Nvidia Unveils Blackwell Ultra AI Chip, Aiming to Advance AI Reasoning Capabilities
Nvidia has disclosed more details about its upcoming Blackwell Ultra artificial intelligence chip platform, designed to enhance AI’s capabilities in reasoning and decision-making. This advancement could transform AI applications beyond simple chatbots into practical tools that can act on a user’s behalf.
Developed upon the foundation of Nvidia’s existing Blackwell chip, the Ultra version promises to boost computational power, enabling AI models to break down intricate queries into multiple steps and evaluate various options effectively—essentially, allowing them to reason. This development has been spurred by the surging demand for AI chips, especially after the introduction of OpenAI’s ChatGPT in 2022, which significantly boosted Nvidia’s share prices.
Nvidia’s chips are critical for powering the data centers behind popular, resource-intensive AI and cloud services offered by tech giants like Microsoft, Amazon, and Google. The company’s January quarter earnings exceeded Wall Street’s predictions, indicating its continued dominance in the market.
Nvidia aims for its chips to be central to reasoning models, like those pioneered by Chinese tech startup DeepSeek. These models sparked market discussions, suggesting the possibility of high-performing AI models running on less expensive hardware. Nvidia states that a prompt handled by DeepSeek R1 that took a minute and a half on their previous Hopper chip would only take 10 seconds on the Blackwell Ultra.
Companies such as Cisco, Dell, HP, Lenovo, and Supermicro are actively developing new servers centered on Blackwell Ultra, with the first product releases expected in the latter half of 2025.
Experts believe that enhanced reasoning capabilities will enable AI apps and agents to manage more complex and specialized questions. Instead of simple answers, chatbots with reasoning capabilities could effectively dissect questions and offer multiple, specific responses based on different scenarios. Nvidia provided an example of using a reasoning model to plan a wedding seating arrangement, taking factors like family preferences and seating positions into consideration.
“The models are now starting to mimic a little bit of human-like behavior,” commented Arun Chandrasekaran, an AI analyst at Gartner, the market research firm.
Google has also updated its Gemini models to improve reasoning skills, and Anthropic launched a hybrid reasoning model named Claude 3.7 Sonnet in February, demonstrating a growing trend.
Furthermore, some experts anticipate reasoning models to lead to the development of “AI agents,” or AI assistants capable of performing actions rather than just supplying answers. Companies like Google, Amazon, and Qualcomm are advancing their concepts for AI-powered helpers that can manage tasks, such as arranging vacations according to personal preferences.
Gene Munster, a managing partner at Deepwater Asset Management, noted, “What agentic AI excels at is multitasks, and being able to reason in each of those tasks is going to make the agents more capable.”
