Why Investors Just Handed a Year Old AI Lab 1 Billion Dollars to Use Less Compute
Flapping Airplanes raised a $1 billion Series A at a reported $6 billion valuation to prove AI can be trained on far less data and far fewer GPUs. Its backers include Nvidia, the company selling the chips it wants to use more efficiently.
The AI industry has spent the past two years treating compute as destiny. Buy more GPUs, scrape more text. Flapping Airplanes just raised $1 billion on the opposite premise, and some of the biggest names in venture capital are paying up for it.
The San Francisco lab, founded in late 2025, announced a $1 billion Series A on Wednesday, co-led by Index Ventures and Lux Capital. Bloomberg reported the round values the company at $6 billion including the new money. Sequoia, Alphabet's GV and Kleiner Perkins also joined, according to TechCrunch. In January the company raised a $180 million seed at a $1.5 billion valuation. That is a fourfold jump in roughly nine months, for a company that, per Bloomberg, has no product on sale.
The bet against brute force
The pitch is efficiency at every layer. Flapping Airplanes wants to train large models on far less data and squeeze far more work out of each chip. Co-founder Asher Spector framed it to Bloomberg as a supply problem.
"The basic thesis is demand for AI is going to outpace chip and data supply." Asher Spector, co-founder, Flapping Airplanes
Spector added that wringing just 10% more performance out of each chip would be "tremendously economically valuable," Bloomberg reported. His earlier ambition, shared with TechCrunch shortly after launch, was far bigger: "A model that's a million times more data efficient is probably a million times easier to put into the economy."
The technical centerpiece is what the founders call "the world's most complete megakernel framework," software that bundles training, inference and other tasks into a single GPU program. They claim it "can overlap memory-bound inference with compute-bound training,". Co-founder Aidan Smith told Bloomberg the team has been writing its own version of the software that runs Nvidia's graphics chips. The founders say they want intelligent systems built "without ingesting half the internet."
An efficiency lab with Nvidia on the cap table
Here is the twist. Nvidia is an investor, according to TechCrunch. A startup whose thesis is that the industry wastes GPUs counts the world's dominant GPU seller among its backers. Index Ventures argues there is no conflict, writing in its announcement post that making chips "more flexible and more valuable" benefits "the companies making them" too. The logic holds if software makes each GPU worth more.
Index partners Mark Xu and Shardul Shah laid out the investor logic plainly.
"The winners in AI will be the companies that get the most out of every dollar, every megawatt, and every byte of data." Mark Xu and Shardul Shah, Index Ventures
Xu is joining the board. Index says it also led the seed, so this is a firm doubling down on its own early call.
Young, small and very expensive
The team is tiny. Index describes "just a couple dozen people," all working in person in San Francisco. Taken literally, that puts the valuation somewhere around $250 million per employee. The founders are young too. Ben Spector, the CEO, founded the student accelerator Prod while at MIT and was a computer science PhD student at Stanford. His brother Asher earned a statistics PhD at Stanford. Smith is a Thiel fellow who worked at Neuralink during college. Sequoia partner David Cahn called it "the young person's AGI lab."
Flapping Airplanes belongs to a growing class of so-called neolabs, research-first startups raising huge sums before shipping anything. TNW points to Yann LeCun's AMI Labs, which raised about $1 billion in March, and Mirendil, which raised $200 million in June.
A thesis, not a product
Spector told Bloomberg the company plans to eventually sell its technology to make AI cheaper for others. No timeline was given. For investors, a lab built around scarcity doubles as a hedge if the spending spree on chips and data centers hits limits. None of the claims about megakernels or million-fold data efficiency have been independently verified in public results so far. For now, investors are paying $6 billion for a thesis and a team. Whether the efficiency camp has found an edge or just a good story will depend on what this lab ships.
