DeepSeek’s preview release of its V4 large language model on Friday triggered an immediate market split in Hong Kong, with semiconductor stocks surging although several AI application developers slid sharply.
Shares of Semiconductor Manufacturing International Corp rose 10 percent by the close and Hua Hong Semiconductor gained 15 percent, extending a broader rally among mainland chipmakers as investors rotated into companies expected to benefit from rising demand for computing power. In contrast, Knowledge Atlas Technology, likewise known as Zhipu AI, fell 9.1 percent, and MiniMax dropped 9.4 percent, reflecting pressure on application-layer firms as the new model intensified competition in China’s generative AI sector.
The move follows DeepSeek’s pattern of releasing open-source models that challenge the cost and performance assumptions of Western AI leaders. The Hangzhou-based startup said V4 achieves strong performance against domestic rivals, particularly in agent-based tasks, knowledge processing, and inference, and is optimized for use with tools like Anthropic’s Claude Code and OpenClaw. Like its predecessors, V4 is available in both “pro” and “flash” versions, allowing developers to download, run, and modify the code locally in most cases.
DeepSeek first gained global attention in late 2024 with its V3 model, trained on less powerful chips at a fraction of the cost of systems from OpenAI and Google. In January 2025, its R1 reasoning model disrupted markets by matching or exceeding leading LLMs despite being built in two months for under $6 million using lower-capacity Nvidia hardware—a feat that questioned the U.S. Lead in AI and the scale of Large Tech’s infrastructure spending. Since then, no subsequent release has matched R1’s impact, as traders had already priced in the reality that Chinese AI is both competitive and cheaper to deploy.
“DeepSeek’s V4 preview is a serious flex,” offering lower inference costs than previous models, Neil Shah, vice president of research at Counterpoint Research, told CNBC. Inference costs—the computational and financial expenses of running a trained AI model to generate outputs—are a key metric in model efficiency. According to Counterpoint’s principal AI analyst Wei Sun, V4’s benchmark profile suggests it could deliver “excellent agent capability at significantly lower cost.”
However, analysts note the timing of V4’s release differs from R1’s debut. Ivan Su, senior equity analyst at Morningstar, said V4 is unlikely to shock markets the way R1 did, because the competitive pricing and performance of Chinese AI models are now widely anticipated. “This is a framing that didn’t exist with R1,” Su added, “and that alone tells you how much domestic competition has intensified.” Since R1’s launch, firms like Alibaba and ByteDance have accelerated their own model releases, increasing pressure on DeepSeek to maintain its edge.
The launch arrives amid tightening U.S. Export controls on advanced Nvidia chips, which have pushed Chinese firms to accelerate development of domestic alternatives. DeepSeek was reportedly in talks with Tencent and Alibaba for its first funding round, a detail notable given that Alibaba owns the South China Morning Post, one of the sources reporting on the chipmaker rally.
How DeepSeek’s model releases have shifted investor expectations over time
The market reaction to V4 reveals a maturing narrative around China’s AI capabilities. Where R1’s debut forced a reassessment of U.S. Dominance and sparked volatility across tech stocks, V4’s release was met with a more segmented response: chipmakers rose on anticipated demand, while application developers faced renewed competitive pressure. This divergence shows investors now distinguish between infrastructure beneficiaries and application-layer firms in the AI value chain.

What the V4 release signals about China’s domestic AI race
DeepSeek’s positioning of V4 as a direct challenger to other Chinese open-source models reflects a shift from earlier dynamics. When R1 launched, there was little direct comparison to domestic peers; now, Su noted, the framing explicitly pits Chinese models against one another. This internal competition suggests the sector is moving beyond the initial phase of catching up to Western leaders and into a stage where differentiation, cost efficiency, and integration with developer tools become key battlegrounds.
Why did chipmakers rise while AI application stocks fell after the V4 preview?
Investors rotated into semiconductor stocks expecting higher demand for computing power to run and deploy models like V4, while AI application developers faced pressure as the new model intensified competition in the generative AI space, potentially eroding market share for existing players.
How does V4 compare to DeepSeek’s earlier R1 model in terms of market impact?
Analysts say V4 is unlikely to match R1’s market impact because traders have already priced in the competitiveness and cost efficiency of Chinese AI models, whereas R1’s debut came as a surprise that questioned U.S. Leadership and Big Tech’s spending assumptions.
