DeepSeek’s AI Selloff: Market Overreaction or Long-term opportunity?

Yesterday, global equity markets reacted sharply to the announcement that DeepSeek’s latest AI assistant surged to the top of Apple’s iPhone download charts over the weekend, sparking market-wide doubts about America’s dominance in AI. Its claims of a high-performing model developed at a fraction of the cost rattled global investors, triggering sharp declines across AI-related stocks. The Nasdaq was down nearly 5% in pre-market trading, with several AI-related stocks dropping 10-20%. European technology names, including ASML and Siemens Energy, also saw similar declines. This immediate response highlights how markets often overreact to attention-grabbing headlines, and how sentiment drives short-term volatility.

However as always, it’s important to take a step back and put things into perspective. The excitement surrounding DeepSeek is largely centred on the fact that its app became the most downloaded on the app store and that it is open-source—a notable achievement, but one that doesn’t derail the broader AI theme. If anything, it reinforces the importance and growth of AI by showcasing the investment flowing into the sector and the increasing accessibility of AI tools. Far from undermining AI’s potential, this development reflects the global drive to integrate AI into everyday applications, which in the longer term should strengthen the theme as Jevon’s paradox will most likely still hold true even in the age of AI.

The initial market reaction also failed to account for the nuances behind DeepSeek’s claims that its cheaper and can do the same as much more expensive models. The company itself acknowledges a “4x compute disadvantage” compared to global leaders, requiring significantly more data and computing power to achieve comparable results. Their reliance on a Mixture-of-Expert (MoE) architecture has allowed for cost efficiencies but comes with trade-offs, including challenges with generalisation and scalability in real-world use. Moreover, DeepSeek’s reported efficiency in using fewer chips likely masks the significant prior compute resources needed to refine their techniques—resources that aren’t fully disclosed. For instance, DeepSeek has previously operated Asia’s first 10,000 A100 cluster and reportedly has access to a 50,000 H800 cluster, in addition to cloud-based resources. It is likely that extensive trial and error tests on these systems significantly contributed to their results.

Then there is also a layer of irony in DeepSeek’s transparency. While Chinese firms like DeepSeek openly share model weights and methodologies to attract global attention, Western companies such as OpenAI and Microsoft have chosen to keep their most advanced models private, making comparisons like those announced by DeepSeek more or less irrelevant. Yet, like all other Chinese-made AI models, DeepSeek self-censors on politically sensitive topics, such as Tiananmen Square, President Xi Jinping, or the possibility of China invading Taiwan. This could prove jarring for international users unfamiliar with Chinese chatbots and highlights how such models are shaped by domestic constraints and likely have ties to the Chinese government.

The sharp market swings triggered by news like this highlight how sentiment often overshadows reality in the short term. While headlines can dominate market movements, the longer-term implications—such as increased investment in AI, its democratisation, and the ongoing integration of these technologies—are what truly matter.