DeepSeek V4 Launches with Huawei Ascend Integration and 73% Cost Reduction — China's AI Independence Stack Matures

DeepSeek released a preview of its V4 model series on April 24, 2026 — one year after R1 disrupted global AI markets in January 2025. The flagship V4-Pro features 1.6 trillion parameters and a 1 million token context window (up from 128K in V3), with validated compatibility on both Nvidia GPU and Huawei Ascend NPU platforms. Huawei issued a separate statement confirming its Ascend 950 chips fully support V4, with its 'Supernode' technology combining large Ascend 950 clusters underpinning inference infrastructure. DeepSeek published API pricing showing a ~73% inference cost reduction vs V3: V4-Pro costs 1 yuan/M tokens (cached input), 12 yuan/M uncached, and 24 yuan/M output; the lighter V4-Flash costs 0.2/1/2 yuan per million tokens. The launch triggered a 9% surge in SMIC shares and a 15% jump in Hua Hong Semiconductor in Hong Kong trading, reflecting investor confidence in accelerating domestic chip demand. DeepSeek remains fully open-source under a permissive license, continuing its strategy of winning global AI distribution. The launch arrived one day after Commerce Secretary Lutnick's Senate confirmation of zero Nvidia H200 chip purchases by China — together, these events validate China's deliberate strategy of building competitive AI capability entirely within the domestic hardware ecosystem. DeepSeek's V4 Ascend validation closes the loop: China is not merely blocked from US chips, but has engineered a complete domestic AI stack where the model, hardware, and inference infrastructure are co-designed outside US supply chains.