NVIDIA gives away AI for free, but earns the most of all: the Nemotron 3 Ultra strategy
On June 4, 2026, NVIDIA released Nemotron 3 Ultra — the largest open-source AI model in the Nemotron 3 lineup. The weights, training data, and training methodologies are released under a permissive license. This is not charity, but a subtle market maneuver: the company gives away "free" intelligence to sell hardware.
Unlike closed giants like ChatGPT or Claude, Nemotron 3 Ultra can be downloaded, fine-tuned on your own data, and run on your own infrastructure. The bet here is not on maximum intelligence, but on openness, efficiency, and control. And it works.
Hybrid Architecture: Why It Matters
Nemotron 3 Ultra is not just a "scaled-up transformer." It is based on a hybrid architecture of three components: Mamba-2 layers, Attention layers, and a Latent Mixture of Experts (Latent MoE). Mamba-2 processes long texts quickly and efficiently — costs grow linearly, not exponentially. Attention layers accurately retain large context volumes. And Latent MoE compresses data before passing it to experts, forcing each to work narrowly and precisely.
The result: the model has around 550 billion parameters, but only about 55 billion are activated for processing each token. It thinks like a giant but behaves like a compact system in terms of cost. With a context window of 1 million tokens and a speed exceeding 300 tokens per second, this yields 5–6 times greater throughput and approximately 30% lower task execution costs.
NVIDIA's Strategy: Ecosystem, Not Model
The main value of the release is not the model itself, but the ecosystem NVIDIA is building around its hardware. Anyone running Nemotron is almost certainly doing so on NVIDIA GPUs, fine-tuning it with its software tools, and deploying it on its software stack. Openness here is not charity, but a way to funnel developers back into purchasing the company's hardware.
NVIDIA's financial capabilities are incomparable to the model's costs. With a market capitalization exceeding $5 trillion, training Nemotron 3 Ultra likely cost hundreds of millions of dollars — a nearly negligible expense for the company. GPU sales more than cover the research, so NVIDIA can give the model away for free and still earn more than closed competitors charging for access.
The political context adds weight: an open American model can be inspected, modified, and run on private servers — making it attractive for countries building independent national AI, from Europe to Southeast Asia. Such a model cannot be remotely shut down, and this is especially valuable given recent restrictions around closed models.
Weaknesses and Prospects
Despite its strengths, Nemotron 3 Ultra is not the smartest model on the market. In the independent Artificial Analysis Intelligence Index, it scored 48 points — the best result among open US models, but globally it trails leaders like Kimi K2.6 (54 points) and DeepSeek. Open models, according to analysts, lag behind closed ones by three to seven months.
But this lag matters less and less if an open model is simply good enough 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 expose confidential information to outsiders.
My view: NVIDIA's bet on efficiency, rather than test benchmarks, may prove more far-sighted. In mass AI adoption, the operational cost of a model takes center stage, and one that is almost as smart but five times cheaper wins in real-world deployment. NVIDIA has the resources, motivation, and distribution channels to release increasingly powerful open models faster than any other company. The ecosystem will only grow stronger.