[DigitalToday reporter Chi-gyu Hwang] Nvidia is moving faster toward developing its own open-source AI models. It is developing and distributing its own open-source AI models as part of an effort to expand hardware demand, but some see it as creating a delicate situation in which Nvidia could end up competing with AI companies that buy its hardware.

The Information recently reported that Nvidia is developing the open-source large language model Nemotron 4 with at least 1 trillion parameters.

ㆍNvidia gains momentum in open-source AI model development..."Developing Nemotron 4 with 1 trillion parameters"

With parameters about twice the level of Nemotron 3 Ultra, which Nvidia unveiled in June, it may not be the world’s largest open-source model. But it is seen as likely to rank among the top tier. Chinese AI startup Moonshot AI recently unveiled its large language model Kimi K3 with 2.8 trillion parameters, while DeepSeek V4-Pro has 1.6 trillion.

Nemotron 3 Ultra ranks second among U.S. open models, behind Thinking Machines Lab’s Inkling, on performance benchmarks including Arena and Artificial Analysis, which evaluate agent performance and intelligence. It has a wide gap with top Chinese open models and sits outside the top 40 overall.

If Nemotron 4 is released, Nvidia’s ranking in the open-source AI model field is expected to rise sharply.

As it develops the Nemotron 4 series, Nvidia has released Nemotron 3.5 Lightning, a lightweight open-source model. It has also rolled out free model routing software to help companies more easily develop model routers that connect suitable models for specific AI tasks.

Nvidia has also significantly increased spending on computing resources needed to train its own models.

Nvidia secures computing resources by leasing back AI servers from some cloud providers that buy its chips. As of April, Nvidia’s multiyear cloud service contracts running through early 2031 had grown to $28 billion, about three times the level a year earlier, The Information reported.

Even if it has no intention of directly competing with AI companies, some say Nvidia’s increasingly aggressive push into open-source AI models could complicate relationships with AI companies that develop models on Nvidia GPUs, whether open-source or closed.

More companies are adopting Nvidia AI models. Palantir said in June it would work with Nvidia to use Nemotron models for U.S. government customers.

These companies have long developed AI models using Nvidia GPUs. Against that backdrop, Nvidia invested $30 billion in OpenAI, a major AI model developer. Nvidia also made substantial investments in U.S. open-source AI startups including Reflection AI and Thinking Machines Lab.

Nvidia says diversity in open models tied to its own open-source AI model development could ultimately lead to higher demand for its chips. It is making clear it has no intention of competing with companies that make a living developing AI models.

From Nvidia’s perspective, it benefits if more companies, from startups to large corporations, use high-quality, low-cost models optimized for its hardware to develop and use AI. The same is true when Nvidia’s open-source strategy encourages development of other open-source models.

Nvidia believes its in-house model development will spur competition and lead to more open model development, ultimately boosting GPU demand, The Information reported, citing a Nvidia executive.

Kari Briski (카리 브리스키), vice president of generative AI at Nvidia, said, "We are investing in Nemotron because Nvidia believes every company and country needs accessible state-of-the-art open models to enhance safety and security, accelerate innovation, and establish a foundation they can trust and use even as generations change."

Nvidia’s own open-source AI models are not only about boosting GPU demand. They are also an important factor in strengthening Nvidia’s hardware design capabilities.

ㆍ[Tech Insight] Why does Nvidia keep developing LLMs?

Nvidia has long developed its own AI models to obtain information needed for hardware design.

Briski previously stressed in an interview with Alex Kantrowitz (알렉스 칸트로위츠), who runs the tech podcast and newsletter "Big Technology", that "presence in the LLM field is the foundation of core technology development capabilities." She said that understanding core technology properly is necessary to communicate more clearly with partners.

She said, "We need to understand how to train these models at scale and how to run them at scale through inference." She said, "This can inform not only GPU architecture but also storage and networking."

Open-source AI has emerged as a heavyweight geopolitical variable as China expands its global market share with open-source AI, and Nvidia CEO Jensen Huang has actively supported open source. In an email in July, he stressed that "open models advance safety, cybersecurity, scientific progress and national security."