DeepSeek blew up markets year ago. Why hasn't it done so since?

Almost a year ago, DeepSeek made headlines by significantly disrupting the AI landscape. The announcement of an advanced model from this lesser-known Chinese AI lab sent shockwaves through the market, leading to a dramatic plunge in the stock prices of major Western tech companies. Nvidia experienced a staggering 17% drop, resulting in a loss of nearly $600 billion in market capitalization. Similarly, Broadcom and ASML saw their stocks fall by 17% and 7%, respectively, within a single day. Fast forward eleven months, and these tech giants have not only recovered but thrived. Nvidia became the first company to reach a $5 trillion valuation in October, while Broadcom's shares surged by 49% over 2025, and ASML's stock climbed by 36%. According to Haritha Khandabattu, a senior director analyst at Gartner, the initial release from DeepSeek prompted a significant reevaluation of the global perspectives on AI competitiveness and cost structures, particularly in relation to semiconductors and hyperscalers. However, since that pivotal moment, DeepSeek has rolled out seven updates to its models, but none have generated the same level of market disruption as its January release. Founded in 2023, DeepSeek launched its free, open-source large language model (LLM) named V3 late in 2024, claiming it was trained using less powerful chips and at a fraction of the cost compared to leading models from OpenAI and Google. The subsequent release of its reasoning model, R1, in January 2025, matched or exceeded the performance of many top LLMs, catching the market off guard. The initial excitement about DeepSeek's capabilities led to concerns about the potential effects on AI infrastructure demand and revenue streams for companies like Nvidia. However, contrary to those fears, spending on AI technologies continued unabated in 2025, with forecasts suggesting even greater expenditure in 2026 and beyond. Analysts noted that the updates DeepSeek has released since January have mostly been enhancements of existing models rather than groundbreaking new introductions, leading the market to perceive them as a continuation of progress rather than a groundbreaking transformation. Analysts believe that DeepSeek's limited releases may stem from constraints on computing resources. The company has faced difficulties in training its R2 model, initially slated for release in May, due to challenges associated with using Huawei chips, as Chinese authorities have pushed for reduced reliance on U.S. technology amid ongoing export restrictions. Chris Miller, author of "Chip War," highlighted that the shortage of advanced computing power in China, largely due to U.S. chip sales restrictions, hampers the development of cutting-edge AI models. Despite these challenges, DeepSeek recently acknowledged its limitations compared to advanced closed-source models like Gemini 3, particularly regarding computing capabilities. Meanwhile, the market has seen the continued emergence of advanced models from Western competitors, including OpenAI's GPT-5 and Google's Gemini 3, which have alleviated fears of a sudden shift towards commoditization. Looking ahead, there are indications that DeepSeek may be preparing for a significant model release soon. On New Year’s Eve, the company published a research paper outlining a more efficient approach to AI model development. Analysts like Dan Ives from Wedbush Securities believe the market can expect more impactful developments from DeepSeek in the near future, hinting that the company may still have surprises in store for investors.