Nvidia on Tuesday revealed more details about its next artificial intelligence chip platform, Blackwell Ultra, which it says will help applications reason and act on behalf of users.
Blackwell Ultra, which the company announced at its annual GTC conference, builds on Nvidia’s existing sought-after Blackwell chip. The company said the additional computing power in the new Ultra version will make it easier for AI models to break down complex queries into multiple steps and evaluate different options, or in other words, reason.
Demand for AI chips has surged following OpenAI’s ChatGPT in 2022, sending Nvidia’s stock price soaring. Its chips power the data centers that power popular, power-hungry AI and cloud-enabled services at companies like Microsoft, Amazon and Google.
But the reasoning ability and supposedly low cost of DeepSeek, a Chinese tech startup that shocked Wall Street with its R1 model, has led to speculation that expensive hardware may not be necessary to run high-performance AI models. But Nvidia appears to be overcoming such concerns, as evidenced by January quarter earnings that beat Wall Street expectations.
Nvidia wants its chips to be at the core of the reasoning models that the Chinese tech startup helped popularize. It claims that a DeepSeek R1 query that took a minute and a half to answer on Nvidia’s previous-generation Hopper chip will take just 10 seconds with the Blackwell Ultra.
Cisco, Dell, HP, Lenovo and Supermicro are among the companies working on new servers based on Blackwell Ultra. The first products with Blackwell Ultra are expected to ship in the second half of 2025.
Experts say being able to reason or think about a response before responding will allow AI applications and agents to handle more complex and specific types of questions. Instead of simply providing an answer, a reasoning chatbot could examine a question and provide multiple, specific responses that account for different scenarios.
Nvidia gave the example of using a reasoning model to help create a seating arrangement for a wedding that takes into account preferences, such as where to seat the parents and in-laws and seating the bride on the left side.
“Models are now starting to mimic somewhat human-like behavior,” said Arun Chandrasekaran, an artificial intelligence analyst at market research firm Gartner.
DeepSeek and OpenAI aren’t the only ones creating models that can reason. Google also updated its Gemini models with more reasoning capabilities late last year, and Anthropic introduced a hybrid reasoning model called Claude 3.7 Sonnet in February.
