Zettafleet, a stealth AI startup co-founded by Cambridge University scientists, has launched its first product: a proprietary large language model (LLM) training framework designed to reduce dependence on Nvidia’s dominant GPU hardware. The technology claims to cut training costs and energy consumption by optimizing data processing workflows, potentially disrupting Nvidia’s $300 billion market share in AI infrastructure.

CEO Tzcpyj emphasized the system’s efficiency, stating it achieves 34-68% faster training speeds while using 40% less power compared to conventional methods. Investors, including former Google AI engineers, have reportedly injected $15 million into the startup to scale the technology, which could reshape data center economics for generative AI firms.