A Pre-Product AI Lab Just Raised $1 Billion at a $6 Billion Valuation

Flapping Airplanes, a roughly year-old San Francisco AI research lab with no commercial product and no disclosed revenue, has closed a $1 billion Series A at a $6 billion valuation, according to a Bloomberg report corroborated by TechCrunch, SiliconAngle, TheNextWeb and other outlets this week. Index Ventures and Lux Capital co-led the round, with Sequoia, Alphabet's GV, Kleiner Perkins, Valor Equity, Atreides, Jump Trading and Nvidia also putting in money. That brings the company's total raised to roughly $1.18 billion since its $180 million seed round in January, when it was valued at about $1.5 billion — a fourfold jump in nine months with nothing on the market to show for it.

The pitch: AI is learning the wasteful way

Founders Ben Spector, Asher Spector and Aidan Smith argue that today's frontier models are absurdly data-hungry — by their own estimate, humans learn 100,000 to 1,000,000 times more efficiently than current systems that are trained on enormous slices of the internet. Their first year of work reportedly went not into a product but into GPU-level plumbing: a 'megakernel' framework that runs an entire program, including training and inference and even a video game, as a single kernel on one chip. It's a legitimate systems-engineering feat — Ben Spector co-authored the 2024 ThunderKittens GPU-kernel paper — but it's infrastructure for a thesis, not evidence the thesis works at model scale.

A model that's a million times more data efficient is probably a million times easier to put into the economy.

A valuation built on pedigree and a scarcity story, not traction

What's actually being priced here is not a product but a story: that AI demand will keep outrunning the supply of training data and chips, and that whoever cracks data-efficient learning first captures enormous value in robotics, trading and life sciences. Investors are underwriting that story largely on founder pedigree and on a pattern that's repeating across the industry. Yann LeCun's AMI Labs raised roughly $1 billion in March on a similar 'rethink how machines learn' pitch. Ex-OpenAI researcher Jerry Tworek's Core Automation has reportedly been seeking $500 million to $1 billion for a related bet. Thomas Reardon's Flourish and the smaller Mirendil have drawn comparable checks. None of these companies has shipped a product; several of the numbers describing them (valuations, raise totals) trace back to a single Bloomberg report rather than company disclosure, which is worth flagging rather than treating as confirmed fact.

What isn't known

There is no independent benchmark showing the megakernel framework or any resulting model actually delivers the data efficiency gains the founders describe — the claims so far are the company's own framing, repeated by press. There's no stated timeline for a sellable product, no disclosed revenue, and no detail on how the $6 billion figure was derived beyond Bloomberg's sourcing. And the core wager — that chip and data scarcity will bite hard enough to make a 10% efficiency gain per chip 'tremendously economically valuable,' as Asher Spector put it — assumes that cheaper inference and falling model prices elsewhere in the market (price wars among Chinese open-weight labs, for instance) don't undercut the scarcity premise before Flapping Airplanes has anything to sell against it.