China’s DeepSeek AI scores important victory against US tech hegemony – Friends of Socialist
DeepSeek, a Chinese AI startup, made headlines this week after its app became the most-downloaded free app in the U.S. on Apple’s App Store. On Monday, global tech stocks dropped sharply as investors reacted to DeepSeek’s rapid rise. Nvidia and Broadcom lost a combined $800 billion in market value.
A new report from SemiAnalysis, a semiconductor research firm, estimated that DeepSeek’s hardware expenses are “well higher than $500 million.” The report noted that DeepSeek’s total costs include research, development, and computing power for training AI models. DeepSeek’s own research paper stated that the official training cost for its latest AI model was $5.576 million. However, this number only includes direct training expenses and excludes prior research and experiments.
DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of the hedge fund High-Flyer. The startup grew from High-Flyer’s AI research unit and now focuses on large language models and artificial general intelligence (AGI). The company remains fully owned and funded by High-Flyer.
Earlier this month, DeepSeek released its latest AI model, R1, an open-source system that rivals OpenAI’s models. Experts praised its capabilities, especially given the restrictions placed on U.S. chip exports to China. Some analysts said this development highlights how the U.S. is falling behind in AI, a market expected to reach $1 trillion in revenue within a decade.
Despite its success, DeepSeek has faced scrutiny. OpenAI’s CEO, Sam Altman, publicly acknowledged DeepSeek’s achievements but suggested that the company may have used OpenAI’s data without permission. DeepSeek has not commented on the allegation.
As DeepSeek continues to grow, tech giants are taking notice. Some experts believe this could change the competitive landscape of AI. Others worry about its impact on existing AI leaders and the global tech industry.
#DeepSeekAI #TechShock #AIGameChanger #OpenAIScandal #ChinaVsUS
