On March 11 (US time), Nvidia announced a US$26 billion five-year investment in open-source large language models and introduced Nemotron 3 Super, its most powerful hybrid mixture-of-experts model to date. The company said the model outperforms OpenAI's open-source GPT-OSS in several benchmark tests.
The launch forms part of a broader open-source strategy. The US$26 billion investment will fund model development while stress-testing AI compute, memory, and networking infrastructure.
Operational data from these deployments will feed directly into Nvidia's future chip and system architecture, enabling large-scale model training to inform hardware design.
The strategy reinforces Nvidia's dominance in AI training infrastructure. Its GPUs remain the industry standard for large-scale model training, and open-source models optimized for Nvidia hardware could deepen its grip on the global AI compute market.
More broadly, the move signals Nvidia's transition from a supplier of chips and software stacks to a full-stack AI company competing with leading AI labs.
Rather than confronting rivals such as OpenAI, Anthropic, or DeepSeek directly, Nvidia treats open-source models as a mechanism to expand the AI ecosystem — reinforcing a strategy driven by hardware and models in tandem.
Open-source models as an AI ecosystem catalyst
Nvidia CEO Jensen Huang recently underscored the importance of open-source models in the global AI ecosystem.
Huang noted that a large share of global AI models are open-source, with enterprises, research institutions, and governments relying on them for AI development. Once these models reach advanced capability levels, they stimulate demand across the wider technology supply chain.
Huang cited DeepSeek R1 as a telling example, saying the model helped trigger a global shift toward open-source AI. Its release accelerated application development while stoking demand for AI training compute, infrastructure, chips, and energy.
Such a single model breakthrough can therefore ripple across the entire AI industry chain.
Open-source strategy builds Nvidia's developer ecosystem
For Nvidia, the open-source strategy also carries significant commercial implications.
When releasing models, Nvidia publishes model weights and technical documentation, enabling start-ups and researchers to build on the technology. This approach cultivates a developer community anchored to Nvidia hardware, reinforcing long-term demand for its chips.
By actively shaping the open-source ecosystem, Nvidia steers developers worldwide to build innovation on top of its technology platform.
AI models guide Nvidia's hardware roadmap
The initiative is also closely tied to Nvidia's hardware roadmap, extending beyond model competition.
According to Kari Briski, Nvidia vice president of generative AI software for enterprise, the company's AI models will support chip development while optimizing the architecture of supercomputing-scale data centers.
Nvidia vice president of generative AI software for enterprise, Kari Briski. Credit: GeekWire YouTube
"We build it to stretch our systems and test not just the compute but also the storage and networking, and to kind of build out our hardware architecture roadmap," Briski said.
Nvidia also outlined the "Super + Nano" layered architecture used in the Nemotron 3 series.
Nemotron 3 Nano, introduced in December 2025, handles targeted single-step tasks in AI agent workflows, while Nemotron 3 Super focuses on complex multi-step reasoning and planning.
In software development, Nvidia suggests Nano could process simple merge requests, while complex coding tasks requiring deep understanding of a codebase would be handled by Super. Highly specialized tasks could then call on third-party proprietary models.
The layered design aims to help enterprises balance cost and capability while deploying AI agents at scale.
Article translated by Levi Li and edited by Jerry Chen
