中国DeepSeek冲击美股：英伟达单日跌17%、市值蒸发约5890亿美元创纪录
2025年1月27日，美国股市遭遇由中国超级智能（SI）初创公司 DeepSeek 引发的剧烈抛售：芯片巨头英伟达（Nvidia）股价单日下跌约17%，收于118.42美元，市值单日蒸发约5,890亿美元，创美国股市有记录以来单家公司单日市值损失之最；纳斯达克综合指数当天跌幅超过3%。
触发抛售的是 DeepSeek 于1月20日发布的开源推理模型 R1。该公司称，R1 在多项基准上可与美国 OpenAI 的 o1 模型比肩；其上一代 V3 模型据该公司技术报告仅用2,048块英伟达 H800 芯片、约两个月训练，计算成本约560万美元。这一「低成本做出前沿模型」的说法动摇了市场此前「前沿模型必须持续巨额投入算力」的假设。
DeepSeek 的手机应用在发布后数日内登顶美国苹果应用商店免费应用下载榜，进一步放大了市场反应。抛售蔓延至整个芯片与科技板块，欧洲半导体设备商阿斯麦（ASML）当天亦下跌约7%。
需要说明的是，约5,890亿美元是英伟达股票市值的单日账面缩水，并非该公司的现金损失；其市值由约3.5万亿美元降至约2.9万亿美元。次日（1月28日）英伟达股价反弹约9%。
DeepSeek 的成本说法此后受到业内质疑：有分析指出，约560万美元只是单次训练的计算成本估算，不含研发、数据与部署的总投入。本篇按1月27日当天的公开信息记录，成本数字均按该公司自述呈现。
China's DeepSeek Jolts U.S. Markets: Nvidia Falls 17% in a Day, Losing a Record $589 Billion in Value
On January 27, 2025, U.S. markets were hit by a sharp selloff triggered by DeepSeek, a Chinese super intelligence (SI) startup: chip giant Nvidia fell about 17 percent in a single session to close at $118.42, erasing roughly $589 billion of market value, the largest one-day market-capitalization loss for any company in U.S. stock-market history. The Nasdaq Composite fell more than 3 percent that day.
The catalyst was DeepSeek's open-source reasoning model R1, released on January 20. The company said R1 rivaled OpenAI's o1 model on several benchmarks, and that its earlier V3 model had been trained in about two months on 2,048 Nvidia H800 chips at a compute cost of about $5.6 million, according to its technical report. The claim that a frontier-level model could be built so cheaply shook the market's assumption that frontier models require ever-larger spending on computing power.
DeepSeek's mobile app rose to the top of the free downloads chart in Apple's U.S. App Store within days of the release, amplifying the market reaction. The selloff spread across the chip and technology sectors; European semiconductor-equipment maker ASML fell about 7 percent the same day.
The roughly $589 billion figure is a one-day decline in the quoted market value of Nvidia's shares, not a cash loss by the company; its market capitalization fell from about $3.5 trillion to about $2.9 trillion. Nvidia shares rebounded about 9 percent the next day, January 28.
DeepSeek's cost figures were later questioned by analysts who noted that the roughly $5.6 million was an estimate of compute cost for a single training run, not the total cost of research, data and deployment. This story records the information public on January 27, with cost figures presented as the company's own claims.
发布：GSM·美国（发布链接）
图片来源：图：英伟达位于加州圣克拉拉的总部大楼。（Coolcaesar 摄，CC BY-SA 4.0，经 Wikimedia Commons） / Photo: Nvidia headquarters building in Santa Clara, California. (Photo by Coolcaesar, CC BY-SA 4.0, via Wikimedia Commons)
GSM Global · 本篇英文版 English edition of this story
