According to Beating AI News, ByteDance is negotiating to purchase 50000 AI chips from Tiantian Zhixin. The foundation of China's AI computing power is being quietly replaced and is undergoing localization substitution. ByteDance beanbag has a monthly life of 330 million, ranking second in the world. Behind such a large user base is a massive demand for inference computing power. ByteDance's newly purchased chip is a Chinese company, Tiantian Zhixin, which just rang the bell on the Hong Kong Stock Exchange in January 2026（ http://9903.HK ）. After Nvidia was banned, its market share in China continued to decline. But demand will not disappear out of thin air. Domestic GPU manufacturers already captured nearly 41% of the domestic AI accelerator server market last year, and this replacement speed is much faster than most people imagine. Taking a look at Tiantian Zhixin, 16800 pieces were shipped in 2024, and by the first half of 2025 alone, 15700 pieces were released, almost equaling the entire year of 2023. Accumulatively delivered over 52000 pieces and served over 290 customers. Huatai Securities expects its shipment volume to exceed 100000 units in 2026, a year-on-year increase of 139%. This Nima is no longer a trial run, it is a large-scale production. Key question: Why? Why it? The answer is easy to migrate. Tiantian Zhixin is deeply compatible with the CUDA ecosystem, allowing enterprises to migrate from Nvidia at almost zero cost. This is the real reason why big companies are willing to pay. In terms of performance, the average performance of the Tianshu architecture in the DeepSeek V3 scenario is about 20% higher than that of the NVIDIA Hopper architecture. The average price of the intelligent armor series of inference chips is about 12000 yuan, and the cost-effectiveness is also in place. The logic behind ByteDance's large-scale purchase of intelligent armor inference chips is that the user base of Doubao is there, and inference computing power is in high demand. Domestic chips can catch up at this point in time. Tencent is also buying AI chips from Baidu Kunlun Chip, and the National Development and Reform Commission has clearly stated its intention to promote the adaptation of domestic large models to domestic computing power. So, you know, besides, Nvidia currently does not have an advantage in performance for China, so domestic chips can meet the demand for inference. Now it's not just one company looking for a replacement, the entire industry chain is also being restructured. The foundation of China's AI computing power is undergoing a silent but thorough overhaul. The gap left by the US ban has been filled by domestically produced chips, and the entire computing infrastructure has been developed. However, this script has just turned the first page.