Flapping Airplanes Raises $1bn, Reportedly at $6bn

Flapping Airplanes, an AI lab barely a year old with a couple of dozen staff, has raised $1bn on the premise that today’s models waste most of what GPUs can do. Index Ventures said on 21 October that it was leading the Series A, with partner Mark Xu joining the board.

TechCrunch reports that the round was co-led by Index and Lux and came “reportedly at a $6 billion valuation”. Neither the company nor Index has confirmed that figure. If it holds, it would mean the valuation has quadrupled in about nine months: TechCrunch says the lab launched with a $180m seed round from GV, Sequoia and Index that valued it at $1.5bn.

TSN tracks major AI investments on its AI funding rounds hub.

Who is backing it

Earlier backers Sequoia, GV and Kleiner Perkins took part in the new round, according to TechCrunch, which also lists Valor, Atreides, A* and Nvidia among the company’s investors. It does not say how the $1bn is split.

The lab was founded by Aidan Smith and the brothers Asher and Ben Spector. TechCrunch says Smith is a former founder and Thiel fellow who worked at Neuralink while at university; Asher Spector earned a statistics PhD at Stanford; and Ben Spector, the chief executive, was a computer science PhD student at Stanford who founded Prod, a student-run start-up accelerator, while at MIT. Sequoia’s David Cahn has described the venture as “the young person’s AGI lab”.

The bet on efficiency

Index says the founders started the company believing that “AI could be orders of magnitude more efficient”. Their tool for that is the megakernel. A kernel is a program that runs on a GPU, and a megakernel, in Index’s words, runs “an entire program on a GPU as a single kernel”, cutting the idle time between steps.

The founders call their system “the world’s most complete megakernel framework”, TechCrunch reports, and claim it “can overlap memory-bound inference with compute-bound training”, so the chip is rarely left waiting. The longer-term aim is models that learn from far less data, built “without ingesting half the internet”. “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 after the launch.

The team works across the whole stack, from GPU kernels through pre-training, post-training and inference, all in person in San Francisco, according to Index. It singles out an engineer who joined while still at school and built the agent infrastructure that now automates much of the team’s work, and another who had never trained a model before leading the company’s pre-training run.

Why investors care

Index’s case is that compute and data will stay scarce relative to demand, so “the winners in AI will be the companies that get the most out of every dollar, every megawatt, and every byte of data”. It names robotics, trading and life sciences as markets where AI is still too costly or data too thin.

The claims so far are the founders’ and their investors’. Flapping Airplanes has not published a product, and $1bn is a large sum for a lab of a couple of dozen people. What it ships with that money will show whether efficiency can carry a $6bn price.