不到一年多以前，一家小型中国人工智能创业公司推出了几乎可以与最强西方模型媲美、但成本仅为其一小部分的模型，这一消息短暂引发了市场恐慌。英伟达等 AI 基础设施供应商的股价一度下跌，但随后回升。

4 月 24 日，该实验室发布了 v4 模型，但市场反应平淡。DeepSeek 的 v4 在性能上接近领先的美国模型，而且其最强版本“Pro”仅比 3 到 6 个月前发布的美国模型略低。

v4 对客户来说也很便宜，因此投资者对最初的恐慌进行了修正，对 AI 基础设施需求崩塌的担忧没有实现。上下文在“Intel, market capitalisation, $bn”处中断，因此目前只能确认“低成本”与“只差三到六个月性能代际”的数字化趋势。

A little more than a year ago, a small Chinese AI startup released models that performed almost as well as top Western models while costing only a fraction of the price, and this briefly triggered market panic. Nvidia and other AI-infrastructure providers’ shares fell for a moment, then recovered.

On April 24, the lab released the v4 model, but the market reaction was muted. DeepSeek’s v4 reached near the performance level of leading American models, and its strongest Pro version was only marginally behind those models launched three to six months earlier.

V4 is also cheap for customers, so investors eventually dialed back their initial fear and demand for AI infrastructure did not collapse as feared. The passage cuts off at “Intel, market capitalisation, $bn,” so so far only the quantitative trend of low cost versus earlier models and the three-to-six-month performance gap can be confirmed.

Source: Sequel struggles

Subtitle: Why DeepSeek's new model has been met with a shrug

Dateline: The Economist May 2nd 2026