The key facts
- Flapping Airplanes raised $1 billion in a Series A at a $6 billion valuation — just nine months after launching with a $180 million seed round at a $1.5 billion valuation
- The round was co-led by Index and Lux, with participation from Sequoia, GV, Kleiner Perkins, and Nvidia
- Founders Aidan Smith and brothers Asher and Ben Spector are developing a 'megakernel' framework for more efficient model training
A Record Round for a One-Year-Old Startup
Flapping Airplanes, a startup founded a year ago, announced it has raised $1 billion in a Series A round co-led by the Index and Lux funds. According to available information, the company's valuation reached $6 billion. The round came just nine months after the company launched with a $180 million seed round from GV, Sequoia, and Index, when the startup was valued at $1.5 billion — meaning the valuation has quadrupled. The new round also included existing investors Sequoia, GV, and Kleiner Perkins; the company's investors also include Valor, Atreides, A*, and Nvidia.
The Founders and Their Silicon Valley Connections
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 neural implant company Neuralink while in college. Asher Spector is a PhD student in statistics at Stanford, and Ben Spector is a PhD student in computer science at Stanford and the founder of Prod, created while he was studying at MIT. Prod is a five-year-old student startup accelerator that Sequoia partner David Cahn, an investor in Flapping Airplanes, called 'one of the most powerful talent incubators in Silicon Valley.' According to Cahn, the founders are building 'an AGI lab for the young.'
The Technology: a Megakernel Framework
The founders describe their development as 'the world's most complete megakernel framework,' capable of running full training cycles, serving inference, running video games — and even doing all of that simultaneously in a single kernel. A kernel is a program that runs on a GPU, and a megakernel combines many tasks into one program. According to the founders, the framework can 'overlap memory-bound inference with compute-bound training,' that is, combine tasks where the chip is mostly waiting for data with tasks where it is mostly computing — so the processor rarely sits idle. Their premise: AI has barely tapped the capabilities of GPUs, and with this approach intelligent systems can be built 'without swallowing half the internet.'
The Goal: Saving Data and Money
The startup aims for systems that are cheaper to train and operate. 'A model that is a million times more efficient with data is probably a million times easier to deploy into the economy,' Asher Spector told TechCrunch shortly after the company launched. Now the startup has a fresh billion dollars to hire staff and try to turn that goal into reality.
