Nvidia Invests $26 Billion in Open-Weight AI Model Development, Filings Reveal

Nvidia plans to invest $26 billion over the next five years to advance open-source artificial intelligence model development, as indicated in a recent financial filing for 2025. Executives shared insights about this significant venture during interviews with WIRED, marking a pivotal move for the company.

Nvidia’s Strategic Investment in AI

This investment aims to transform Nvidia from a leading chip manufacturer into a serious player in AI model development, potentially rivaling established entities like OpenAI and DeepSeek. The new open-source models will be specifically optimized for Nvidia’s hardware, reinforcing its position in the AI market.

Understanding Open-Source Models

• Open-source models allow public access to weights or parameters, enabling widespread use.
• These models can be modified and executed on personal machines or cloud platforms.
• Nvidia is also sharing technical advancements that simplify modifications for startups and researchers.

Nemotron 3 Super: Nvidia’s Latest AI Model

On Wednesday, Nvidia unveiled the Nemotron 3 Super, its most advanced open-weight AI model to date. With 128 billion parameters, it equates closely with the largest version of OpenAI’s GPT-OSS but claims superior performance across various benchmarks.

• Nemotron 3 Super scored 37 on the Artificial Intelligence Index, surpassing GPT-OSS, which scored 33.
• Chinese AI models have shown higher performance in certain benchmarks.

Benchmark Performance

Nvidia’s model excelled in a secret test on PinchBench, focused on controlling OpenClaw, where it secured the top rank. This achievement reflects Nvidia’s commitment to pushing the boundaries in AI model capabilities.

Innovative Training Techniques

Nvidia has introduced several innovative training techniques for Nemotron 3, enhancing its reasoning abilities and responsiveness. Key improvements include:

• Architectural enhancements for better long-context handling.
• Techniques to boost responsiveness to reinforcement learning.

Competitive Landscape in AI

Since the launch of Llama by Open Frontier Meta in 2023, the open-model space has experienced significant changes. However, OpenAI’s open-weight models remain less adaptable compared to proprietary models.

Many Chinese AI companies, including DeepSeek and Alibaba, provide openly accessible weights for their models, attracting startups and researchers worldwide. This accessibility fosters an ecosystem where innovation thrives.

The Future of Nvidia’s AI Models

With the release of the first Nemotron model in November 2023, Nvidia is not slowing down. The company recently completed pretraining a massive 550-billion-parameter model, reinforcing its dedication to advanced AI technologies.

Future AI developments aim not only to enhance chip performance but also to optimize supercomputer-scale datacenters. Kari Briski, VP of generative AI software for enterprise, emphasizes the importance of these efforts in advancing Nvidia’s hardware architecture roadmap.

This substantial investment solidifies Nvidia’s vision for a future deeply integrated with AI, shaping the landscape for open-source model development.