Flapping Airplanes Raised a Billion Dollars Nine Months After Launch: the AI Lab Is Reportedly Worth 6 Billion, Four Times More Than at the Start

Nine months ago Flapping Airplanes appeared with a seed round of 180 million dollars and a valuation of 1.5 billion dollars. This Wednesday the artificial intelligence lab announced a Series A round of one billion dollars, led by the funds Index and Lux, and according to reports its valuation is now 6 billion dollars. Four times more in less than a year. What exactly is being bought at that price?

Returning to the table are early investors Sequoia, GV and Kleiner Perkins, and the backers also include Valor, Atreides, A* and Nvidia. That last name is worth noting: the company that sells the chips is investing in a lab whose whole idea is to squeeze much more out of graphics processors.

The founders are Aidan Smith and brothers Asher and Ben Spector. Smith is a former founder and fellow of Peter Thiel's programme who, as a student, worked at Neuralink, the brain implant company. Asher Spector was doing a PhD in statistics at Stanford, and Ben was a computer science PhD student, also at Stanford, and while still studying at MIT founded Prod - a five-year, almost secret student programme for startups. "It became one of the most powerful talent incubators in Silicon Valley," wrote investor David Cahn of Sequoia, who says the three are building "an AGI lab for young people."

The founders claim they have "the world's most complete megakernel framework" - a single program on the graphics processor that can run an entire model training, answer requests and even play video games, all at the same time. The idea is that the chip almost never sits idle: while one task waits for data from memory, another computes. The ultimate goal is intelligence built "without swallowing half the internet."

"A model that's a million times more data-efficient is probably a million times easier to bring into the economy," Asher Spector said when the startup launched. In a world where the biggest labs spend billions on data centres, cheaper training is a real question. But a billion dollars nine months after launch is a bet on a promise, not on a proven result.

The new money will go on hiring and research. If the idea works, it's bad news for those building ever-bigger data centres. And if models really do become many times cheaper to train, could smaller markets, like the Balkans, finally train their own models instead of renting someone else's?