Crypto news

Something unusual is happening in the world of artificial intelligence. NVIDIA, a company primarily known as the manufacturer of the most expensive and sought-after hardware for AI, has released its largest open model — Nemotron 3 Ultra. The release took place on June 4, 2026, and under a free license, not only the model weights were published, but also the training data and training methodologies. At first glance, it's a generous gesture. But, like everything in Jensen Huang's strategy, there is a cold market calculation behind it.

An Architecture That Breaks Stereotypes

Nemotron 3 Ultra is not just another "scaled-up transformer." It is based on a hybrid architecture combining three approaches: Mamba-2 layers, classic attention layers, and latent mixture of experts (Latent MoE). Mamba-2 layers process long texts quickly and efficiently: their costs grow linearly, not exponentially like the standard attention mechanism. Attention layers, in turn, accurately retain large volumes of context in memory. And Latent MoE compresses data before passing it to experts, forcing each of them to work narrowly and precisely, without unnecessary computational costs.

The result: with a total volume of about 550 billion parameters, only approximately 55 billion are used to process each token. The model thinks like a giant system but behaves in terms of cost like a much more compact one. A context window of 1 million tokens and a speed of over 300 tokens per second provide five to six times greater throughput and approximately 30% lower task costs compared to analogs.

Strategy: Not a Model, but an Ecosystem

The main value of this release is not the model itself, but the ecosystem that NVIDIA is building around its hardware. The logic is simple: whoever runs Nemotron almost certainly does so on NVIDIA graphics cards, fine-tunes it using its software tools, and deploys it on its own software. Openness here is not charity, but a way to bring developers back to purchasing the company's hardware.

And it works. With a market capitalization of over $5 trillion, the costs of training Nemotron 3 Ultra, which are likely in the hundreds of millions of dollars, are almost negligible for the company. Graphics card sales more than cover the research, so NVIDIA can give away the model for free and still earn more than closed competitors with paid access.

The political context adds additional weight to the release. An open American model can be inspected, modified, and run on one's own servers — this has made it attractive for countries building independent national AI, from Europe to Southeast Asia. No one can remotely disable such a model, and this is especially valuable against the backdrop of recent restrictions surrounding closed models.

Weaknesses and a Look to the Future

Despite all its advantages, Nemotron 3 Ultra is not the smartest model on the market. In the independent Artificial Analysis Intelligence Index rating, it scored 48 points — the best result among open US models, but globally it lags behind leaders such as Kimi K2.6 (54 points) and DeepSeek. According to analysts' estimates, open models trail closed ones by three to seven months.

But this lag matters less and less if the 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 closed data, kept within its own secure perimeter, and not share confidential information with outsiders.

My conclusion: NVIDIA's bet on efficiency rather than test records may prove more far-sighted than it seems. 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.