China Telecom has developed the country’s first artificial intelligence models with the innovative Mixture-of-Experts (MoE) architecture that are trained entirely on advanced chips from Huawei Technologies. According to a technical paper published last month by China Telecom’s Institute of Artificial Intelligence (TeleAI), the TeleChat3 models, ranging from 105 billion to trillions of parameters, were trained on Huawei’s Ascend 910B chips and its open-source deep learning AI framework, MindSpore. TeleAI researchers stated that the Huawei stack met the “severe demands” of training large-scale MoE models across a range of sizes. “These contributions collectively address critical bottlenecks in frontier-scale model training, establishing a mature full-stack solution tailored to domestic computational ecosystems,” they added . China Telecom’s model lags behind OpenAI’s GPT-OSS-120B MoE architecture distributes tasks to multiple specialized submodels, or “experts.” Therefore, AI models developed with it can scale up capacity without significant increases in computational overhead. MoE was popularized by DeepSeek’s V3 model, released in December 2024, and has since become the norm for leading-edge Chinese AI models. MoE models, however, were considered more technically demanding to train and run. China Telecom’s self-reported performance scores for its TeleChat3 models showed that they lagged behind those of OpenAI’s GPT-OSS-120B, released in August, on several benchmarks. Last week, Tsinghua University spin said its new image-generation model was trained on Huawei chips, making it the first open-source model developed on an entirely domestic training stack to achieve industry-leading scores in image generation. Beijing-based Zhipu AI was blacklisted by Washington last January. The US has placed several Chinese technology companies, including Huawei and iFlytek, on export-control blacklists. This effectively bars them from receiving US-origin chips, semiconductor tools, and other advanced technology. Ant Group researchers, a fintech affiliate of Alibaba Group Holding, also said they successfully trained a 300-billion-parameter MoE model “without premium GPUs”. However, they did not specify whether they had exclusively used domestically designed chips. Meanwhile, as reported by Cryptopolitan, a Nasdaq-style index of local Chinese tech stocks has jumped nearly 13% just this month. A second gauge tracking Hong Kong-listed Chinese tech firms is up 6%, and both are leaving the Nasdaq 100 behind. Nvidia stock tanks as Beijing declares self-reliance Nvidia said that its advanced GPUs and machine-learning frameworks were the best tools in the world for training large-scale MoE models. However, Beijing has made self-reliance across the entire AI stack a key priority for the country in the next five years due to US trade restrictions that block Chinese firms’ access to advanced US chips. The US government recently gave the go-ahead for Nvidia to sell the H200, the firm’s second-most-powerful chip, to China. However, China moved to block shipments of advanced chips. Cryptopolitan reported that Beijing could be considering restrictions to advance local chip development or strengthen its negotiating position with the US. As a result, suppliers paused production of H200 components after the block. Nvidia had expected more than 1 million orders from Chinese customers, with suppliers gearing up for March deliveries, but customs officials reportedly refused entry for the chips. Nvidia shares have since slid about 3% after reports. According to analysts, Nvidia faces a clear risk. If China continues blocking H200 shipments, the stock could break a key near-term support. Should approvals ease, the boost could come fast, but policy uncertainty swings both ways. On the other hand, other chipmakers showed mixed moves as AMD climbed 1.7%, Intel fell 2.8%, while the S&P 500 ETF SPY dipped roughly 0.1%. Meanwhile, market watchers are looking out for NVDA’s upcoming February 25 quarterly earnings and any fresh details on its China export situation. If you're reading this, you’re already ahead. Stay there with our newsletter .
