NVIDIA Corporation (NVDA.US) is ramping up its efforts in open-source AI models! It has launched Nemotron 3.5 Lightning and is currently developing the trillion-parameter Nemotron 4.

NVIDIA Corporation (NVDA.US) is further expanding its landscape of open-source artificial intelligence models. The company has recently launched Nemotron 3.5 Lightning and simultaneously released the open-source smart routing tool NeMo Switchyard to improve the efficiency of long-running AI agent applications. Meanwhile, reports indicate that NVIDIA Corporation is developing the next-generation Nemotron 4 series, with the largest model's parameter scale expected to reach at least one trillion, aiming to rank among the world's leading open-source AI models in terms of performance. NVIDIA Corporation's accelerated push for the Nemotron series reflects its strategic shift from AI chips and infrastructure to model layers. The underlying logic is to drive more enterprises to adopt open-source models, lower the thresholds for AI deployment, and further expand the overall demand for GPU computing power within the AI ecosystem. The launch of Nemotron 3.5 Lightning focuses on AI agent efficiency. NVIDIA Corporation has announced an expansion of its Nemotron 3 open-source model series with the introduction of Nemotron 3.5 Lightning. The company describes it as the most efficient model for long-running AI agent workloads in its category. Nemotron 3.5 Lightning is a hybrid expert model with 30 billion parameters, primarily designed to help developers build smarter and more efficient AI agent applications. Unlike traditional chat applications that focus primarily on single-question-and-answer scenarios, AI agents typically need to perform multi-step reasoning, call on different tools, and continuously complete complex tasks. Therefore, model operational efficiency, inference costs, and the scheduling capabilities between different models become increasingly important. NVIDIA Corporation aims to penetrate this rapidly growing application landscape further with Nemotron 3.5 Lightning. Meanwhile, the company also launched the open-source library NeMo Switchyard, which provides intelligent model routing capabilities for mainstream AI agent tools. Businesses can establish custom routing systems based on their specific needs. Once deployed, NeMo Switchyard can automatically direct requests to AI models that are better matched in terms of capability and characteristics for different tasks, without requiring developers to rewrite existing applications. NVIDIA Corporation stated that the combination of Nemotron 3.5 Lightning and NeMo Switchyard enables businesses to control how AI is deployed, where it runs, and how efficiently it operates, covering various computing environments including PCs, workstations, data centers, and the cloud. Nemotron 4 is in development, with the largest model parameters expected to reach at least one trillion. In addition to Nemotron 3.5 Lightning, NVIDIA Corporation is also advancing the next-generation Nemotron 4 with more aggressive scale and performance. According to reports from The Information, citing multiple individuals involved in the Nemotron project, NVIDIA Corporation hopes that the largest model in the Nemotron 4 series will compete in performance with the world's top open-source AI models. This project is massive. The research paper for NVIDIA Corporation's previous large model involved 570 authors, and employees have indicated that even more personnel might be involved with Nemotron 4. Several employees engaged in the Nemotron project expect that the largest Nemotron 4 model will contain at least one trillion parameters, approximately double the parameter scale of the Nemotron 3 Ultra launched by NVIDIA Corporation in June of this year. As of now, the final specifications and release date for Nemotron 4 have yet to be determined. Reports indicate that NVIDIA Corporation has made some decisions regarding the pre-training data and model architecture for the large Nemotron 4 model, but final training has not yet begun, and this process could take several months. Two employees anticipate that the model may be ready as early as later this fall, though others believe the release may be later. Joining global AI laboratories to promote open-source frontier models In March of this year, NVIDIA Corporation established the Nemotron Coalition to collaborate with several leading AI laboratories worldwide on developing open-source frontier models. At the time, the company stated that the first model created collectively by the coalition would serve as the foundation for the future Nemotron 4 open-source model series. Kari Briski, NVIDIA Corporation's Vice President of Generative AI, stated that the company invests in Nemotron because it believes every business and country should have access to cutting-edge open-source models to enhance security, accelerate innovation, and provide a foundation that can continue to be used across different technological generations. This strategy also signifies that NVIDIA Corporation is no longer satisfied with merely providing GPUs to AI developers but is beginning to take a more active role in building an open-source model ecosystem. Doubling down on open-source models: Ultimately still pointing to GPU demand The substantial investment by NVIDIA Corporation in open-source AI model development is driven by a more direct business logic: the richer the open-source model landscape and the lower the barriers to use, the broader the potential for enterprises to deploy AI, further stimulating GPU demand. Currently, the high demand for AI chips from NVIDIA Corporation comes from frontier AI laboratories and major cloud computing operators like OpenAI, Microsoft Corporation (MSFT.US), and SpaceX (SPCX.US). However, some of these large clients are actively developing their own AI chips to reduce dependence on NVIDIA Corporation's GPUs. Therefore, further expanding the sources of demand for AI computing power is crucial for NVIDIA Corporation to maintain long-term growth in the GPU market. Reports indicate that NVIDIA Corporation believes its model development work can drive competition in the AI model market, encouraging the emergence of more open-source models, and ultimately creating greater demand for GPUs. In other words, the significance of Nemotron lies not only in whether NVIDIA Corporation can build the world's most powerful AI model, but whether the company can further expand the open-source AI ecosystem, enabling more businesses and developers to deploy AI, and converting these new workloads into demand for computational infrastructure. NVIDIA Corporation is also facing a nuanced situation of competing with its clients and invested companies. However, the Nemotron strategy puts NVIDIA Corporation in a more complex position. On one hand, the company is potentially competing with some open-source AI startups in which it has invested; on the other hand, NVIDIA Corporation is also beginning to experience some degree of business overlap with its most important GPU clients like OpenAI and Anthropic at the model level. OpenAI has long been one of the key drivers of demand for NVIDIA Corporation's AI chips, and the company itself has invested $30 billion in OpenAI. Reports suggest that the flagship model of Nemotron 4 is expected to still not reach the overall capabilities of the most advanced closed-source frontier models from OpenAI and Anthropic, but open-source models are increasingly seen by businesses as a cost-effective and more controllable alternative for certain workloads. Thus, NVIDIA Corporation does not necessarily need Nemotron to beat all closed-source models outright. As long as it can promote the widespread adoption of open-source models and expand AI deployment, it could create positive feedback for its core GPU business. Nemotron is still lagging behind the industry's cutting edge. Despite ongoing investment, Nemotron's actual performance currently falls short of the world's leading models. Reports indicate that NVIDIA Corporation's largest model, Nemotron 3 Ultra, ranks second among open-source models in the U.S. in benchmark testing for specific AI capabilities and intelligence levels, only behind the recently released Inkling from Thinking Machines. However, on a global scale, Nemotron 3 Ultra has not even entered the top 40 overall model rankings. This also means that Nemotron 4 will be a critical generation of products to test NVIDIA Corporation's model strategy. If the next-generation model can significantly narrow the performance gap with the leading global open-source models, NVIDIA Corporation will not only possess GPUs, networking, server systems, and AI software platforms, but will also further grasp this key link of open-source models, forming a more complete AI technology stack. From a business strategy perspective, the ultimate goal of NVIDIA Corporation's push for Nemotron remains closely related to its core hardware business: the more prevalent open-source AI models are, the richer the scenarios in which enterprises deploy AI, and the greater the global demand for AI computing resources and GPUs is likely to expand.