DeepSeek V4 Model Expected Late April 2026, Confirmed to Run on Huawei Ascend Chips

DeepSeek, the Chinese AI research company, is preparing to launch its V4 model in late April 2026, according to multiple Reuters reports citing The Information. The upcoming model represents a strategic shift toward domestic hardware, with DeepSeek confirming it will run on Huawei's latest Ascend chips rather than Nvidia or AMD processors. The company has spent months working with Huawei and Cambricon to rewrite parts of the model stack and conduct testing, signaling a deliberate move toward China's technological independence in AI infrastructure.

While the exact launch date remains unconfirmed, the model appears to be in late-stage preparation with two additional variants reportedly in development. Current public API documentation still lists only DeepSeek-V3.2 models, meaning V4 should be treated as imminent but not yet available for production integration. The January 2026 reports indicated the model would emphasize coding capabilities and extremely long coding prompts, suggesting specialized performance for software development workflows.

This hardware strategy marks a significant departure from industry norms, where most large language models are optimized for Nvidia's ecosystem. By aligning with Huawei's chip architecture, DeepSeek positions itself as a key player in China's push for AI sovereignty while potentially gaining performance advantages through hardware-software co-design. The move could reshape global AI competition by demonstrating viable alternatives to Western-dominated chip markets.