Flapping Airplanes raises $1 billion as Silicon Valley bets on training models with less data

While AI labs continue their frantic race to feed models with billions of digital texts, the Silicon Valley startup “Flapping Airplanes” leapt to the top of the investment scene with a completely opposite thesis based on training intelligent systems with less data. After only one year since its founding, the company announced the close of a first funding round worth $1 billion, jointly led by Index and Lux, lifting its estimated market value to $6 billion, which is four times its previous valuation recorded just nine months earlier.

Quadrupling the valuation in nine months reveals the size of the bet on breaking the big-data equation.The jump was not accidental; the company launched an initial round of $180 million with participation from GV, Sequoia and Index, valuing it at roughly $1.5 billion at the time. In the new round, previous investors increased their stakes alongside Sequoia, GV and Kleiner Perkins, and prominent investors such as Valor, Atridis, A Star and Nvidia joined, reflecting a consensus among major infrastructure financiers on the viability of this architectural shift.

Behind the project are three founders who combine advanced research backgrounds with early entrepreneurial experience: Eden Smith, a former fellow in the Thiel program who worked during his university studies at the brain-chip company Neuralink, and the brothers Asher and Ben Specter. Asher holds a PhD in statistics from Stanford University, while Ben was a PhD researcher in computer science at the same university after founding the accelerator “Broad” while at the Massachusetts Institute of Technology; David Kahn, a partner at Sequoia and an investor in the company, described Broad as one of the strongest talent incubators in Silicon Valley, considering the team a nucleus for building a general-AI laboratory led by a young generation.

Merging training and inference tasks within a single software core eliminates wasted processing cycles in chips.The company describes its technology as the most complete framework in the world for what is known as the giant core, or a unified operating system on graphics processors. Instead of separating computing tasks, this framework enables full training runs, inference services, and even video-game execution simultaneously on a single core. The mechanism relies on overlapping inference operations, which are normally stalled waiting for data to be transferred from memory, with computationally intensive training, so the chip performs arithmetic while data is being fetched, eliminating processor idle periods that drain the efficiency of conventional servers.

The founders of Flapping Airplanes start from a technical hypothesis that the industry has tapped only a small fraction of the actual capabilities of graphics processors, and that achieving ultra-intelligent systems does not require consuming half of the Internet’s data. Asher Specter summarized the vision, noting that a model that is a million times more data-efficient would be a million times easier to integrate into the economic cycle. The ultimate goal here is not limited to shortening compute time, but aims to dramatically reduce the capital and operating costs of building and running models.

This shift in chip-efficiency philosophy reshapes the technical infrastructure priorities of data-center operators and institutions across the Arab region. In the Gulf, where billions are poured into building ultra-high-performance computing centers and attracting the latest GPU clusters, the main operational challenge remains the energy-consumption cost and chip idle rates during data exchange; therefore, the ability to squeeze silicon power and synchronize training with inference means reducing the capital bill for sovereign projects and major banks. In Egypt and the Levant, where startups and development teams face financing constraints that prevent them from leasing massive compute clusters in hard currency, the move toward integrated software cores opens an alternative path that emphasizes computational efficiency and hardware engineering rather than betting on server stacking, making competitiveness depend on advanced architectural-optimization skills.

The company now has fresh cash liquidity of $1 billion to hire talent and turn its architectural promises into tangible production reality. The real test of this massive bet will be proving that these giant cores can compete with established software frameworks in global data centers, and confirming whether pure computational efficiency can indeed dispense with the endless data-gobbling appetite of open-ended models.