Company teardown · Flapping Airplanes
Flapping Airplanes raised $1.18 billion for a twenty-person lab that publishes nothing. One of its hires came straight from high school.
Flapping Airplanes raised $1 billion in October 2026, nine months after a $180 million seed. Its pitch is models that need a thousandth of the data. Its record is GPU systems: the CEO first-authored Stanford's megakernel work, and the product is a megakernel framework. Its investor Index says a hire who joined from high school built the agent infrastructure. Cursor runs through the founders' story in five specific places, the investors sit in the hiring loop, and the lab publishes nothing, on purpose. We read all twenty employee profiles, two founder interviews and four investor posts. The lab sells a data-efficiency dream; what it has shown is GPU craft.
0.1 — The whole story · start dates from all 20 current profiles
Three people in October 2025, twenty a year later. In nine of those months the valuation went from $1.5 billion to $6 billion.
— The 20 profiles read, by the year they joined
Everything here is public: LinkedIn, two founder appearances, four investor posts and the funding press. Three profiles give a start year without a month. The reading is ours.
0.2 — The pitch and the product · skip if you know what a GPU kernel is
The pitch is models that learn from far less data. The product is a GPU program that never stops to wait.
— How models run today
Each step of a model is a separate GPU program, a kernel. Between kernels the chip sits idle, waiting for the next one to load.
— A megakernel
The whole program runs as a single kernel, so the gaps disappear. Flapping says its framework can train, serve inference and play a video game in one kernel at once. No public code or benchmark yet.
§ 01 — Part one of four · The founders
A GPU-kernel engineer, a statistician and a Neuralink programmer. The first one's papers explain the company.
Who the founders are, the five places Cursor touches their story, and why a lab worth $6 billion has published nothing.
1.1 — The founders
A kernel engineer, a statistician and a brain-interface programmer. Only one has published on the GPU work the company builds.
1.2 — The Cursor thread · five specific links
Cursor comes up in almost every account of the company. Here are the five places it actually touches it.
1.3 — No papers, on purpose
No papers, no public code, no benchmark. The founders say the literature is the thing to get away from.
— Aidan Smith, TechCrunch, February 2026
“They think about things in a way that many established researchers just can't because they haven't been polluted by the context of thousands and thousands of papers.”
Self-report. Asher Spector on the algorithms: “We're not going to talk about them cuz they're our core IP.” Ben Spector: “You don't need two PhDs.”
Incremental work, Ben Spector says, forces you “very far up the scaling ladder” before you learn anything. The absence of publications is the strategy, not a gap in the data.
Talent brief
Want the full visual breakdown?
Download the PDF version of this teardown: the headcount chart month by month, the megakernel drawn out, the five Cursor links, and five lessons for founders building a research team from scratch.
§ 02 — Part two of four · The roster · October 2026
Seventeen of twenty do research. The deepest skill on the team is making GPUs work harder.
The two hires the investors point to, the student core, the credentials, and the systems engineers around them.
2.1 — Two hires the investors point to
Index says one hire joined from high school and built the agent infrastructure. Another had never trained a model and ran the pre-training.
“They hire for skill and trajectory over age or pedigree, finding talent in pockets most companies overlook.”
Mark Xu and Shardul Shah, Index Ventures, Series A post, October 2026.
2.2 — The student core · eight researchers, enrolled or just out
The research bench is mostly undergraduates, hired for competition records rather than publication records.
Student researchers are not named. The credentials, verified against official results: two Science Talent Search finalists, one Putnam and USAMO medallist, a physics-olympiad honourable mention and two national debate champions (the Chief of Staff and Asher Spector). There are no IMO or IOI medals on the roster.
2.3 — The systems layer · five hires, February to September 2026
The research lab's deepest bench is GPU and systems engineering. All five of these hires arrived in 2026.
Index: the team “works across the stack, from the kernels that run on GPUs through pre-training, post-training, and inference.” With the CEO, this layer is the lab's proven expertise: making GPUs work harder, not yet using less data.
§ 03 — Part three of four · How they hire
No recruiter on the roster. The investors say they help with hiring.
Investors who say they meet candidates and whose posts double as job ads, the networks the twenty came from, and the three people who keep the lab running.
3.1 — The investors in the hiring loop
Sequoia's partner says he met nearly every candidate. The investors are part of the hiring process, not just the cap table.
Nobody on the twenty-person roster has a recruiting title. Founders and, by their own account, investors do the recruiting.
3.2 — Where the 20 came from · schools, labs and employers anywhere on the CV
Twelve of twenty have Stanford on the CV. The rest came through quant desks, an emulator project and the founders' incubator.
— Three people run everything that isn't research
Asher Spector: “If we start by signing big enterprise contracts, we're going to get distracted.” None of the three sells, because there is nothing to sell yet.
§ 04 — Part four of four · What to make of it
$300 million of valuation per employee. The price assumes the gaps close.
What the public record leaves out, and what we took from it.
4.1 — What the record doesn't show
Five gaps in the evidence, and a price set before any of them closed.
4.2 — What we learned
Five moves worth copying.
— The takeaway
Twenty people, $6 billion, no papers.
The bet is that GPU craft and a young bench beat the literature.
Methodology & limitations
All twenty profiles read, and every self-report labelled.
— Sources
- The LinkedIn profiles of all 20 Flapping Airplanes employees, collected October 13–14, 2026. Three investor partners on the People page are left out of every count.
- Degrees, graduation years and competition results from personal sites, department pages, MAA and Regeneron results and thesis records; Google Scholar and Semantic Scholar citations.
- Two founder appearances: TechCrunch and Sequoia AI Ascent.
- Investor posts from Sequoia, GV and Index (seed, Series A); TechCrunch and Bloomberg for the Series A; Forbes for the July talks; Cursor's Mixture-of-Kittens post and paper; Neo's account via Dealroom.
— Limitations
- Founder quotes are self-report. Investor claims are neither self-report nor independent, and are labelled.
- Profile locations conflict with Index, which says everyone works in person in San Francisco.
- The lab publishes nothing, so publication and code signals undercount the team by design.
- Groupings (the student core, the systems layer) are our reading of titles and CVs, not an org chart the company publishes. Founders are named as in the press and other staff as they show themselves on LinkedIn; student researchers are not named.
- Citation counts move daily; they are rounded and dated to collection. Teardowns like this are how our searches begin — this one's on us.