U.S. tech stocks faced significant pressure on Monday as investors reassessed the economic implications of Chinese startup DeepSeek's rapidly advancing models, including its newly unveiled Janus-Pro-7B image generator.
The Hangzhou-based company announced that its new model outperforms OpenAI's DALL-E 3 in benchmarks. Additionally, DeepSeek's assistant app has overtaken ChatGPT in U.S. App Store downloads. The company also revealed that it had temporarily limited new user registrations following a cyberattack.
Analysts warned that the selloff highlighted deeper vulnerabilities: high valuations, concerns about overhyped infrastructure spending, and growing evidence that U.S. chip export controls are failing to stifle Chinese AI progress.
Nvidia led the decline, plunging nearly 17 percent to erase almost $600 billion in market value—the largest single-day loss in Wall Street history. The Nasdaq tumbled 3.1 percent, and the Philadelphia Semiconductor Index cratered 9.2 percent, its worst day since March 2020. Oracle, Broadcom, and data center firm Vertiv Holdings each fell over 13 percent, while power utilities tied to AI infrastructure demand—including Vistra and Constellation Energy—sank by up to 28 percent. The rout spread globally, hitting Japan's SoftBank and Europe's ASML.
\"If it's true that DeepSeek is the proverbial 'better mousetrap,' that could disrupt the entire AI narrative that has helped drive the markets over the last two years,\" said Brian Jacobsen of Annex Wealth Management. \"It could mean less demand for chips, less need for a massive build-out of power production to fuel the models, and less need for large-scale data centers.\"
Model Momentum Meets Market Mania
DeepSeek's dual releases—the Janus-Pro-7B and the DeepSeek-R1, just days ago—intensified existing doubts. The Janus-Pro-7B, an upgrade to December's image model, uses 72 million synthetic images to achieve what developers call \"significant advancements\" in both multimodal understanding and text-to-image instruction-following capabilities. Meanwhile, the DeepSeek-V3-powered assistant's sudden U.S. popularity highlighted China's consumer AI prowess. Most jarring to markets, researchers claimed the V3 model was trained on just $6 million worth of Nvidia's restricted H800 chips—a fraction of industry norms—and that the R1 model's API is sold at around 1/30th of OpenAI's prices.
Silicon Valley figures amplified the alarm. Venture capitalist Marc Andreessen called DeepSeek's rise a \"Sputnik moment,\" likening it to the Soviet satellite that sparked the space race.
Strategic Fault Lines
The selloff exposed divides over AI's future. Daniel Morgan, senior portfolio manager at Synovus Trust Company, which owns almost a million Nvidia shares, called the reaction overblown, saying DeepSeek's AI model is for use on mobile phones and PCs rather than data centers, where real profits lie.
Yet even optimists acknowledged existential questions. Founded in 2023 by quant trading veteran Liang Wenfeng, DeepSeek has leapfrogged from obscurity to Silicon Valley admiration in months. Its open-source strategy and ability to accumulate chips have made it a case study in China's innovation despite sanctions.
Reference(s):
cgtn.com
