04
Cost over time, not just the first invoice
Build costs a lot up front and little per month. Buy costs little up front and more every month as seats grow. LLM costs almost nothing to start and then charges per call. Change the inputs and switch eras to see where the curves cross.
build effort ×0.4 · maintenance ×0.6
integration ×0.5 · token price −60%/yr
integration ×0.5 · token price −60%/yr
Cumulative cost · 36 months
BuildBuyLLMfaint = pre-AI
cheapest at month 6
Buy
$23k
cheapest at month 12
Buy
$44k
cheapest at month 24
LLM
$100k
cheapest at month 36
LLM
$133k
Build is usable after about 1.6 months with a team of 3 (pre-AI: 4.0 months). Buy and LLM go live within weeks.
Table view
| month | Build | Buy | LLM |
|---|---|---|---|
| 0 | $0 | $8k | $9k |
| 3 | $75k | $15k | $23k |
| 6 | $84k | $23k | $36k |
| 12 | $104k | $44k | $60k |
| 18 | $123k | $71k | $81k |
| 24 | $144k | $105k | $100k |
| 30 | $166k | $152k | $117k |
| 36 | $188k | $211k | $133k |
- why it mattersSwitch to Pre-AI and watch the Build curve's up-front slope. That slope is what made Buy the default for anything that wasn't core. In the AI era it's less than half as steep.
- why it mattersPush LLM calls per day up to a million. Per-call pricing grows linearly with volume just like per-seat pricing does. What's different is that token prices keep falling.
- why it mattersUsage growth is the slider that decides the most. At 8% per month, anything you pay per seat or per call ends up costing more than code you own.