China’s growing pipeline of artificial intelligence data centre projects highlights a significant weakness in its technology development strategy. Despite years of efforts to reduce dependence on American semiconductor technology, a substantial number of Chinese computing projects continue to be designed around Nvidia processors. The scale of these commitments demonstrates the difficulty of replacing established foreign technology once infrastructure investment has already been planned around it.
A review of more than 150 procurement notices, investment plans, construction contracts and corporate disclosures identified Chinese AI computing projects worth over 23 billion yuan, equivalent to approximately $3.2 billion. The projects involve restricted Nvidia processors, including the H100, H200 and H800 models, and cover different regions, institutions and stages of development.
The total value includes planned and signed projects, infrastructure and equipment commitments. It should not be interpreted as the amount spent exclusively on Nvidia chips or as confirmation that all the facilities have received their intended processors.
Nevertheless, the figures reveal the extent to which Nvidia technology remains embedded in China’s AI expansion, even as American export restrictions have made access increasingly difficult.
Regional Data Centres Have Created Long-Term Commitments
China’s AI infrastructure expansion is not restricted to its largest technology centres. Regional authorities and local companies have also supported computing projects in an effort to attract investment, strengthen industrial development and establish new technology businesses.
These projects can serve several purposes. Data centres provide computing resources for research institutions, technology companies and industrial users. They can also become part of wider regional development strategies involving cloud computing, digital services and AI applications.
However, the decision to build infrastructure around specific processors creates long-term dependencies. The buildings, power systems, cooling arrangements and networking equipment must be compatible with the computing hardware selected for the facility.
If access to those processors becomes restricted, project developers may have to reconsider their original plans. This could involve replacing equipment, modifying software systems or delaying operations until alternative hardware becomes available.
For regional governments, the consequences could extend beyond individual companies. Delayed facilities may affect investment expectations, employment plans and the commercial value of supporting infrastructure.
The H100 and H200 Restrictions Create a Difficult Transition
Nvidia’s advanced processors have become important components of AI computing because of their performance and the software ecosystem supporting them. The H100 and H200 models are designed for demanding computing workloads, while the H800 was developed in the context of earlier US restrictions.
American export controls have sought to restrict China’s access to advanced computing capabilities. These measures have complicated the procurement of processors required by Chinese AI developers.
The difficulty for China is that semiconductor substitution is not simply a matter of purchasing a different chip. AI infrastructure depends on compatibility between hardware, software, networking and applications.
A company moving from one processor architecture to another may need to modify its software, test performance and train technical staff. The costs can become particularly significant when a facility contains large numbers of interconnected processors.
Chinese manufacturers have been developing domestic alternatives, but the challenge is to deliver systems that can compete across performance, availability, software support and operating costs.
The transition may therefore take place gradually rather than through an immediate replacement of all foreign technology.
A Pipeline Does Not Mean All Projects Will Be Completed
One important distinction in assessing China’s AI infrastructure expansion is the difference between planned investment and operational capacity.
A project listed in a procurement notice or investment document does not necessarily indicate that construction has been completed or that the required processors have been delivered. Some projects may remain at the planning stage, while others may be under construction or already operating.
This distinction becomes especially important when evaluating the potential financial impact of export restrictions. A project that has not yet committed substantial resources can be postponed more easily than one that has completed construction and entered into long-term operational agreements.
The reported pipeline should therefore be treated as an indication of demand and planned investment rather than a confirmed inventory of functioning Nvidia-powered facilities.
For investors and policymakers, the status of individual projects will matter more than the headline value of the overall pipeline.
Domestic Alternatives Face a Scale Challenge
China has made the development of domestic semiconductor technology a strategic priority. Its manufacturers are working to provide alternatives for AI training and inference, reducing the country’s dependence on foreign suppliers.
However, developing a competitive semiconductor ecosystem requires more than manufacturing processors. It involves advanced production capabilities, software development, semiconductor design, networking and a large community of developers.
Nvidia’s commercial position has been strengthened by its established software environment and relationships with technology companies. Chinese developers seeking to migrate their applications may have to undertake considerable technical work before domestic alternatives can deliver comparable results.
The country’s large domestic market could help support this transition. Government-backed procurement and investment may provide local manufacturers with opportunities to improve their products and expand production.
But there is a difference between encouraging domestic adoption and establishing an economically competitive technology ecosystem. The latter requires sustained improvements in performance, reliability and manufacturing capacity.
Geopolitical Decisions Will Influence Commercial Viability
The future of China’s Nvidia-based data centre projects will depend partly on decisions made in Washington and on the availability of alternative technologies within China.
Export restrictions can change the economics of projects even when companies have already secured funding and completed construction. They may also encourage developers to postpone investment until the regulatory environment becomes clearer.
Nvidia’s leadership transition adds another dimension to the situation. The departure of Jensen Huang, who had publicly supported continued access to the Chinese market, may influence the company’s approach to international commercial relationships, although the immediate effect of his resignation on export policy remains uncertain.
For Chinese project developers, the central challenge is reducing exposure to external supply decisions without sacrificing the computing performance needed to compete globally.
The reported $3.2 billion pipeline illustrates how deeply the current infrastructure expansion remains connected to Nvidia technology. It also shows why China’s transition towards domestic AI hardware is likely to be a prolonged industrial and technological process rather than a quick policy-driven change.
(Spource:www.livemint.com)
