Flapping Airplanes Secures $1B at $6B Valuation to Build Data-Efficient GPU Megakernels
AI research lab Flapping Airplanes raises $1 billion to develop megakernel frameworks and data-efficient models, pushing its valuation to a reported $6 billion.
San Francisco-based AI research lab Flapping Airplanes Inc. has raised $1 billion in a new funding round, reaching a reported $6 billion valuation just twelve months after its founding. Bloomberg first reported the valuation for the Series A deal, which brings the young company's total funding raised to date to $1.18 billion. Despite having not yet launched a commercial product, the lab has commanded significant investor interest by targeting one of artificial intelligence's most persistent bottlenecks: the extreme data and compute requirements of modern frontier models.
Redefining GPU compute with megakernels
Flapping Airplanes spent much of its initial year developing fundamental new primitives for graphics processing units. Announcing the progress on X, the team claims this foundational work paves the way for new algorithms capable of drastically reshaping the overall cost structure of artificial intelligence. At the core of their technical approach is what the founders describe as the world’s most complete megakernel framework.
In an overview published by Index Ventures Management, a megakernel operates by executing an entire program on a GPU as a single, unified kernel. This approach prevents the processing chip from losing valuable execution time between separate steps. According to Flapping Airplanes, its custom framework can concurrently handle a complete model training run, serve real-time inference, and run a video game—all within a single kernel on the same processor. By enabling memory-bound inference tasks to overlap directly with compute-bound training workloads on a single chip, the system aims to maximize hardware utilization.
The technical foundation relies on a small, highly specialized team of roughly two dozen people working in person in San Francisco. Co-founder and CEO Ben Spector, currently a computer science doctoral student at Stanford University, previously founded the student accelerator Prod at MIT and was lead author on ThunderKittens, a 2024 open-source toolkit designed for writing fast GPU kernels. His brother, co-founder Asher Spector, holds a doctorate in statistics from Stanford. The third co-founder, Thiel Fellow Aidan Smith, previously worked at Neuralink Corp. during his college years.
Investor backing and data efficiency
The startup's main objective is creating AI systems that require dramatically less training data than current industry standards. Speaking to TechCrunch, Asher Spector emphasized the practical implications of this research, noting, "A model that's a million times more data efficient is probably a million times easier to put into the economy."
Investors have backed this thesis aggressively. The $1 billion Series A round was co-led by Index Ventures and Lux Capital Management, with Mark Xu of Index taking a seat on the board. The deal roughly quadruples the startup's valuation compared to its $180 million seed round completed in January. Early investors Sequoia Capital Operations and GV Management Co. both returned to participate in the new round. Additional institutional backers listed by the startup include Nvidia Corp., Kleiner Perkins Caufield & Byers, Valor Equity Partners LP, Atreides Management LP, Jump Trading Group, and XTX Ventures.
Index partners Mark Xu and Shardul Shah noted that market demand for artificial intelligence continues to expand faster than the available supply of both data and hardware compute. They specifically highlighted fields like robotics, quantitative trading, and life sciences as key sectors that urgently require more efficient AI paradigms.
What it means for developers
If Flapping Airplanes successfully proves its megakernel model, the underlying economics of model training and inference could shift noticeably. Today, hardware constraints and GPU context-switching costs contribute heavily to high API pricing and infrastructure expenses for software engineers. Unifying workloads onto single GPU kernels and dramatically reducing required data volumes could lower the operational bar for deploying advanced AI.
While research labs work on long-term hardware and algorithmic efficiency, engineers today still need practical access to current frontier infrastructure. For instance, developers can try top AI models cheaply through one API at https://apixoai.online to build and test applications across various providers without managing complex infrastructure.
As Flapping Airplanes transitions from underlying primitive engineering to building practical models, the broader developer community will be watching to see if low-data, single-kernel AI can deliver on its promises. If successful, the lab's innovations could make high-performance models vastly more accessible across resource-intensive industries.
Source: Year-old AI research lab Flapping Airplanes raises $1B at a reported $6B valuation — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
