NVIDIA

Poolside has concluded an agreement with NVIDIA valued at 12 billion dollars, described as an inverse execuhire: NVIDIA licenses Poolside's Model Factory and hires 109 of its employees, almost its entire technical staff. According to the founders, including Eiso Kant, the employees joining NVIDIA share approximately 6 billion dollars, while the founders, who remain at the head of Poolside, receive approximately 1 billion dollars. The company, whose latest model was built by fewer than 70 people and whose research and engineering efforts involved fewer than 115 in total, had already established an investment partnership with NVIDIA before this transaction.

At the end of 2025, Poolside had a six-week window to raise 2 billion dollars to finance a cluster of 40,000 GB300 GPUs scheduled to go into service in January 2026; the funding did not materialize in time and the cluster was lost. In parallel, Poolside Infraco, an infrastructure subsidiary spun off in January 2026, would aim for a capacity of 7 gigawatts for a neocloud. Summary and translation produced by Le Fil IA from Latent Space. AI by Google, Scale AI by Meta, or Instacart by OpenAI, where typically executives leave with a comfortable check, leaving the company and its employees behind.

Poolside, founded approximately three and a half years ago, became known for its foundation models intended for software development, with the ambition to compete with frontier labs despite more limited resources. The founders state that they are not yet ready to unveil the company's new direction, but several clues given in their communications suggest a vision where human-level intelligence capabilities will be largely commoditized by open models, while superintelligence will remain a domain reserved for a few players. They also distinguish two categories of economically valuable problems: those limited by intelligence, such as software development or accounting, which the increase in model power can solve, and those limited by real experimentation, such as medical research, which require empirical feedback loops that no artificial intelligence alone can fill.

This distinction could guide the new strategy of Poolside Infraco, whose stated ambition of 7 gigawatts of neocloud capacity suggests a pivot towards computing infrastructures rather than the development of models themselves, a field now occupied by NVIDIA itself.