Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-source AI model that rivals top systems in the U.S. [1].
The launch marks a significant escalation in the global AI race, demonstrating China's ability to produce massive models that compete with Western technology. However, the development has sparked a dispute over the hardware used to train the system.
Reports suggest that Alibaba Group, which backs the Beijing-based Moonshot AI, entered an agreement allowing the startup to use Nvidia H200 GPUs [2]. These high-performance chips are critical for training models of this scale, but their acquisition in China is often subject to strict trade restrictions.
Alibaba denied the existence of the chip-supply agreement [2], the company said.
Despite the hardware controversy, the Kimi K3 is being described as the largest open-source model ever released [1]. By making the model open-source, Moonshot AI allows a broader community of developers to analyze and build upon the 2.8-trillion-parameter architecture [1].
The reliance on Nvidia hardware remains a central point of contention for observers of the Chinese tech sector. While the Kimi K3 demonstrates high-tier performance, the alleged use of H200 chips suggests that top-tier AI progress in the region may still depend on U.S.-made silicon [2].
“Kimi K3 is the largest open-source model ever released.”
The emergence of Kimi K3 proves that Chinese firms can architect models with trillion-parameter scales, but the dispute over Nvidia H200 chips highlights a critical vulnerability. If China's most advanced AI is dependent on restricted U.S. hardware, the sustainability of its AI growth remains tied to the volatility of trade policy and supply chain access.
