Nvidia is getting into the artificial intelligence agents game with the release of a software platform that helps businesses build their own autonomous bots.
The platform, called NeMo microservices, is available for all customers to use to build their “AI teammates,” the chip maker said Wednesday.
The Santa Clara, Calif.-based company said it has developed a better way of building AI agents that relies on open-source AI models like those provided by Meta Platforms and the startup Mistral AI.
Nvidia is betting on open-source, or open-weight, AI technologies because they tend to offer businesses more flexibility and control than proprietary models. Popular proprietary models are offered by vendors like OpenAI and Anthropic. AI models that are open-weight share the numerical parameters, or “weights,” that underlie them.
That flexibility is critical for enterprises with a lot of confidential data but still want to use agents, said Joey Conway, a senior director of generative AI software for enterprise at Nvidia.
“We wanted to focus on places where enterprises need the full control of open-weight models,” Conway said. “And we don’t see much of the market moving there.”
Nvidia joins the ranks of companies like OpenAI, Microsoft, Amazon and Google that aim to help businesses build their own AI agents, which are technologies that can independently perform tasks across various functions.
Nvidia pegs the size of the AI agent market at $1 trillion—roughly the same as the enterprise software market that it predicts agents will replace. But over the past year, the technology has struggled to gain widespread adoption among enterprises.
It has been difficult for developers to train models while readying their corporate data, Conway said. That’s one gap Nvidia’s NeMo microservices aims to fill, he said, by making agent-building easier with a system for incorporating private business data.
By selling software for agents, Nvidia continues a tactic it started with its popular Compute Unified Device Architecture, or CUDA, a programming language that lets developers write applications for graphics processing units.
That strategy of selling software alongside hardware keeps Nvidia’s GPU business strong, said Dave McCarthy, a research vice president at research firm International Data Corp.
Another reason businesses might want to use Nvidia’s software to build agents is because it isn’t tied to any particular cloud platform, McCarthy added.
“In many cases, it’s an easy choice for an enterprise to say, ‘If you don’t want to lock yourself into a particular cloud provider or hardware company, Nvidia is a good choice,’” he said.
Write to Belle Lin at belle.lin@wsj.com
