Nvidia Corp. announced a sweeping product lineup at Computex 2026, introducing the Vera CPU, the Nemotron 3 Ultra AI model, and the Cosmos 3 physical AI foundation model. CEO Jensen Huang revealed that Anthropic, OpenAI, and SpaceX are among the first major users of the upcoming Vera processors. The new CPU delivers 1.8x faster performance for AI workloads compared to x86 chips, according to the company.
The announcements underscore Nvidia's push beyond graphics into domains spanning robotics, autonomous vehicles, and enterprise AI. Vera represents a direct challenge to Intel and AMD in data center compute, while Nemotron 3 Ultra targets the open-model AI race. Cosmos 3, an open AI world model, aims to help robots and autonomous systems better understand and predict real-world environments with limited training data.
Nemotron 3 Ultra is a 550B-parameter mixture-of-experts (MoE) open model, with roughly 55 billion parameters active at any given time. According to benchmarking firm Artificial Analysis, it is the smartest open U.S. model but trails the Chinese model Kimi K2.6. Additionally, Nvidia launched the Alpamayo 2 Super, which it calls its most powerful open AI model for robotaxis.
Early adoption by leading AI labs and a space exploration company signals strong industry confidence in Nvidia's roadmap. The Vera CPU could reshape data center economics for AI workloads, while Cosmos 3 and Alpamayo 2 Super may accelerate autonomous vehicle and robotics development. Nvidia also outlined a DGX Spark roadmap for laptops and desktops, suggesting the AI hardware push is extending into personal computing.
Critics note that Nemotron 3 Ultra still lags behind a Chinese competitor in raw performance benchmarks, and the Vera CPU's success will depend on software ecosystem support and adoption beyond Nvidia's existing customer base.