Flapping Airplanes, a startup founded just one year ago, has garnered attention in Silicon Valley thanks to its impressive founding team and its mission to find ways to train large models with less data. On Wednesday, the company announced it has completed a $1 billion Series A round led jointly by Index and Lux, with a reported valuation of $6 billion.
This funding round occurred just nine months after the company’s launch. Flapping Airplanes initially raised $180 million in seed funding from investors including GV, Sequoia Capital, and Index, valuing the company at $1.5 billion—meaning its valuation has now quadrupled.
Other participants in this round of financing include early investors Sequoia Capital, GV, and Kleiner Perkins. Valor, Atreides, Kleiner Perkins, A*, and NVIDIA are also investors in the company.
Flapping Airplanes was founded by Aidan Smith, along with his brothers Asher Spector and Ben Spector. Smith previously founded a company, was a Thiel Fellow, and worked at Neuralink, a brain-computer interface company, during university. Asher Spector was a Ph.D. candidate in statistics at Stanford University, and Ben Spector was a Ph.D. student in computer science at Stanford, where he also founded Prod during his time at MIT.
Prod is a five-year-old, almost secret society-style student-run startup accelerator. In an article earlier this year explaining his investment rationale, David Cahn, a partner at Sequoia and investor in Flapping Airplanes, wrote that Prod has “become one of Silicon Valley’s most powerful talent incubators.” Cahn described the founders as building “a young people’s AGI lab.”
The founders describe their work as "the world's most comprehensive megakernel framework—it can handle the entire training process, provide inference services, run video games, and even perform all three simultaneously within a single kernel." (A kernel is a program running on a GPU, while a megakernel bundles multiple tasks into one program.)
The founder said this framework "allows memory-bandwidth-limited inference to overlap with compute-limited training," meaning it enables a task that spends most of its time waiting for data to run in parallel with a task that spends most of its time performing mathematical calculations, keeping the chip rarely idle. The core premise is that AI has only just begun to fully leverage GPU capabilities; they believe intelligent systems can be created without needing to "swallow half the internet" of data.
Its goal is to create systems with lower training and operational costs. Shortly after the company launched, Asher Spector told TechCrunch: “A model that is a million times more data-efficient could also be a million times easier to integrate into the economy.”
Now, they have secured an additional $1 billion in funding to hire talent and work toward making this goal a reality.