The self-host vs API decision for Indian founders has shifted three times in 12 months, and the new equilibrium is interesting. With Qwen 3.6 35B running 80 tok/s on a 16GB consumer GPU, the break-even token volume vs GPT-5 INR pricing has dropped from 200M/month to roughly 50M/month.

For most early-stage Indian SaaS — the kind doing 10-30M tokens/month — the API still wins on operational simplicity. Past 50M, self-host on Hetzner Helsinki (₹3,800/month) wins decisively. The new question is not whether to self-host, but at what scale to switch.