NVIDIA gives away powerful AI for free — and earns more than its competitors from it
On June 4, 2026, NVIDIA released its largest language model, Nemotron 3 Ultra, to the public. Unlike closed flagship models like ChatGPT or Claude, this model is distributed under a free license: weights, training data, and training methodologies are available to everyone. It is designed for long-lived autonomous agents and complex reasoning.
NVIDIA's strategy here is fundamentally different. Instead of selling access to AI as a service, the company gives away the "gold" for free but makes money by selling the "shovels." The launch of Nemotron 3 Ultra is virtually guaranteed to run on NVIDIA graphics cards, fine-tuning uses its tools, and deployment relies on its software. Openness here is not charity, but a mechanism to lock developers into the ecosystem.
What makes Nemotron 3 Ultra unique?
The model's architecture is hybrid. It combines three approaches: Mamba-2 layers for fast and cost-effective processing of long texts (costs grow linearly, not exponentially), classic attention layers for precise retention of large context volumes, and a Latent MoE (Latent Mixture of Experts) mechanism. The latter compresses data before passing it to "experts," forcing each to work narrowly and precisely, without unnecessary computation.
The model has approximately 550 billion parameters in total, but only about 55 billion are utilized for processing each token. This allows it to think like a massive system while behaving like a much more compact one in terms of cost. With a context window of 1 million tokens and a speed exceeding 300 tokens per second, this provides five to six times greater throughput and roughly 30% lower task costs compared to analogs.
Ecosystem strategy
NVIDIA's financial capabilities are incomparable to the model's costs. With a market capitalization exceeding $5 trillion, training Nemotron 3 Ultra, which likely cost hundreds of millions of dollars, is a nearly negligible expense. Graphics card sales more than cover the research, allowing the company to give away the model for free and still earn more than its closed competitors.
The political context also plays into NVIDIA's hands. An open American model can be inspected, modified, and run on one's own servers—this makes it attractive for countries building independent national AI, from Europe to Southeast Asia. No one can remotely disable such a model, which is especially valuable amid recent restrictions surrounding closed models.
Where does the model fall short?
Despite all its advantages, Nemotron 3 Ultra is not the smartest model on the market. In the independent Artificial Analysis Intelligence Index ranking, it scored 48 points—the best result among open US models, but trailing leaders like Kimi K2.6 (54 points) and DeepSeek. Open models generally lag behind closed ones by three to seven months.
However, this gap matters less and less if an open model is simply sufficient for real-world tasks. A bank deploying Nemotron 3 Ultra to process loans on its own servers doesn't need flagship-level intelligence—it needs a model that can be fine-tuned on private data, kept within a secure perimeter, and not share confidential information with outsiders.
Analytical conclusion: NVIDIA's bet on efficiency, rather than test records, may prove more farsighted. With mass AI adoption, the cost of running a model comes to the forefront—and one that is almost as smart but five times cheaper wins in real-world operation. The open ecosystem will only strengthen: NVIDIA has the resources, motivation, and distribution channels to release increasingly powerful open models faster than any other company.