Chinese artificial-intelligence start-up DeepSeek has postponed the launch of its R2 large language model after failing to complete training on Huawei’s Ascend 910 series processors, the Financial Times reported on 14 August, citing people familiar with the matter. Engineers encountered unstable performance and connectivity limits on the domestic chips, forcing the company to revert to Nvidia hardware for training while still aiming to run the finished model on Huawei silicon for inference. The setback pushes the release beyond an already-slipped timetable that had moved from an initial May target to the second half of August. The episode underscores the technical gap between Huawei’s Ascend line and Nvidia’s GPUs—as well as the challenges Beijing faces in its drive to replace U.S. components in advanced AI systems. A delay for one of China’s highest-profile ChatGPT rivals also preserves demand for Nvidia’s restricted H20 accelerators, complicating government efforts to build a fully self-reliant AI supply chain.