NVIDIA is giving away its most powerful AI for free: a genius move or a hidden threat to competitors?
On June 4, 2026, NVIDIA took a seemingly paradoxical step—it released its largest language model, Nemotron 3 Ultra, as open source. For free. With open weights, training data, and training methodologies. But, as practice shows, in the world of technology, "there's no such thing as a free lunch," and this story is a brilliant confirmation of that rule.
An Architecture That Targets Efficiency, Not Brute Force
Nemotron 3 Ultra is not just another "giant transformer." It is a hybrid architecture combining three different approaches: Mamba-2 layers for fast and economical processing of long texts, classic Attention layers for precise context retention, and an innovative Latent MoE (Mixture of Experts) mechanism. The latter compresses data before passing it to the "experts," forcing each to work as accurately and narrowly as possible, without unnecessary computational costs.
The result is impressive: with a total volume of about 550 billion parameters, only 55 billion are spent on processing each token. This allows the model, with a context window of 1 million tokens and a speed exceeding 300 tokens per second, to provide 5-6 times greater throughput and approximately 30% lower task execution costs compared to its counterparts.
The "Ecosystem" Strategy: Give Away the Model, Sell the Shovels
NVIDIA's true genius here lies not in the model itself, but in the strategy. The company, with a market capitalization exceeding $5 trillion, can afford to spend hundreds of millions on training a model and give it away for free. Why? Because whoever runs Nemotron is almost certainly doing so on NVIDIA graphics cards, using its software tools for fine-tuning, and deploying it on its own software stack.
This is not charity, but a brilliant marketing move. Graphics card sales more than offset the research costs. Openness here is a way to bring developers back to purchasing hardware, not altruism. Moreover, the model is attractive to countries building independent national AI—from Europe to Southeast Asia. It cannot be remotely disabled, which is especially valuable given the restrictions surrounding closed models.
Not the Smartest, but the Most Pragmatic
In the independent Artificial Analysis Intelligence Index ranking, Nemotron 3 Ultra scored 48 points. This is the best result among open US models, but it lags behind leaders like Kimi K2.6 (54 points) and DeepSeek. However, this gap of just 3-7 months is not critical for real-world tasks.
A bank deploying this model to process loan applications on its own servers does not 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 third parties.
My expert opinion: NVIDIA's bet on efficiency, rather than record-breaking test scores, is a far-sighted move. With mass AI adoption, the cost of running a model comes to the forefront. Nemotron 3 Ultra, almost matching in intelligence but five times cheaper, wins in real-world operation. The ecosystem will only strengthen: NVIDIA has the resources, motivation, and distribution channels to release increasingly powerful open models faster than any other company.