Flapping Airplanes raises $1B to train AI on less data

Flapping Airplanes, an AI research lab about a year old, raised $1 billion in a Series A round on October 21, 2026. Index Ventures and Lux Capital co-led the round. Bloomberg reported that it values the lab at $6 billion, roughly four times its value in January, though the company has not released a product.

The round and its backers

The raise was announced by the company, TechCrunch reports. The $6 billion figure comes from Bloomberg, not from Flapping Airplanes itself. SiliconANGLE and TechCrunch both describe the valuation as reported.

Detail	Figure
Round	Series A, $1 billion
Co-leads	Index Ventures, Lux Capital
Reported valuation	$6 billion (Bloomberg)
Earlier round	$180 million seed in January, at $1.5 billion
Team size	"A couple dozen" people, in person in San Francisco

Sequoia and GV, Google's venture arm, returned from the seed round, SiliconANGLE reports. Kleiner Perkins, Valor, Atreides, Jump Trading, XTX Ventures and Nvidia are also named as investors in the company. Mark Xu of Index Ventures joins the board.

What the lab is trying to build

Flapping Airplanes wants AI models that need far less training data. Today's large models learn from huge amounts of text, which makes them costly to train and run. The founders say they want to get there "without ingesting half the internet," TechCrunch reports.

Co-founder Asher Spector told TechCrunch: "A model that's a million times more data efficient is probably a million times easier to put into the economy."

The lab spent most of its first year on low-level GPU software, SiliconANGLE reports. Its main result is a "megakernel" framework. A kernel is a small program that runs on a graphics chip, and a megakernel runs an entire program on the GPU as one kernel. The founders call it "the world's most complete megakernel framework" and say it can run full training jobs and serve a model's answers.

Index Ventures, in its own post on the deal, says the team works across GPU kernels, training and inference. It also builds internal agent tools that automate much of the team's work. Index calls it possibly "the most vertically integrated company per capita in AI."

Who runs it

The lab has three co-founders, according to TechCrunch and SiliconANGLE:

• Ben Spector, the chief executive, is a computer science PhD student at Stanford. Ben started the student accelerator Prod while at MIT and led ThunderKittens, an open-source toolkit for fast GPU kernels, in 2024.
• Asher Spector, Ben's brother, holds a statistics doctorate from Stanford.
• Aidan Smith is a Thiel Fellow who worked at Neuralink during college.

Sequoia partner David Cahn says the founders "are building the 'young person's AGI lab.'"

What this means for developers

A billion dollars for a lab with no product is a bet on research, and the useful signals for developers sit in that research.

• Watch the megakernel work. ThunderKittens is already open source. If the lab publishes more of its GPU tools, teams running their own models could get faster training and inference on the same hardware.
• Data efficiency changes who can train models. If models need far less data, smaller companies with private datasets could train useful models without scraping the web.
• Treat the valuation as reported, not confirmed. The company announced the amount but not the $6 billion price. Any comparison with other labs should say so.
• Expect hiring pressure. The funding is for hiring, TechCrunch reports, and a couple dozen people now hold $1 billion. Expect the lab to compete hard for GPU-kernel specialists.