We’ve raised from a16z speedrun to keep building the software stack that makes any model run well on any hardware.
What we’re spending it on
Backend coverage, the agentic harness, and the benchmark infrastructure that keeps both honest.
What we’re deliberately not building
Not a model. Not a cloud. Not a serving product that only works if you move your whole deployment onto it. The bet is that the optimization layer is valuable precisely because it is portable.