DeepSeek shakes Wall Street, what does Goldman Sachs think? Cautiously bearish on U.S. tech stocks, keep an eye on the giants' earnings reports on Wednesday

This Monday, DeepSeek, which became an overnight sensation overseas, scared Nvidia and burst the AI bubble in the US stock market, shook Wall Street. Goldman Sachs analyst Rich Privorotsky asserted in a report on Monday that DeepSeek will be the hot topic of the week, especially before the earnings reports of the US tech giants known as the "Magnificent Seven" (Mag 7), where the focus will be on AI-related capital expenditures.

Why is DeepSeek so popular? Wall Street Insights previously mentioned that this Chinese AI startup's open-source model has relatively low costs and uses only a small number of chips, yet it has performance comparable to models from American AI giants like OpenAI. Goldman Sachs' Privorotsky believes that DeepSeek's popularity "stems from many complex reasons, fundamentally due to a significant improvement in reasoning efficiency. According to some assessments, it is 40 to 50 times more efficient than other models. If you can do more with fewer resources, then naturally, questions will arise about whether you need so much capacity."

Considering the significant increase in market capitalization related to the DeepSeek theme, Privorotsky believes that the trend in the US stock market was severely reversed last week, with a basket of previously rising stocks turning to decline, and US semiconductor/AI/tech stocks were under pressure on Monday. Goldman Sachs' FICC trader François Theis closely observed the sharp reversal in market sentiment, viewing China as a disruptor in the AI industry, stating that as the Nasdaq fell sharply, the Asian AI ecosystem is in a complete profit mode.

Goldman Sachs emphasized that recent model improvements and cost efficiencies may come from the mixture of experts (MoE) architecture, emphasis on post-training, cost optimization, and reinforcement learning of models. Goldman Sachs' Ronald Keung also highlighted that open-source Chinese models, including DeepSeek R1/V3 and Alibaba's Tongyi Qianwen, have attracted significant interest from developers since their launch due to their transparency and much lower pricing per token compared to global models. In a global environment where the supply of Chinese chips is constrained, many Chinese industry participants have maximized their model output. Not only DeepSeek, but many established Chinese companies also have a large number of AI products.

Goldman Sachs believes that DeepSeek has brought about the following key impacts:

• First, SoftBank's return on investment and capital expenditure efficiency for the "Stargate" AI project remain a major focus (companies related to supply chains and energy demand will benefit from this project).
• Due to the lowered entry barriers, there may be competition between capital-rich internet giants and startups.
• Compared to pre-training, the computational resources required for training to more inference have significantly decreased (this relates to the recent revival of Goldman Sachs' Asian grid stocks)
• Chinese companies have the potential for further global expansion, but considering geopolitical factors, it is clearly necessary to remain vigilant.
• For Chinese cloud/data center companies, Goldman Sachs still believes that the focus in 2025 will be on chip supply conditions, the capability of CSPs (Cloud Service Providers) to increase relevant revenue contributions through AI-driven cloud revenue growth, and how AI workloads and AI-related services, aside from infrastructure/GPU leasing, will promote future growth and profits.
• Beneficiaries: Thanks to its ecosystem, Tencent is in the most advantageous position to launch To-C AI intelligent application apps. The cloud business of internet giants (Alibaba is the largest public cloud hyperscale provider in China) and data centers (GDS, VNET) are expected to benefit from the sustained demand for public cloud and AI computing driven by years of increased AI application rates.

Last year's Goldman Sachs report discussed the debate over the potential for productivity growth in the U.S., with the primary question being whether generative AI has too much investment and too little return.

Goldman Sachs analyst Keita Umetani wrote in the report that based on several conversations he had with investors on Monday morning, the initial thoughts were similar, all believing that DeepSeek's "negative impact on capital expenditures and the significant shift in the narrative of the U.S. AI sector is concerning, whether from the perspective of investment or returns in ROI. If there are no returns, can the investment still be justified?"

The report then pointed out,

"The current clear consensus is to cautiously bearish on U.S. tech stocks until we hear the earnings to be announced this week, including those from Meta and Google, of the mag7."

But Umetani raised a question: Everyone is focused on the earnings reports from the giants this Wednesday, but will you wait until the earnings are announced on Wednesday to conduct the options trading he suggests? For example, the Nasdaq 100 put options NDX 21Feb25 97% SQ Put and the put spread NDX 21Feb25 Put Spread