From OSS to startups: how vLLM, SGLang and FlashAttention teams monetize inference research

AccBalanced · x · 2026-11-21

sonnetxu maps a new category of AI infra companies emerging from open-source systems research: vLLM → Inferact (enterprise inference software/support), SGLang → RadixArk (managed inference and RL infra), Orca/continuous batching → FriendliAI (high-performance inference platform), speculative decoding work → Novita AI (decoding speedups and model serving), open kernels research → Together AI (compiler, kernels, full-stack platform). Key point: OSS success gives credibility, distribution and a tech head start, but customers and business models diverge.

Original post →

More from Venture

Venture channel →