NVIDIA is giving away its most powerful AI for free: a brilliant marketing move or a new leadership strategy?
On June 4, 2026, NVIDIA took what initially seemed like a paradoxical step by releasing its largest language model, Nemotron 3 Ultra, as open source. Unlike closed giants such as ChatGPT or Claude, this model—with open weights, training data, and methodologies—is available for download, fine-tuning, and deployment on any developer's own infrastructure. It would appear the company is giving away a key asset for free. But, as practice shows, this is merely the tip of the iceberg of a complex and far-sighted strategy.
Nemotron 3 Ultra is not just "another large transformer." It is built on a unique hybrid architecture that combines three approaches: Mamba-2 layers, attention layers, and a latent mixture of experts (Latent MoE). Mamba-2 ensures linear cost growth when processing long texts (instead of the quadratic cost of standard attention), which is critical for long-running autonomous agents. Attention layers, in turn, guarantee precise context retention. Meanwhile, Latent MoE compresses data before passing it to "specialists" within the model, allowing each to work narrowly and efficiently without requiring extra computation.
The result is impressive: with a total volume of approximately 550 billion parameters, only about 55 billion are used to process each token. This means the model thinks like a giant but behaves like a much more compact system in terms of cost. A context window of 1 million tokens and a speed exceeding 300 tokens per second deliver 5-6 times higher throughput and roughly 30% lower task costs compared to alternatives.
The "Shovel Seller" Strategy: Openness as the Key to Dominance
The main value of this release, in my assessment, lies not in the model itself, but in the ecosystem NVIDIA is building around its hardware. The logic is simple and elegant: whoever runs Nemotron almost certainly does so on NVIDIA graphics cards, fine-tunes it using its software tools, and deploys it on its software. Openness here is not charity, but a powerful catalyst for hardware sales. With a company market capitalization exceeding $5 trillion, the cost of training Nemotron 3 Ultra (likely hundreds of millions of dollars) is a nearly negligible expense. Graphics card sales more than cover the research, so NVIDIA can give away the model for free and still earn more than closed competitors charging for access.
The political context adds further weight to this move. An open American model that can be inspected, modified, and run on one's own servers becomes highly attractive for countries building independent national AI—from Europe to Southeast Asia. It cannot be remotely disabled, and this is especially valuable in light of recent restrictions surrounding closed models.
Are There Weaknesses? Yes, But That's Not the Main Point
To be fair, 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 overall lag behind closed ones by three to seven months.
My expert perspective: This lag matters less and less if an open model is simply "good enough" for real-world tasks. A bank deploying Nemotron 3 Ultra for loan processing on its own servers doesn't need flagship-level intelligence—it needs a model that can be fine-tuned on proprietary data, kept within a secure perimeter, and not share confidential information with outsiders. NVIDIA's bet on efficiency, rather than test records, may prove far more prescient. In mass AI adoption, the cost of running a model takes center stage, and one that barely lags in intelligence but is five times cheaper wins in real-world operation. NVIDIA has the resources, motivation, and distribution channels to release increasingly powerful open models faster than any other company. The market hasn't fully realized this yet, but I predict this strategy will cement NVIDIA's dominance for decades to come.