I'm bootstrapped, currently at \~$20K MRR across products (main product \~12K). My biggest bottleneck isn’t leads — it’s **infrastructure stability + feature velocity**. **Current stack + costs:** * [Trigger.dev](http://Trigger.dev) → \~$1.1K/month * Vercel → \~$100/month (observability mostly) * Neon Postgres → \~$550/month * Fivetran → \~$2.5K/month (going away once we self-host ETL) * AI (OpenRouter/Speechmatics/etc.) → \~$3K/month **Scale today:** * \~2M workflow runs/month * \~20M webhooks received/month I expect \~10× scale when we hit $100K MRR (so \~20M workflows + 200M webhooks). **Options I’m weighing:** 1. **Cloudflare Workers + Workflows only** * Dirt cheap (\~$600/month even at 250M events). * Downsides: 128MB memory cap, terrible observability, primitives (idempotency, retries, concurrency) all DIY. Dev velocity feels slow, and debugging at scale is painful. 2. **Upstash + Vercel Functions** * More expensive (\~$10K–$12K/month at 250M events). * Upsides: 2GB memory per function (handles our more demanding LLM analysis workloads), good observability, primitives built-in. Much faster feature shipping, less founder stress. At $100K MRR, with \~$30K AI + \~$5K DB spend, I’d still have 50–60% gross margin even if workflows cost $10K/month. **The real trade-off:** 👉 Is it smarter (as a bootstrapped founder) to **take the margin hit and move faster**, or to **squeeze every $10K out of infra and risk slowing down growth**? Also curious if anyone here has migrated *off* “toy infra” (Upstash, Resend, Tinybird, etc.) later on. Did it bite you? Or was it fine because by the time you migrate, revenue is high enough to absorb the pain? **Why not AWS?** Lambda pricing doesn’t really solve this. AWS bills per GB-second, not CPU time. Our workflows are mostly I/O-bound, so on Vercel’s “fluid compute” we save a ton. On AWS, we’d probably be paying similar to [Trigger.dev](http://Trigger.dev) (tens of thousands/month at scale) while taking on AWS complexity. That’s not attractive for a bootstrapped solo founder.