Are DeepSeek moments now the new normal?

China's Moonshot Al's performance and development cost raise questions about Silicon Valley tech giants' huge outlays.

- Catherine Thorbecke

A little-known Chinese artificial intelligence (AI) company recently released an open-source reasoning model that challenged Western dominance and was developed at a fraction of the cost. And no, it’s not DeepSeek.

When Moonshot AI, a Beijing-based laboratory, launched Kimi K2 Thinking earlier in November, it went viral in tech circles.

A partner at prominent Silicon Valley venture capitalist firm Menlo Ventures called it a “turning point in AI”. The model now ranks second on Artificial Analysis’ intelligence index, behind only OpenAI’s GPT-5.1 - and ahead of the latest offerings from Alibaba Group Holding and DeepSeek, as well as US titans like xAI and Anthropic.

Looking at another benchmark measuring more complex, problem-solving “agentic” tasks, it even outperformed OpenAI.

This time around, however, markets barely batted an eye. As Bloomberg Economics analyst Michael Deng noted: “The contrast with January’s DeepSeek panic, which wiped almost US$600 billion (S$782 billion) off Nvidia in a single day, reveals how quickly investors have internalised that Chinese labs can match frontier capabilities at lower cost.”

Have we already reached the point where matching the best in AI on a shoestring budget is no longer a shock?

It’s true that it has become increasingly hard to judge model performance based on benchmarks alone. Moonshot’s latest release joins an especially crowded domestic market.

Launches and updates from Alibaba, Zhipu and MiniMax have come at a frenetic pace in 2025.