token usage, and API pricing to see payback period. You either guess or build a spreadsheet yourself. The gap is small but real—it's a specific calculation people actually need when making a purchase decision.

Build angle

Start by validating whether people actually use this before building. Post the calculator as a simple web form on Hacker News, Twitter, or relevant Discord communities. Track: (1) How many people use it, (2) What hardware/models they input, (3) Whether they actually buy hardware after using it. If 100+ people use it monthly and express it changed their decision, build a polished version. Add features only if users request them—like comparing multiple hardware configs, factoring in electricity costs, or tracking actual vs. predicted usage.

Strengths

  • Solves a real micro-decision people face when considering local inference
  • Simple enough to build as MVP in a weekend
  • Natural audience: ML engineers, indie hackers, people on HN/Twitter who discuss this exact topic
  • Low competition—no existing tool does this specific calculation
  • Could become a reference point in discussions about local vs. cloud inference

Risks

  • Market is tiny: only people actively considering $2k+ hardware purchases need this. That's maybe 5,000-10,000 people globally per year
  • Payback period math is straightforward—users might just do it in a spreadsheet instead of visiting your site
  • API pricing and model performance change constantly, making the tool stale quickly
  • No clear monetization path. Can't charge users (too cheap to calculate). Affiliate links to hardware are weak. Sponsorships from API providers create bias
  • Usage is one-time or infrequent. No reason to return after calculating payback period
  • Actual purchasing decisions depend on factors beyond ROI: ecosystem lock-in, privacy concerns, latency requirements, model quality—the tool only addresses one dimension

Questions about this idea?

FlyBot reads the scoring and gives you a second opinion on “Sunk Cost – How long until a local LLM rig pays for itself?”.

Open FlyBot