Sources: Project SGLang spins out as RadixArk with $400M valuation as inference market explodes

$400M for an inference optimization library is wild. SGLang's tech is legitimately clever—token scheduling, smart caching, dynamic batching.

But inference is commoditizing. OpenAI's got vLLM, there are fork fests everywhere, everyone's solving the same problems.

This valuation assumes a moat that doesn't exist yet. First-mover advantage in infrastructure doesn't last.