A renowned semiconductor and AI research firm has published analysis suggesting that the competitive landscape among Chinese AI labs is yielding significant results.
The report states that the newly released Kimi K3 model has significantly outperformed NVIDIA's flagship open-source model, Nemotron 3 Ultra, in benchmark tests. The analysis directly addresses NVIDIA CEO Jensen Huang, arguing that the "Nemotron Committee" development model he championed has proven ineffective as the right path for open-source AI in the United States, and calls for NVIDIA to reconsider its strategic direction.
The firm's view is that Chinese AI labs, through continuous iteration and rapid delivery in a free market competitive environment, have established a clear advantage in the open-source domain. In contrast, if the U.S. continues with its current committee-based collaborative model, it risks falling further behind in this race.
Committee Model Fosters "Groupthink"
The failure of the Nemotron Committee model is attributed to its structural flaws. According to the analysis, when Jensen Huang established the Nemotron Committee, the mechanism restricted the free flow of different technical approaches, creating groupthink. This fundamentally conflicts with the core value of open source, which is the capacity for free experimentation.
The problems extend further. The report reveals that mistakes occurred during Mistral's pre-training phase within Nemotron. Because NVIDIA employed a single committee structure, an error by one member company negatively impacted the entire committee's final output, leading to cascading dissatisfaction among members.
Perhaps more concerning is a crisis of trust. The analysis indicates that, based on its understanding, some alliance members are reluctant to share their best ideas, highest-quality datasets, and evaluation methods with the consortium, resulting in internal friction. The firm concludes that committee-based training is "extremely clearly" not the way forward.
Chinese Model Offers a Contrast
The competitive ecosystem of Chinese AI labs is presented as a contrasting reference. The report notes that while these labs compete with each other, they also learn from each other's proven successful experiences. This process has fostered various innovations such as KDA, MSA, and DSA.
In the firm's assessment, the dominant lead of Kimi K3 over Nemotron 3 Ultra is a direct manifestation of the superiority of this competitive mechanism. True technological progress is driven when independent labs compete to test different ideas, architectures, and technical pathways.
Proposed Roadmap for Reform
The analysis does not advocate for NVIDIA to abandon the committee model entirely but proposes a more specific improvement plan. The suggestion is that if Jensen Huang genuinely wishes to pursue committee-based training, he should at least form three separate, isolated committees that do not communicate with each other to avoid groupthink.
These three alliances should innovate independently and compete against each other in dimensions such as data, reinforcement learning, pre-training, and evaluation. The firm believes this arrangement would align more closely with American free-market values and more closely resemble the actual operational model of Chinese AI labs—iterating through competition and differentiating genuine technological advancements through that iterative process.
Comments