A year-old San Francisco AI lab just raised a $1 billion Series A, reportedly at a $6 billion valuation, betting that smarter use of GPUs can replace training on half the internet.
Most frontier AI labs grow by feeding models more data and more compute. Flapping Airplanes is betting on the reverse. On Oct. 21, the year-old startup announced a $1 billion Series A. TechCrunch reports the valuation at $6 billion.
The round
- Size: $1 billion Series A.
- Leads: Index Ventures and Lux Capital.
- Other participants: earlier backers Sequoia, GV and Kleiner Perkins. TechCrunch also lists Valor, Atreides, A* and Nvidia among the company’s investors.
- Valuation: reportedly $6 billion.
The speed stands out. The company launched about nine months ago with a $180 million seed round from GV, Sequoia and Index that valued it at $1.5 billion. The new round quadruples that in under a year. Crunchbase News listed it among the week’s 10 biggest funding rounds.
Who’s behind it
Flapping Airplanes was founded by Aidan Smith and brothers Asher and Ben Spector. According to TechCrunch:
- Aidan Smith is a former founder and Thiel Fellow who worked at brain-implant company Neuralink while in college.
- Asher Spector earned a statistics PhD at Stanford.
- Ben Spector was a computer science PhD student at Stanford and, while at MIT, founded Prod, a student-run startup accelerator.
Sequoia partner David Cahn, an investor, has described Prod as one of Silicon Valley’s most powerful talent incubators and called the founders’ project “the ‘young person’s AGI lab.'”
The technical bet: megakernels
At the center of the company’s pitch is what it calls “the world’s most complete megakernel framework.” A kernel is a program that runs on a GPU. A megakernel bundles many tasks into one program. The founders say theirs can run full training jobs, serve inference and even play video games, all at once, inside a single kernel.
The key claim is overlap. Inference is often memory-bound: the chip spends its time waiting for data. Training is compute-bound: the chip spends its time doing math. The company says its framework can run both together, so the GPU rarely sits idle.
Why data efficiency matters
The larger aim is models that cost less to train and run, built “without ingesting half the internet,” as the company puts it. Crunchbase News describes it as an AI lab focused on models that are cheaper to train and run.
Asher Spector made the business case to TechCrunch: “A model that’s a million times more data efficient is probably a million times easier to put into the economy.”
If that holds up, it would change the economics for everyone else. Startups and enterprises could adapt models without huge datasets or compute budgets, and AI would rely less on scraping the web, which is under growing legal and ethical pressure.
What to watch
- Proof. So far the claims are the company’s own. Published benchmarks or open releases would show whether the megakernel approach beats conventional training stacks.
- Hiring. TechCrunch notes the new money is mainly for hiring. Watch who joins.
- Nvidia’s role. Having the leading GPU maker as an investor gives a lab built on squeezing more out of GPUs a close hardware partner.
Our take
A $6 billion valuation for a year-old company with no product on the market is a bet on people and an idea. Data efficiency is one of the most important open problems in AI. If Flapping Airplanes makes real progress on it, its tools could matter far beyond its own models.
