Nvidia builds 1-trillion-parameter model, trailing Chinese labs

Nvidia's investment in Nemotron 4 reflects a strategic pivot into competing with the world's best open-weight models, but the scale trajectory reveals a lag behind Chinese laboratories. At one trillion parameters, Nemotron 4 would reach a milestone that Moonshot AI (with Kimi K3 at 2.8 trillion) and DeepSeek (with V4 Pro at 1.6 trillion) have already surpassed. This gap translates to measurable performance: Nemotron 3 Ultra, which was the strongest open US model at launch in June, now scores 38 points on the Artificial Analysis Intelligence Index compared to Kimi K3's around 60—a substantial differential despite Nemotron 3 Ultra's initial market position.

Nvidia's $28 billion commitment through 2031 signals conviction in the open-model strategy, which aligns with the company's commercial interests: more companies self-hosting open models increases demand for GPUs. However, Nemotron 4 creates a direct competitive conflict with major customers like OpenAI, potentially straining those relationships. Simultaneously, Nvidia has joined a petition opposing regulation of open models, even as the Trump administration weighs targeted bans on Chinese models—a positioning that protects Nvidia's open-model business but also benefits Chinese competitors operating in the same regulatory space.