Nvidia Launches Blackwell Ultra and Rubin AI Chips

Nvidia's recent announcements at the annual GTC conference have made waves in the tech community. CEO Jensen Huang introduced two new families of chips: Blackwell Ultra and Vera Rubin. This marks an important step in Nvidia's ongoing evolution in the artificial intelligence space.

The Blackwell Ultra chips are set to ship in the second half of this year and promise enhanced performance for AI tasks. They are designed to generate more content within a shorter time frame compared to their predecessors. According to Nvidia, these chips could allow cloud providers to significantly increase their revenue from AI services.

On the other hand, the Vera Rubin setup, expected in 2026, consists of a custom CPU named Vera and a next-generation GPU called Rubin, named after the famous astronomer Vera Rubin. This combination aims to push boundaries with its computational capacity, expected to manage 50 petaflops of inference, doubling the capabilities of the current model.

Nvidia's sales have skyrocketed, increasing more than sixfold since the launch of OpenAI's ChatGPT in late 2022. This surge is largely due to the company's dominance in the GPU market, especially for AI development and training purposes.

Developers and investors are closely monitoring the new chips to determine if they provide enough performance to encourage major clients, like Microsoft, Google, and Amazon, to invest further in data centers built around Nvidia technology.

This year marks a shift in Nvidia's release strategy, as the company aims to unveil new chip families annually, moving away from its previous biannual cycle. The GTC conference, held in San Jose, California, had a record attendance, showcasing the growing interest in AI technology and its applications.

Overview of Vera Rubin

The Vera Rubin system combines a custom-designed CPU and GPU to enhance AI workloads. The new CPU design is expected to significantly outpace previous versions, achieving double the performance of last year's models.

The Rubin GPU itself represents a major leap forward, delivering a substantial increase in performance for AI inference tasks. Nvidia's decision to develop a custom CPU stems from the lesson learned by other tech giants that customized designs can optimize performance more effectively than off-the-shelf solutions.

Introduction to Blackwell Ultra

The Blackwell Ultra series is designed to increase processing capacity, enabling faster content generation. This capability is essential for cloud service providers looking to enhance their AI offerings. The chips will also come in various configurations to meet different server requirements, ranging from single GPUs to multi-GPU setups.

Despite challenges in the competitive landscape, including new entrants such as China's DeepSeek, Nvidia remains optimistic about its ability to lead in performance and efficiency. The new Blackwell Ultra chips are built to meet the growing demands of reasoning-based AI models, solidifying Nvidia's position in the market.

Nvidia is set to continue innovating, as hinted by plans for future chip architectures, including a forthcoming design named after physicist Richard Feynman, anticipated to launch in 2028.

Nvidia, AI, Chips