Nvidia Reportedly Building Trillion-Parameter Nemotron 4 to Match Chinese Open Models
Nvidia is reportedly building Nemotron 4, an open-weight model with at least one trillion parameters — double the size of Nemotron 3 Ultra. The company has tripled its cloud spending on in-house training to $28 billion through 2031, with an earliest possible release this fall.
Nvidia is developing Nemotron 4, an open-weight model that will scale to at least one trillion parameters, according to a report from The Information. That would make it twice the size of Nemotron 3 Ultra, the company's current flagship open model.
Nvidia has tripled its cloud spending on in-house model training to $28 billion through 2031, the report states. The earliest possible release window for Nemotron 4 is this fall.
Still Behind Chinese Labs
Even at one trillion parameters, Nemotron 4 would only reach a scale that Chinese labs have already surpassed. Moonshot AI's Kimi K3 reportedly has 2.8 trillion parameters, and DeepSeek V4 Pro has 1.6 trillion — both larger than Nvidia's planned model.
The performance gap tells a similar story. Nemotron 3 Ultra launched in June as the strongest open US model on the Artificial Analysis Intelligence Index, but it still trailed Kimi K2.6 at release. On the current version of the index, Nemotron 3 Ultra scores 38 points, while Kimi K3 scores around 60 — a substantial lead for the Chinese model.
Policy Backdrop
Nvidia is among the signatories of a petition opposing regulation of open-weight models. This comes as the Trump administration reportedly considers targeted bans on specific Chinese models, a policy direction that would directly affect competitors like Kimi and DeepSeek that currently outperform Nvidia's open offerings.
A Conflicted Position
Nvidia's strategy here is unusual for the company. Nvidia's primary business is selling GPUs, and self-hosted open models drive that demand — the more companies run open weights on their own infrastructure, the more hardware Nvidia sells. Building a top-tier open model in-house supports that flywheel.
But Nemotron 4 also puts Nvidia in direct competition with major customers, including OpenAI, which buys enormous volumes of Nvidia compute while building its own frontier models. Whether Nvidia can credibly compete on model quality while remaining a critical infrastructure supplier to labs it now rivals remains an open question.
No benchmark scores, pricing, context window, or exact release date have been disclosed for Nemotron 4. The one-trillion-parameter figure and fall release timeline are described as the earliest possible scenario, not confirmed specifications.
What This Means
Nemotron 4's reported scale illustrates how far ahead Chinese open-weight labs have moved. A trillion parameters was frontier territory a year ago; today it would place Nvidia below Kimi K3 and DeepSeek V4 Pro on raw size, and Nemotron 3 Ultra's current Intelligence Index score of 38 versus Kimi K3's roughly 60 suggests the gap isn't just parameter count — it's capability. Nvidia's $28 billion training commitment through 2031 signals a long-term bet on owning a frontier open model rather than ceding that ground entirely to Chinese labs or US closed-model providers. The regulatory angle adds friction: Nvidia lobbying against restrictions on open models while the US government weighs bans on the very Chinese models currently leading that category is a tension worth watching as Nemotron 4's release approaches.
Related Articles
Nvidia Releases Nemotron 3.5 Lightning: A 31.6B-Parameter Open Model Built for Speed, Not Peak Intelligence
Nvidia's Nemotron 3.5 Lightning, a 31.6B-parameter open-weight model with only 3.6B active parameters, matches OpenAI's gpt-oss-120b on the Artificial Analysis Intelligence Index while delivering the fastest throughput in its class at nearly 670 tokens per second. The model posts especially large gains on agentic benchmarks, beating both gpt-oss-120b and the larger Nemotron 3 Super.
NVIDIA Releases Nemotron 3.5 Lightning 30B-A3B: 3B-Active MoE Model With 1M-Token Context, Quantized for Single-GPU Depl
NVIDIA has published NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4, a 30-billion-parameter Mixture-of-Experts model with only 3B active parameters, a hybrid Mamba-2/MoE/Attention architecture, and support for up to 1 million tokens of context. The NVFP4-quantized checkpoint is designed to run on a single DGX Spark (GB10) or H100 GPU.
NVIDIA Releases Alpamayo 2 Super, a 34B Vision-Language-Action Model for Autonomous Driving
NVIDIA has released Alpamayo 2 Super, a 34B-parameter foundation model for autonomous vehicle development that combines a 32B vision-language backbone with a 2.3B-parameter diffusion action decoder. The model handles trajectory prediction, visual question answering, 2D grounding, and auto-labeling, and posts a Lingo-Judge score of 79.2 on LingoQA reasoning evaluation.
NVIDIA Releases Magpie TTS Multilingual Update: 364M-Parameter Open-Weights Model Now Supports 12 Languages, Sub-50ms La
NVIDIA's Magpie TTS Multilingual, a 364M-parameter open-weights text-to-speech model, now supports 12 languages after adding Modern Standard Arabic, Korean, and Brazilian Portuguese. The model achieves 32ms time-to-first-audio on B200 GPUs and improves speech quality across French, Spanish, and German.
Comments
Loading...