Chinese AI model shakes Big Tech

DeepSeek’s success in producing a cost-effective model that runs on less-advanced chips has “cast doubt” on rich valuations for companies like US chipmaker Nvidia, Bloomberg reported, and could “potentially derail the investment case for the entire AI supply chain,” the director of an asset management firm told the outlet. DeepSeek’s own chatbot predicted as much in a conversation with technology professor Warwick Powell: “Investors may become more cautious about funding new AI ventures… They may shift focus to startups leveraging open-source technologies or those with innovative business models.” Still, some analysts told the Financial Times the market reaction was overdone, and that chips would become cheaper and more capable over time.

The DeepSeek model is AI’s “Sputnik moment,” venture capital investor Marc Andreessen wrote on X: the Chinese firm’s stripped-down approach “defies the usual assumption that you must have near-infinite resources to compete at the top tier,” which has in turn raised fears that China’s AI strategy is resilient in the face of Western sanctions and export controls, another investor argued for CTech. The achievement may even demonstrate just how counterproductive these restrictions have been, insofar as Chinese AI companies have been forced to experiment with novel approaches, thus “spurring breakthroughs that might not have occurred otherwise,” a legal expert argued in the Financial Times.

DeepSeek’s bot is subject to a layer of pro-Chinese Communist Party censorship, according to analysis of its system prompts by tech expert Jim the AI Whisperer: The bot refuses to answer questions about Tiananmen Square, the persecution of the Uyghurs, and even comedy mocking Chinese leader Xi Jinping, having been instructed to “avoid any content that undermines the stability, unity, or dignity of the People’s Republic of China.” But when restricted to yes/no answers, DeepSeek was seemingly able to bypass the censorship, answering “Y” to “Tank man right?”