Bootstrapper runs multiple $10K MRR businesses on a $20/month stack

A solo developer details the exact infrastructure behind multiple profitable SaaS products that collectively cost under $20 a month to operate. The core approach: a single $5-10 VPS on Linode or DigitalOcean, Go backends compiled to static binaries, SQLite with WAL mode instead of Postgres, and local AI inference on a used RTX 3090 running VLLM for batch workloads. For user-facing LLM features requiring frontier models, OpenRouter provides a single integration point with automatic fallback routing across providers.

The piece is a direct counter to the default 2026 startup playbook of Kubernetes clusters, managed databases, and ballooning cloud bills before finding product-market fit. The author argues that near-zero infrastructure costs provide the same effective runway as venture funding, without the pressure or complexity. Practical tips include exploiting GitHub Copilot’s per-request pricing for heavy agentic coding sessions, using swap files to stretch 1GB VPS instances, and structuring local LLM workloads as batched async requests to maximize GPU throughput.

The underlying thesis is that VCs actively penalize capital efficiency - the author keeps getting rejected from pitch nights precisely because the business doesn’t need money. For indie builders and bootstrappers, the message is clear: modern tooling has made it possible to run real revenue-generating products on infrastructure that costs less than a streaming subscription.