NVIDIA doubles down on becoming a major model maker. Plans to increase investments in open-source tech.
The market’s beloved chip designer, NVIDIA, just unveiled a family of open-source models called the Nemotron 3.
It has made fortunes supplying chips to the market giants. But now it’s vamping its roadmap. NVIDIA is trying to expand its offerings, especially given that some market leaders have now begun designing and manufacturing their own capable-enough chips. Be it Anthropic, Google, or OpenAI.
That’s crucial for NVIDIA. But it has already found a roundabout- the family of open-source models- Nano (30 billion parameters), Super (100 billion parameters), and Ultra (500 billion parameters).
Open-source AI models are extremely substantial to AI research and development. That’s what most companies experiment with, prototype, and build upon. Right now, Chinese counterparts enjoy the dominance. Because even though Google and OpenAI also offer smaller models, they aren’t updated and refined as regularly.
But with Nemotron 3, NVIDIA might become the best of the best.
According to the company’s press release ahead of the launch, NVIDIA published specific benchmark scores. These scores showcase that these models are very easily downloadable and modifiable. And they run on one’s own hardware.
“Open innovation is the foundation of AI progress,” asserts Jensen Huang.
And with the Nemotron 3, NVIDIA plans to transform advanced AI. And offer developers the toolkit to efficiently and seamlessly develop scalable agentic AI systems. That remains the roadmap for now. To empower engineers and developers with transparency and efficiency.
And to further differentiate itself from its US rivals, NVIDIA is being quite flexible and transparent with the data used to train Nemotron. Because it’s not just a glimpse into user privacy and ethical practices, but opens up a segueway for developers to modify the model easily. Something that NVIDIA’s competitors moved away from in the past year due to fear of their research being stolen.
Additionally, the company is also launching tools for fine-tuning and customization, along with a new hybrid latent mixture-of-experts model architecture and libraries.
The only hindrance for NVIDIA? Its silicon has become a bargaining chip. It’s substantial to the AI and global economy. And this could work against the company as we witness intensifying competition in this sector.
