Gary Marcus 引述 Gerstner 观点：Nvidia、CoreWeave 股价大跌，AI 泡沫争议升温
Gary Marcus 拿 12 个月前的“没有泡沫”论和现在的股价对照，Nvidia 跌 16%、CoreWeave 跌 55%，还顺带澄清 DeepSeek 600 万美元训练成本是被误读的。做 AI 投资判断或想看清这轮周期的人值得花两分钟看看这组数字。
Gary Marcus 在社交媒体转述投资人 Brad Gerstner 的言论，对比后者 2025 年 1 月“没有 AI 泡沫”的判断与随后 12 个月的市场现实：Nvidia 股价下跌 16%，CoreWeave 下跌 55%，Oracle 下跌 52%，Anthropic 估值在数周内缩水 30%。Marcus 据此认为市场狂热期可能已经结束。
帖子同时回顾了围绕 DeepSeek 训练成本的争论。Gerstner 曾指出，网传 DeepSeek 仅花 600 万美元训练、而 OpenAI、Anthropic 和 xai 花费数十亿美元的说法不实；同口径对比应是 600 万美元对 1500 万美元，且 DeepSeek 晚了近一年，成本本应再低至少 50%。Dario Amodei 方面确认 DeepSeek 性能落后 7-10 个月，实际训练开销在数千万美元量级。
这条内容把二级市场股价、私募估值变化与 DeepSeek 成本争议放在同一时间线里，说明此前支撑“无泡沫”论断的部分数据后来被修正，对关注 AI 板块估值的投资人和从业者有直接参考价值。
Gerstner: “No AI bubble” (January 2025) Reality, over last 12 months Nvidia is down 16% CoreWeave is down 55% Oracle is down 52% the credit markets have had enough and Anthropic’s valuation dipped 30% in a matter of weeks. The party is probably over. Brad Gerstner @altcap DeepSeek should not be dismissed but is NOT a Chinese break thru on AI cost which caused Monday’s panic. The reports they spent $6 M vs. billions for @OpenAI , @AnthropicAI & @xai are fake news. Apples for Apples is $6 M vs $15 M. And given they did nearly a year later, should be at least 50% less costly. Anthropic confirms today - Deepseek performance 7-10 months behind; right compare is 10s of millions; and you would expect 3-4x cheaper per yr due to falling compute cost. @DarioAmodei
This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include DeepSeek, Nvidia, Anthropic, OpenAI.
Source attribution: curated from Gary Marcus (@GaryMarcus). Cross-source corroboration: 11 independent sources have covered this development (including 小互, Latent Space (swyx), TestingCatalog), confirming industry-wide signal strength.
