DeepSeek Launches V4 Open-Source Model, Challenging US AI Rivals and Boosting Chinese Chipmakers
DeepSeek V4 delivers top-tier AI performance at 15-20% of US rivals' inference costs.
DeepSeek, the Hangzhou-based AI startup, has unveiled its V4 open-source model, which analysts say could trigger a broad reassessment of stocks across the AI and semiconductor supply chain. V4 delivers top-tier performance on coding benchmarks and shows significant advantages in reasoning and agentic tasks, competing directly with OpenAI and Anthropic. Crucially, Morgan Stanley reports that Chinese models like V4 achieve similar performance to US peers at just 15–20% of inference costs, lowering the threshold for SMEs and individuals to adopt high-performance AI. This efficiency, combined with DeepSeek's reported funding talks with Tencent and Alibaba, positions V4 as a catalyst for broader commercial AI adoption in China.
The V4 launch is a major boon for China's domestic chip ecosystem. DeepSeek optimized the model for homegrown hardware, benefiting chipmakers such as Cambricon Technologies, Moore Threads, Hygon, and foundry SMIC. The AI chip market in China is projected to grow from $19.8B in 2024 to $196.2B by 2029 (54% CAGR). Stock markets have already reacted: the Star Market 50 Index jumped 25% in April, and Cambricon hit a record high. Meanwhile, brokers expect MiniMax and Zhipu to join the Hang Seng Tech Index, drawing up to $1.75B in passive inflows. DeepSeek's collaboration with domestic chipmakers accelerates the development of a self-controlled, sanctions-resilient AI computing stack in China.
- DeepSeek V4 matches US AI models in coding/reasoning at 15-20% of inference costs.
- Model optimized for Chinese chips, boosting Cambricon, SMIC, and Moore Threads.
- China's AI chip market projected to grow 54% annually to $196.2B by 2029.
Why It Matters
DeepSeek V4 democratizes AI in China, challenging US dominance while fueling a domestic semiconductor boom worth $196B.