Flapping Airplanes raises $1bn a year after its founding
The pre-product lab works on training models with far less data and on GPU kernel software; Bloomberg, not the company, reported the $6bn valuation.
- Money & business
- Labs & people
- Minor
Flapping Airplanes, a San Francisco AI research lab founded about a year ago, announced a $1 billion funding round on 21 October. SiliconANGLE reported the company’s announcement of the raise; its lead investor, Index Ventures, published a post calling it a $1 billion Series A. SiliconANGLE reported that Lux Capital co-led with Index, that Sequoia Capital and GV returned from the seed round, and that Kleiner Perkins, Valor Equity Partners, Atreides, Jump Trading, XTX Ventures and Nvidia also took part.
The valuation is not from the company. Bloomberg reported it as $6 billion, and SiliconANGLE and The Next Web both cite Bloomberg for it. By Bloomberg’s count, according to SiliconANGLE, that is roughly four times the level at a $180 million seed round in January, for $1.18 billion raised in total. Index’s post does not mention a valuation, and none of the coverage reviewed shows the company confirming the figure.
The lab has no commercial product. Its stated aim is models that learn from much less data than current frontier systems; Index’s post describes work aimed at AI that is “orders of magnitude more efficient”, spanning GPU kernels, pre-training, post-training and inference. SiliconANGLE reported that most of the first year went into new GPU programming primitives, including what the company calls a megakernel framework, which runs a whole program on a GPU as a single kernel to reduce idle time between steps. The company says it can run a training job, inference and a video game at once in one kernel. Those are the company’s claims; no benchmark or independent test was cited.
Ben Spector is chief executive; the other co-founders are Asher Spector and Aidan Smith. Asher Spector, who holds a Stanford statistics doctorate, said that “a model that’s a million times more data efficient is probably a million times easier to put into the economy”. Index describes a team of a couple of dozen people working in person, and its partner Mark Xu is joining the board.
Index’s partners Mark Xu and Shardul Shah argue, according to SiliconANGLE, that demand for AI will keep outpacing the supply of compute and data, and they name robotics, trading and life sciences as markets that need more efficient models. The company has not shown a model yet, so the investment rests on that thesis and the team rather than on a released result.