He spent two years building his own AI model. A weekend experiment made more money, so he killed it. Eugene Cheah is the founder of Featherless AI, an AI inference platform that gives instant access to more than 40,000 open source AI models. He spent 2 years building his own open source model under the Linux Foundation. Then a side experiment made more revenue over its launch weekend than the platform he had built for 2 years. He killed the original product, renamed the company, and grew Featherless to multiple seven figures in ARR within about a year, raising a Series A led by Airbus Ventures and AMD Ventures. Stay for 26:42 where he explains why removing information from his own homepage kept improving conversion, until they took their own research off the top of the page. SPONSORED BY HOBBES Hobbes runs personalized product demos on your website 24/7 and books qualified buyers straight onto your team's calendar. Don't book a demo, take one: https://saasclub.io/hobbes š KEY LESSONS š Let the experiment beat the plan: A side experiment out-earned the main platform over one launch weekend, so he rebuilt the company around what customers actually paid for. š§ Attachment to your own tech is the trap: The pivot was emotional, not technical. People wanted the models more than his model, and he was holding his own mission back. š° Flat pricing sells to the CFO, not the engineer: Per-token billing meant teams could not answer "what will this cost?" A fixed monthly rate unblocked procurement. šÆ Removing explanation improved conversion: Stripping the technical story off the homepage lifted conversion every time, even after they removed their own research. š Go where nobody is competing: The top 100 models have ten providers each. Beyond that, Featherless is usually the only one hosting them. š¤ First customers came from where the complaints already were: Reddit's LocalLlama and Ollama communities and Discord were full of people asking how to run models they could not host. ā” A constraint you solve for yourself can become the product: GPU hot-swapping was built to avoid buying thousands of GPUs. That workaround turned out to be the company. ā±ļø TIMESTAMPS 0:00 Introduction 1:30 What Featherless AI does 2:52 Starting as an open source model project 3:51 The GPU problem: one model per GPU 4:22 Building GPU hot-swapping 6:15 The experiment that beat the platform 11:42 Finding demand on Reddit and Discord 20:04 Realizing he was holding his own mission back 21:13 Why flat-rate pricing instead of per-token 26:42 Removing explanation, improving conversion 37:56 Hosting the long tail of models 39:31 Competing where no one else is 41:33 Lightning round š§ Full show notes: https://saasclub.io/490 š Join 5,000+ SaaS founders: https://saasclub.io/email Featherless AI: https://featherless.ai Eugene on X: https://x.com/picocreator #SaaS #AI #Startups #ProductMarketFit #AIInference