What NVIDIA’s GTC tells us about the future of enterprise AI

NVIDIA’s GPU Technology Conference (GTC) is one of the most anticipated tech events of the year. And this year’s “Super Bowl of AI,” as it’s frequently dubbed in media, was no exception: during a week that kicked off with NVIDIA CEO Jensen Huang’s two-hour keynote, the chip giant revealed various key innovations to monitor in the coming months.

From open-source foundation models to partnerships with Microsoft, GM and IBM, all the evidence points to a common theme: NVIDIA is no longer just a chipmaker. “From being a hardware company, it now [also] wants to become a gen AI software company,” says IBM Distinguished Engineer Kunal Sawarkar in an interview with IBM Think. “NVIDIA is getting into areas where players were not traditionally seeing it as a competition.”

According the Huang, the AI industry is at a USD 1 trillion computing inflection point with the rise of reasoning and agentic AI driving AI computing demand.

Here are some developments that IBM experts say they will be keeping an eye on in the coming months.

Two robots—the one a cute droid, the other a slightly creepy, life-size humanoid—stole the show at Huang’s keynote. But the real story was about the innovations that power these robots, according to IBM Principal Research Scientist Kaoutar El Maghraoui and IBM Master Inventor Nathalie Baracaldo on a recent episode of Mixture of Experts.

Chief among these innovations: GR00T N1, the first open-source foundation model designed specifically for robotics, and the Isaac GR00T Blueprint, a framework that makes it dramatically easier to generate synthetic motion data for training robots before unleashing them in the wild.

While hype around robotics is nothing new, open-sourcing these new technologies has the potential to "democratize [robotics] AI," El Maghraoui says—dramatically accelerating the development of humanoid robots.

Robotics innovation had long been stymied by a lack of data needed for simulations. But the Isaac GR00T Blueprint "allows us to basically create a huge environment where we can test [these robots] and make the whole development cycle much faster and much safer,” Baracaldo says.

Enter NVIDIA’s third new AI tool for creating real-time simulations of robots: the Newton physics engine, developed in partnership with Google DeepMind and Disney Research. By “lining up the right collaborators, [NVIDIA] is making it very hard for others to catch up,” says El Maghraoui.

Finally, factor in the world's smallest AI computer—NVIDIA's new DGX Spark, which the company also debuted at GTC—and now data scientists and hobbyists alike can use GR00T N1 and the Isaac GR00T Blueprint "to fine-tune and deploy these robotics" from home, says El Maghraoui.

The chip giant is also launching a new family of reasoning models designed to provide a strong foundation for agentic AI in business. Built on Llama models, the Llama Nemotron models come with reasoning capabilities that enable the creation of advanced AI agents tailored for enterprise needs. The models are available in different sizes and can be optimized for various business applications.

“NVIDIA’s open reasoning models, software and tools give developers and enterprises everywhere the building blocks to create an accelerated agentic AI workforce,” said Huang.

“These reasoning models, along with the infrastructure stack, will generate interest in the coming weeks or months because agentic AI is where everybody will be focused in 2025,” Sawarkar tells Think.

With DeepSeek making waves earlier this year by promising cheap, open-source reasoning models, NVIDIA is doubling down on its own. Its models are topping standard reasoning and agentic benchmarks, according to the company. And to further strengthen its place on the leaderboard, NVIDIA is already working with major tech players—Microsoft, Accenture, Box and SAP—to integrate these models into real-world applications.

AI storage is not flashy, but it plays an increasingly critical role in accelerating data processing and AI models. IBM Storage Scale is software-driven storage that enables organizations to better manage unstructured data across distributed storage environments such as clouds, data centers and out to the edge.

“So much enterprise data is unstructured—locked inside emails, PDFs, presentations, audio and video files, and other opaque formats—[and] is hard to interpret,” says Vincent Hsu, VP and CTO for IBM Storage.

"Managing exponential data growth has been a consistent challenge and just 1% of enterprise data is found in LLMs today,” says Ric Lewis, IBM SVP of Infrastructure, in an interview with Think. “Content-aware storage helps bring more enterprise data to existing AI models in a way that is still controlled and protected from making copies of it. This can help make models more meaningful to an organization and unlock new data insights that can set them apart. These new capabilities are based on work done at both IBM Research and NVIDIA."

Hsu sees many use cases for CAS, such as helping ensure that AI assistants have the latest contextual information, which can help improve accuracy and address hallucinations. Ultimately, with IBM’s content-aware Storage Scale combined with NVIDIA's new AI Data Platform for enterprise AI infrastructure, companies are able to extract more value from their unstructured enterprise data, one example of several NVIDIA-IBM partnerships announced in a busy week.