While the latest Deepseek model was launched with much less fanfare, compared to last January’s first launch, its partial use of Chinese domestic chips for model training has signalled China’s advancement in the AI supply chains self-sufficiency drive. Mali Chivakul, emerging markets economist at J. Safra Sarasin Sustainable Asset Management, shares her insights.
Despite technological constraints, Chinese AI models have become better and account for a higher share of global top models. While Chinese investment in data centres is small, compared to in the US, its production of hardware for domestic and global data centre build-out has been significant. The localisation of the semiconductor supply chains, supported by government policy, has benefited domestic AI hardware manufacturers, lifting their revenues significantly. The trend is likely to continue throughout the year. Solid growth of AI-related exports of services is also expected.
China has caught up in advanced chips production
It has been more than a year since Deepseek’s first model, R1, was launched in January 2025. The company has recently launched another major version of its model, Deepseek V4, with much less fanfare. Still, both versions of the V4 models are currently considered among the top models in the world. More importantly, the release suggests that Deepseek V4 was partly trained on Huawei’s AI chips, unlike its R1 model which relied entirely on Nvidia’s chips. While Chinese domestic chips remain inferior to American chips, the US’ export controls seem to have accelerated China’s determination in advanced chips self-sufficiency.
Chinese AI models are deployed more frequently globally
Beyond building world-class models, China has taken a different approach to AI compared to the US, focusing on efficiency, wide adoption and physical integration. Constraints on advanced chips have forced Chinese AI companies to be efficient. The strategy of providing open-source models has led to higher global adoption of Chinese models than American models which are closed-source and require paid subscriptions. Driven by their low costs and efficiency, Chinese models are deployed more globally than their American counterparts, according to token usage. Progress in physical integration, the so-called “embodied AI,” has been significant in industrial applications and robotics, aided by data.
Steady investment in electricity supply
Can we see the AI boom in China’s macro data? AI capex investment in the US has concentrated in data centre construction. Due to its technological constraints and strategic focus, Chinese investment in data centres this year is only expected at less than 15% of the US’ $765bn. On energy, fixed assets investment in electricity grew 16% on average between 2000 and 2025, helping to offset the drag from real estate. Year-to-date, it has grown 9%.
Integrated circuit production boosts industrial production
While data centre investment is not large compared to the US, China’s production of hardware for global data centre construction has contributed robustly to its industrial production. While China continues to catch up on advanced chips technology, it is already a major global supplier of mature-node integrated circuits, which are also required in global AI capex build-out. Rising production has supported robust exports, mainly to ASEAN countries to support data centre investment there. At the same time, imports of more advanced chips especially from Korea and Taiwan have also accelerated.
Semiconductor supply chain localisation lifts domestic AI hardware companies
How has the development impacted the market? Chinese manufacturers of semiconductors and related goods have benefited from the AI boom. Localisation of AI hardware amidst strong domestic demand, supported by the government’s AI adoption target of 70% by 2027, has given a strong boost to a number of companies. Huawei, for example, expects revenue from AI chips to jump at least 60% this year. The STAR 50 Index, which consists of 50 largest technology-focused stocks on the Shanghai Stock Market has risen 17% year to date, compared to CSI 300 Index’s 4% return. It is dominated by AI hardware firms such as Cambricon Technologies and Montage Technology that have seen their earnings increasing significantly in Q1. It is important to note that these companies are in earlier stage compared to Hong Kong-listed internet platform companies. The STAR 50 Index’s market capitalisation is less than half of the Hang Seng Tech Index.
AI hardware manufacturers to benefit from the trend
Given the strategic importance of the hardware industry, we expect the trend to continue this year. Strong global demand should continue to drive the performance of legacy chips producers. At the same time, domestic demand for AI integration should remain strong, supporting high-end hardware firms where “involution” should not be an issue in this segment. At the macro level, investment in advanced manufacturing and exports of AI hardware goods will continue to support growth this year.
Strong growth of exports of AI-related services
AI-related business has also supported exports of services. Exports of telecommunication, computer and information services have been growing at double digits in the last 12-18 months and have reached $75 billion, almost 20% of total services exports. AI model deployment, data storage and analytics, gaming and digital content are important drivers. Platform companies are the main players in this segment. While this remains much smaller compared to chips exports ($230 billion), strong growth is expected in the coming quarters.
