Flapping Airplanes raises $1B at a $6B valuation nine months after its seed

On Wednesday, Flapping Airplanes said it raised a $1 billion Series A round co-led by Index and Lux, reportedly at a $6 billion valuation. The company is a year old and has no product on sale.

The round lands nine months after the San Francisco lab launched with a $180 million seed from GV, Sequoia and Index at a $1.5 billion valuation. The new price is four times that. Bloomberg, which first reported the valuation, counts $1.18 billion invested in the company in total.

Sequoia, GV and Kleiner Perkins, all early backers, returned for the new round. Valor, Atreides, A* and Nvidia also appear on the investor list. Index partner Mark Xu joins the board, according to SiliconANGLE.

A team investors already knew

The company was founded by Aidan Smith and brothers Asher and Ben Spector. Smith is a former founder and Thiel fellow who worked at the brain implant company Neuralink during college. Asher Spector was a statistics PhD at Stanford. Ben Spector was a computer science PhD student at Stanford and founded Prod, a student-run startup accelerator, while at MIT.

Prod is five years old and has been described as something close to a secret society. Sequoia partner David Cahn, who invested in Flapping Airplanes, wrote earlier this year that Prod had "become one of the most powerful talent incubators in Silicon Valley." He said the founders "are building the 'young person's AGI lab.'"

That reputation carried the round as much as the technical pitch did. The company is a couple of dozen people working in person in San Francisco, and investors are paying for the team before there is revenue to measure.

One program, several jobs

The technical pitch centers on what the founders call a megakernel framework. A kernel is a program that runs on a GPU. A megakernel bundles many tasks into a single program so the chip spends less time switching between them.

In the founders' words, it is "the world's most complete megakernel framework. It can perform entire training runs, serve inference, play video games, and even do all three simultaneously in a single kernel." They say it "can overlap memory-bound inference with compute-bound training," which pairs a task that mostly waits for data with one that mostly does arithmetic. The chip sits idle less often.

The premise behind that work is that AI has barely scratched the surface of what GPUs can do. The founders argue intelligent systems can be built "without ingesting half the internet." The goal is models that cost less to train and run, which is where the data-efficiency mission comes in.

"A model that's a million times more data efficient is probably a million times easier to put into the economy," Asher Spector told TechCrunch shortly after the startup launched.

What the money has to prove

None of those capabilities have been shown in a product or measured in independent benchmarks. They are the founders' own claims, and the $6 billion price rests on them.

Asher Spector told TechCrunch in February that he could not give a timeline for commercialization, according to Tech Funding News. He told Bloomberg the lab intends to sell its technology eventually, so other companies can run AI more cheaply.

The near-term test is narrower than the mission. A megakernel has to produce measurable speedups on hardware customers already own, before anyone can judge whether overlapping training and inference works outside a demo.

Flapping Airplanes belongs to a small group of labs that raise at large valuations on research rather than products. Its first year went into low-level work on GPU primitives, and the company says that work makes new algorithms possible. The $1 billion buys time to show the rest of it.