Goldman Sachs deeply interprets DeepSeek's investment logic: AI shifts from hardware to software, impacting "American exceptionalism," overweighting Chinese concept technology stocks

Last week, the domestic large model DeepSeek caused a huge stir in the global technology market, especially dealing a heavy blow to European and American computing power stocks: NVIDIA lost $600 billion in market value overnight, and chip giants like TSMC, AMD, and ASML were all wiped out. Although there was a rebound afterward, this round of sharp decline has prompted global investors to rethink the investment logic of technology stocks, and "American exceptionalism" is facing unprecedented challenges.

Goldman Sachs' latest report pointed out that the rise of DeepSeek signifies that the development of the AI industry is shifting from hardware infrastructure to software applications. This trend not only impacts "American exceptionalism" but also provides new opportunities for diversified development in the global market, especially for Chinese concept technology stocks.

Goldman Sachs noted, the investment logic of AI is shifting from hardware to software, and the high exposure of the MSCI China Index to AI software is expected to enable it to achieve excellent performance in the short term. It is recommended to overweight Chinese concept technology stocks, especially those companies with innovative capabilities and market advantages in the AI application layer. Although the weight of A-shares in hard technology is relatively high, they have also been actively laying out AI software in recent years, and thus will benefit from this.

"American exceptionalism" is challenged, AI investment logic reshaped — from hardware to software

The performance of the U.S. stock market has long outperformed other major global markets, a phenomenon known as "U.S. stock exceptionalism." The main reason behind this is that the U.S. holds a leading position in global economy, technology, and innovation, making its stock market uniquely attractive and performing well over the long term.

One manifestation of this is the significant investment by U.S. companies in research and development and capital expenditures. Goldman Sachs' David J. Kostin team’s latest research shows that the growth investment ratio (GIR) of U.S. companies is significantly higher than that of other regions globally, especially in the AI field. U.S. companies have made substantial investments in AI infrastructure, which has allowed them to dominate the AI hardware and semiconductor sectors.

However, DeepSeek's R1 model achieved performance comparable to leading models like GPT-4 and Llama at a cost of less than $6 million, delivering a significant blow to "American exceptionalism."

From the market perspective, Goldman Sachs' research report found that there was a noticeable rotation in the market last week, with AI hardware/semiconductor stocks facing significant valuation pressure in the short term, while software and service stocks performed strongly.

The Phase 2 basket of stocks focused on AI infrastructure that we track performed poorly, declining 3% compared to an equal-weighted S&P 500 index. Meanwhile, the Phase 3 basket, which includes companies that can increase revenue through AI technology, rose by 4%.

This shift indicates that the investment logic of AI is moving from hardware to software, the Kostin team pointed out:

This evolution indicates that the breadth of the U.S. stock market is expanding, benefiting not only large AI infrastructure companies but also a broader range of companies that can create new revenue sources or enhance productivity through AI technology.

This trend also suggests that the global market is gradually expanding: although AI infrastructure companies are primarily concentrated in the United States, the potential user base is spread across the globe.

The Advantage of China's "Soft Technology" and Overweighting Chinese Concept Technology Stocks

To measure the exposure of the Asian stock market to developments related to artificial intelligence (AI), the Goldman Sachs Timothy Moe team’s latest research report summarizes the exposure of relevant markets to "hard technology" (technology hardware and semiconductors) and "soft technology" (internet, software, and services). The results show that Chinese concept technology stocks have a significant advantage in the AI software sector.

According to Goldman Sachs' analysis, the profit share of soft technology in the MSCI China Index is as high as 37%, with a market capitalization share of 32%. Among them, Chinese concept stocks have greater exposure in applications, while the mainland A-share market has a smaller weight in hard and soft technology, with a more balanced distribution. Goldman Sachs pointed out:

Chinese concept stocks have a broad layout in the AI application layer, covering various fields such as natural language processing, computer vision, and machine learning. For example, internet giants like Tencent, Alibaba, and Baidu have made significant progress in the AI application field. The high exposure of these companies in the AI application layer is expected to lead to better short-term performance compared to markets primarily focused on hardware and semiconductors (such as Taiwan and South Korea).

Mainland A-shares have a more balanced layout in the AI field, covering both hard and soft technology. Although A-shares have a relatively high weight in hard technology, they have also been actively laying out the AI application layer in recent years. Therefore, the A-share market also benefits from the development of the AI application layer.

Taiwan's stock market has a higher exposure to hard technology: accounting for 80% of the MSCI Taiwan Index's market capitalization and 77% of its profits, mainly due to the presence of TSMC.

In other Asian markets, the exposure of soft technology in the MSCI Korea and Japan indices is 6% and 3%, respectively, while Indian software companies are even less noteworthy.

The Goldman Sachs report points out that these differences indicate that as the market prices the uncertainty of demand for AI model training, the MSCI Taiwan and Korea indices may face greater short-term pressure, while Chinese concept stocks may benefit from greater exposure to the application layer.

From the perspective of the positioning changes between hard and soft technology, investors may have partially anticipated this shift, but may not have fully considered that China's soft technology is still at a conservative valuation level.

Goldman Sachs maintains an overweight rating on Chinese concept technology stocks, especially in the internet, media, and entertainment sectors. These sectors not only benefit from advancements in AI technology but also have high valuation attractiveness, currently trading at -0.6 standard deviations of their long-term price-to-earnings ratio, with an expected earnings per share growth rate of 20% for 2025/26. In addition, Goldman Sachs believes that Japan's exposure in the AI field is relatively low, and its market performance may be more influenced by domestic economic and policy factors. Although Japan has some layout in hard technologies (such as semiconductors and hardware), its exposure at the AI application layer is limited, which makes it less affected by the decline in AI hardware demand in the short term