高盛喊出做多中国AI：4万亿美元市值背后，全球基金只配了1.2%

TL;DR

• Goldman Sachs recommends buying the China AI value chain basket, covering power, semiconductors, AI infrastructure, models, and applications.
• Goldman Sachs estimates that China's AI-related market cap is approximately $4 trillion, contributing about 16% of global AI-related revenue, yet global mutual funds have only about 1.2% allocated to China within their tech exposure.
• The core of this trade is not a single AI application breakout, but a re-rating opportunity driven by underweight capital, policy investment, and hardware demand.
• Risks lie in the continued need for delivery on data center investment, memory production expansion, IPO financing, and AI hardware exports.

Goldman Sachs' thematic research team is pushing the "China AI value chain" to the forefront of trade discussions.

According to its report titled "Trading Strategy: Long China AI Value Chain," Goldman Sachs recommends a long position in a China AI basket covering power, semiconductors, AI infrastructure, models, and applications. Over the past two years, global AI trades have been dominated by large US tech stocks, the Nvidia supply chain, and cloud capex; Goldman Sachs now sees an opportunity in the misalignment between the market cap, revenue contribution, and global fund positioning of Chinese AI assets.

According to Goldman Sachs' estimates, Chinese AI-related companies have a combined market cap of approximately $4 trillion, contributing about 16% of global AI-related revenue. However, as of January 2026, global mutual fund managers' allocation to China within their global tech exposure stands at only about 1.2%.

These figures constitute the most important trading logic of the entire report: if the Chinese AI industry already holds a double-digit share of revenue, while global capital allocation remains significantly underweight, then there is room for the China AI value chain to be repriced.

The Greatest Contradiction: Significant Revenue Contribution, Minimal Global Capital Allocation

Goldman Sachs' breakdown of global AI assets provides a stark comparison.

Since the end of 2022, global AI-related stocks have created approximately $34 trillion in market cap, of which China's AI-related market cap is about $4 trillion, representing roughly 10% of the global AI market cap. In terms of revenue, China contributes about 16% of global AI-related revenue.

Capital allocation, however, is far below this proportion. Goldman Sachs estimates that as of January 2026, global mutual fund managers' allocation to China within their global tech exposure is only about 1.2%.

This is the core reason Goldman Sachs proposes going long on the China AI value chain. US AI assets have been repeatedly bought by global capital; Nvidia, cloud vendors, semiconductor equipment, and power infrastructure have all been incorporated into the main AI trade narrative. In contrast, while Chinese AI assets have achieved a certain scale of revenue, they remain underweighted in global fund portfolios.

In other words, Goldman Sachs is not simply betting on a "China AI narrative," but on a more specific funding gap: revenue contribution has already emerged, but global holdings have yet to catch up.

This Is Not a Traditional KWEB Trade; Hardware and Infrastructure Take Precedence

Goldman Sachs specifically emphasizes that this trade is different from the traditional KWEB trade.

KWEB typically corresponds to exposure to Chinese internet platforms, where investors think of e-commerce, advertising, online entertainment, and local services. However, the basket Goldman Sachs has constructed this time is the GS China AI Value Chain (GSXACART), covering everything from power, semiconductors, and AI infrastructure to models and applications, more closely resembling a complete Chinese AI supply chain.

Within this framework, hardware and infrastructure take a more prominent position.

China's push for technological self-reliance and advanced computing capacity means that AI hardware, data centers, power support, and semiconductor segments are simultaneously receiving attention from policy, industry, and capital. Goldman Sachs believes the value of these segments is not yet fully reflected in stock prices.

Its research estimates that the potential economic benefits from AI through efficiency gains and creating new profits could be 50% to 100% higher than what is already priced into current AI stocks. This is why power, AI infrastructure, and semiconductors are placed at the core of the basket.

Whether models and applications can take off ultimately depends on the supply of computing power, storage, electricity, and equipment. These are precisely the areas where China possesses capabilities in large-scale manufacturing, engineering construction, and industrial supporting capacity.

Exports, Policy, and IPO Activity Are Strengthening the AI Hardware Thesis

The evolution of China's AI hardware chain is moving from concept towards more tangible orders, exports, and financing milestones.

On the demand side, customs data cited by multiple media outlets shows China's exports grew 19.4% year-on-year in May, the strongest pace in three months; integrated circuit export value grew approximately 111% year-on-year, while export volume grew only slightly. Behind this price and structural shift, AI hardware demand is seen as a key driver. For storage, semiconductor equipment, and upstream materials, such data points to potential improvements in orders and capacity utilization.

On the policy and investment side, according to a Reuters report citing Bloomberg, China is preparing a five-year plan worth approximately 2 trillion yuan, or about $295 billion, to build a national AI data center network. The plan has yet to be officially announced, but if implemented, it would directly boost demand for domestic memory chips, semiconductor equipment, power support, and data center infrastructure.

On the capital markets side, public reports indicate that A-shares, Hong Kong stocks, and some global indices increased the weighting of AI and semiconductors during adjustments in 2026. This would enhance the passive fund visibility of related companies and steer more domestic and international capital toward advanced computing and semiconductors.

Individual company cases and industry developments are also reinforcing this thesis. YMTC's first-quarter 2026 revenue surged approximately 445% year-on-year, its global NAND flash memory market share rose from 8% a year ago to 13%, jumping to a tie for fourth place, and it is advancing its domestic IPO plans to support capacity expansion.

CXMT is considered a key company in China's DRAM industry. Some third-party research estimates its 2026 revenue could exceed $50 billion; the company's prospectus shows first-quarter revenue of 50.8 billion yuan, with a first-half revenue guidance of 110 to 120 billion yuan.

These cases do not mean Chinese memory companies have fully caught up with overseas giants, but they show that China's AI hardware chain is shifting from a "policy concept" to more observable milestones in revenue, market share, financing, and capacity expansion.

Capital Rotation Is Beginning; US AI Remains the Primary Benchmark

Goldman Sachs also notes that the China AI sector has outperformed other China-related assets and shown signs of capital allocation rotation. However, compared to US AI, the performance of Chinese AI assets still lags significantly.

This is where both the appeal and the risk boundary of the trade lie.

The appeal is that if global investors continue to seek growth lines outside of US AI, the underweighted status of China AI could leave room for capital rotation. Especially after US AI leaders have reached high valuations and their capex expectations are fully discussed, the market will naturally look for supply chain and application assets that have not yet been fully held.

The risk is that this remains a trading recommendation, not a foregone industrial conclusion. The 2 trillion yuan AI data center plan depends on policy details and actual execution; the IPOs, capacity expansion, and profitability improvements of companies like CXMT and YMTC also take time; the sustainability of chip export and sales data depends on the global AI hardware cycle and trade environment.

US AI remains the primary reference for global capital. Whether in model capabilities, cloud vendor capital expenditure, the GPU ecosystem, or enterprise application revenue, the US market boasts more mature benchmarks. For Chinese AI to attract more global capital, it cannot just prove it is "cheap and underweighted"; it must continuously deliver on revenue, profit, and technological progress.

The key takeaway from Goldman Sachs' long China AI value chain trade is not about declaring that China AI has caught up with the US, but about highlighting a market dislocation: approximately $4 trillion in market cap and roughly 16% of global revenue contribution, contrasted with a mere 1.2% allocation to China within global mutual funds' tech exposure.

Whether capital can fill this gap will depend on the continued delivery of policy investment, hardware demand, and corporate earnings.