Nvidia rolls out new AI models to cut costs and boost performance

Chip giant Nvidia introduced a fresh set of artificial intelligence tools on Monday that promise better performance at lower costs, jumping into a market where Chinese tech companies have been making big moves.

The California-based company rolled out its third version of AI models called Nemotron. They’re designed to handle writing tasks, computer programming, and other jobs. The smallest version, Nemotron 3 Nano, came out Monday. Two bigger versions will arrive sometime in the first six months of 2026.

Nvidia has built its reputation selling computer chips to companies like OpenAI, which then use those chips to build their own AI systems. But the company also makes its own AI tools.

These are available as open-source software, anyone can use them without paying. Research groups and businesses take advantage of these free offerings. Companies such as Palantir Technologies incorporate Nvidia’s technology into their products.

The new Nemotron 3 Nano model runs more efficiently than older versions. That means cheaper operating costs for users. It also handles complex, multi-step projects better than what came before.

Chinese open-source models gain ground

Nvidia’s announcement comes at a time when Chinese tech firms are gaining ground in AI. Companies including DeepSeek, Moonshot AI, and Alibaba Group Holdings have released their own open-source models that are catching on across the industry. Even major American companies are using them. Airbnb recently revealed it uses Alibaba’s Qwen model.

Meta Platforms might be moving away from open-source models toward closed systems. This would leave Nvidia as one of the few major American providers still offering open-source options.

Meta’s approach to AI has changed a lot over the past year. Last year, CEO Mark Zuckerberg was really confident about the company’s Llama AI models. He predicted they would become industry leaders and bring AI benefits to everyone. He spent a good chunk of time discussing Llama during the company’s January earnings call. By October, though, he barely mentioned the brand name.

Meta is now developing a new AI system with the internal code name Avocado, according to CNBC. People with knowledge of the project said many inside the company expect it to launch before the end of this year.

But someone familiar with the plans said the release is now scheduled for the first quarter of 2026. “The model is going through various performance tests right now to make sure it works properly when it debuts,” they added.

A Meta spokesperson said the company’s model training is proceeding as planned without any major schedule changes.

Security concerns drive U.S. bans on Chinese models

The growing use of Chinese AI models has raised concerns among American officials. Many state governments and federal agencies have prohibited the use of Chinese models because of security worries.

In reaction, China’s market regulator announced that a preliminary investigation found Nvidia violated the country’s anti-monopoly laws, as reported by Cryptopolitan. Beijing said it would continue looking into the matter. The probe relates to Nvidia’s purchase of Mellanox, an Israeli company that makes networking equipment for data centers and servers. Nvidia bought Mellanox in 2020. China approved the deal at the time with certain requirements attached.

Kari Briski leads generative AI software for business customers at Nvidia. She explained the company wants to offer a reliable model. The company is releasing its training information and other tools publicly so government agencies and businesses can check them for security issues and adjust them to meet their needs.

“This is why we’re treating it like a library,” Briski said. “This is why we’re committed to it from a software engineering perspective.”

Claim your free seat in an exclusive crypto trading community - limited to 1,000 members.