Recently, China’s foreign trade has shown an eye-catching new trend. AI-related products have become the main force in export growth, and domestic large-scale models have emerged in overseas markets.
The US dollar-denominated foreign trade data for May released by the General Administration of Customs on June 9 showed that exports increased by 19.4% year-on-year, significantly exceeding market expectations.
Exports of AI-related products such as integrated circuits and computer hardware hit multi-year highs.
Export growth in May was 19.4%, a new high in the past three months, import growth was 27.4%, and the trade surplus for the month was US$105.4 billion.
The main driver of growth is no longer traditional processed products such as clothing, shoes and hats. The entire AI hardware industry chain supports the increase in foreign trade.
Customs breakdown data clearly reflects the industrial transformation. The export of computers and supporting parts increased by 66%, automatic data processing equipment increased by 66.1%, the overall export of high-tech products increased by 50.9%, and the export of integrated circuits doubled year-on-year, setting a new record after 2013.
Memory chip prices continue to rise, industry organizations judge that global demand for AI construction has not peaked, and chips have completely changed the foreign trade structure.
Orders for leading optical module companies are full, and South Korea’s Samsung and China’s InnoLight are reaping the dividends of global computing power construction at the same time.
Overseas customers have locked in hardware orders in advance due to geopolitical risk considerations, further pushing up the scale of exports.
Sub-regional data are full of highlights. Exports to the United States surged 35.4% in May, ending the previous long-term decline.
Analysts compare manufacturing bases in Southeast Asia and find that the tariff disadvantage of local products continues to narrow. Even if the United States adjusts tariffs, domestic manufacturing still maintains stable competitiveness.
The import side exploded simultaneously. In the first five months, the total import volume was 8.77 trillion yuan, with a growth rate of 20.5% outperforming exports.
Imports of high-tech products increased by 47% in May, South Korea’s semiconductor exports to China more than doubled, and the domestic AI industry chain continues to absorb overseas high-end chips.
Foreign trade shows a clear trend of differentiation. Chips and computer hardware contribute half of the export increase. The growth of traditional labor-intensive products has almost stagnated. K-shaped differentiation has become the new normal of foreign trade.
The conflict in the Middle East has driven the export of supporting products for energy storage, photovoltaics, and electric vehicles. The high value-added attributes of high-tech products have also made foreign trade’s ability to withstand exchange rate fluctuations continue to increase.
Exports have remained stable during the RMB appreciation stage, and the core competitiveness of the manufacturing industry is clear at a glance.
Huagong Technology’s overseas orders for optical modules are scheduled until 2027, which is the most intuitive example of the booming overseas demand for AI hardware.
On June 3, the U.S. corporate finance and tax platform Ramp released a list. The domestic large model DeepSeek topped the list of paid procurement software for U.S. companies. A large number of U.S. companies directly paid to use domestic AI services.
The chief economist of the platform bluntly said that it exceeded expectations. In previous years, overseas companies used domestic AI, and most of them downloaded open source code and deployed it locally, without generating any payment.
This round of fundamental changes has occurred. A large number of American companies directly pay to call APIs, and business data is sent to and from domestic servers. This is a strong signal released by the market.
The pricing of large models in the United States has remained high for a long time. Bain & Company research shows that the cumulative investment in AI by large global enterprises has exceeded one trillion US dollars, and cost recovery is far less than expected.
Many companies have begun to compress their AI budgets. Uber has used up its annual token quota early. Salesforce pays Anthropic a service fee of US$300 million every year. Amazon has suspended internal AI consumption rankings to prevent employees from uncontrolled consumption of computing power. Cost control has been tightened across the industry.
DeepSeek launched a permanent price reduction plan in May. The V4 Pro API price dropped to a quarter of the original price, and the comprehensive call cost was only about one-tenth of the mainstream overseas model.
The founder previously proposed the idea of inclusive AI, and the affordable route has now successfully impressed overseas companies.
Data from the third-party platform Open Router proves the popularity. The DeepSeek series model token calls in a single week were 5.74 trillion, surpassing Anthropic and Google for two consecutive weeks. The domestic model camp’s total call volume in a single week was 9.22 trillion, continuing to crush the US manufacturers’ 4.93 trillion. Platform users are mainly overseas developers, and overseas practitioners vote with their feet.
Many U.S. policies have directed technological isolation in an attempt to distance the industrial chains between the two sides, but companies at the market level have made exactly the opposite choice.
The construction of global computing power on the hardware side is a rigid demand, and tariff restrictions are difficult to suppress the demand for chip and server procurement. Domestic AI hardware exports continue to rise, while overseas high-end chips continue to be imported, forming a deeply inter-embedded industrial chain.
The high cost of using local AI on the software side has forced American companies to look for affordable alternatives. DeepSeek’s rise to the top of the list is only a phased result.
Amazon Cloud and Microsoft Azure took the initiative to put the DeepSeek model on the shelves and opened the call channel to global enterprises. The two major American technology giants took the initiative to connect with the domestic AI ecosystem, which later confirmed that the products have irreplaceable cost-effective advantages.
Industry organizations have also objectively warned of risks. Domestic models entering the U.S. corporate system will inevitably face regulatory challenges, and data compliance-related reviews will not be absent.
Even if some companies suspend use in the short term, the budget pressure caused by high computing power costs will not disappear. Finding parity models is a long-term industry trend.
Reaching the top of the list is only a short-term market phenomenon and cannot be directly equated to an overall victory for the industry. Potential obstacles need to be viewed objectively.
Regulatory reviews at the geopolitical level may be implemented at any time. Overseas politicians can easily use data security issues to restrict domestic AI products, and there is uncertainty in short-term overseas expansion.
Competition in the industry continues to intensify, and overseas manufacturers have begun to lower model pricing to proactively respond to the impact of domestic models.
The real core competitiveness does not lie in short-term low prices, but in supporting a complete industrial chain.
The country has the full chain capabilities of AI hardware production, algorithm research and development, and large-scale implementation. Hardware exports continue to increase, and software goes overseas simultaneously. Two-way collaboration between software and hardware is the long-term confidence.
Foreign trade data has proven that technology-driven industrial upgrading can hedge against external policy interference.
AI software and hardware are moving globally at the same time, continuing to deepen the industrial dependence of various countries. It is difficult to achieve long-term implementation simply by relying on barriers to isolate the market.
At the enterprise level, we cannot stop at low-price competition. We must continue to iterate technology and improve overseas compliance programs to transform the current window of demand into a stable and sustainable global market share.
