[Digital Today reporter Jinju Hong] Nvidia unveiled Nemotron 3 Ultra, an ultra-large open artificial intelligence (AI) model with 550 billion parameters, and announced it has started mass production of its next AI server platform Vera Rubin. The company is moving to strengthen AI models and data centre infrastructure at the same time, accelerating its push into the enterprise AI market.

On June 1 local time, online media outlet Gigazine reported that Nvidia CEO Jensen Huang (젠슨 황) unveiled Nemotron 3 Ultra and the Vera Rubin platform in a keynote speech at Nvidia GTC Taipei 2026.

Nemotron 3 Ultra is a large language model (LLM) with 550 billion parameters. Nvidia said it plans to release it as an open model, with the actual release expected to take place this week.

Nvidia introduced Nemotron 3 Ultra as having top-tier performance among open models developed by U.S. companies. Presentation materials said the model showed competitiveness across multiple benchmarks comparing it with major open models including China's GLM 5.1, Kimi K2.6 and Qwen 3.5. Nvidia stressed it delivers better performance than Chinese competing models, especially in cost efficiency.

External assessments also showed a gap with leading Chinese models. Nemotron 3 Ultra scored 48 points on the Artificial Analysis intelligence index, or AAI, published by AI evaluation organisation Artificial Analysis. That was well above the 39 points recorded by Google's Gemma 4 31B, but fell short of the 54 points for the Chinese model Kimi K2.6.

Nvidia emphasised practical usability over pure performance figures. It said Nemotron 3 Ultra showed higher performance in tokens generated per second than competing models with similar intelligence scores. It is positioning processing speed and cost efficiency in real service environments as strengths beyond simple benchmark competition.

Another key part of the announcement was the start of mass production of the next AI server platform Vera Rubin. Vera Rubin is a data centre AI server platform that integrates high-performance storage and network systems around Nvidia's next AI GPU Rubin and its in-house AI CPU Vera.

Nvidia said in particular that Vera Rubin provides higher processing efficiency than previous-generation systems in agent AI environments. As the AI industry evolves beyond simple chatbots to agent AI that performs tasks on its own, related data centre demand is also rising quickly.

Industry watchers are focusing on the fact that the announcement was not simply the separate release of a new AI model or a server. They see it as an assessment that Nvidia has further fleshed out a strategy to build an ecosystem spanning everything from AI model development and inference to data centre infrastructure supply.

At the event, Huang also unveiled new products targeting the Windows 11-based AI PC market. They include the laptop system-on-chip RTX Spark, which combines an Arm-based CPU with an Nvidia GPU, and the high-performance AI workstation DGX Station.

Industry attention is centred on Nemotron 3 Ultra and Vera Rubin. As open-model competition expands into a contest for leadership between Chinese and U.S. companies, Nvidia is seeking to strengthen its presence by emphasising model performance, inference speed and cost efficiency.

Key points to watch going forward are how licensing policy and scope of use will be set after Nemotron 3 Ultra is actually released, and how quickly Vera Rubin can absorb demand in the enterprise AI and agent AI markets after full-scale mass production begins.

Industry watchers see the announcement as another example confirming Nvidia's strategy to expand market dominance by tying AI software and hardware into a single ecosystem.

Nemotron 3 Ultra is coming this week. ⌛️ pic.twitter.com/ShTQv4nTtn