A year-old artificial intelligence lab has joined the billion-dollar funding club. Flapping Airplanes has raised $1 billion at a $6 billion valuation, TechCrunch reported on Wednesday.
The startup's claim is about efficiency at the lowest level of the stack. Its framework uses a 'megakernel' — a single GPU program that can perform entire training runs, serve inference and even play video games simultaneously, overlapping memory-bound inference with compute-bound training so the chip is rarely idle.
The founders argue AI has barely scratched the surface of what GPUs can do, and that intelligent systems do not need to ingest half the internet to work. 'A model that's a million times more data efficient is probably a million times easier to put into the economy,' co-founder Asher Spector told TechCrunch when the startup launched.
The fresh $1 billion is earmarked for hiring and for turning that research premise into working systems. If the efficiency claims hold even in part, the prize is large: models that cost far less to train and run would change the economics of every company building on top of them.
