China will build the world’s largest AI network with its own chips, although they are not enough
The Chinese Government is launching a project that seeks to invest $295 billion in five years in the deployment of a national network of data centers of artificial intelligence (AI). In the current scenario of confrontation in the technological field between the US and China, this plan is not surprising at all. However, we cannot ignore the fine print: at least 80% of the underlying technology, including AI chips, must come from domestic suppliers, such as Huawei or Cambricon. Early October 2024 the Administration sent Chinese AI companies were given a recommendation asking them to use chips produced in China as much as possible. Ten months later, this recommendation became a requirement. The Chinese government forced state-owned data centers across the country to use at least 50% Chinese integrated circuits in their servers. Now, as we have just seen, this figure rises to 80%. Be that as it may, this scenario clearly favors three companies: Huawei, Cambricon and Moore Threads. Nvidia and AMD no longer count for China Huawei invests more than 25 billion dollars annually in developing your hardware for AIso presumably it will not take long to match the performance of the GPUs produced by Nvidia or AMD. However, this company faces an enormous challenge that will probably prevent it from meeting the demand for AI chips from the Chinese market in the short term. In mid-June 2025 Jeffrey Kessler, the Under Secretary of Commerce for Industry and Security at the US Department of Commerce, made this statement in Congress: “Our assessment indicates that Huawei’s Ascend chip production capacity by 2025 will be 200,000 units or less, and we anticipate that most or all of that production will be delivered to companies within China.” This limitation has a specific technical basis: yield per wafer The integration technologies used by the Chinese semiconductor manufacturer SMIC to produce Huawei or Cambricon chips have a very wide room for improvement. The per-wafer performance of SMIC integration technologies has a very large room for improvement SMIC already has the capacity to manufacture 6nm integrated circuits, and will soon also be able to produce 5nm semiconductors, but is limited by the performance of the deep ultraviolet lithography equipment (UVP) that you have in your possession. It is commendable that SMIC and Huawei engineers have managed to refine their integrated circuit manufacturing processes enough to be able to produce 5, 6 and 7 nm chips with ASML’s UVP equipment, but a priori it is very unlikely that with these machines they will be able to go beyond 3 nm. And it is because the technique of multiple patterningwhich is what they are using, imposes important limitations. A note: this strategy broadly consists of transferring the pattern to the wafer in several passes with the purpose of increase resolution of the lithographic process. Its problem is that it usually has an upward impact on the cost of chips and a downward impact on production capacity. For Huawei, it is a big problem not to have the necessary technology to produce cutting-edge semiconductors comparable to those manufactured by Intel, TSMC or Samsung, which is why it is working on the development of its own equipment. extreme ultraviolet photolithography (EUV). This scenario has caused the Chinese semiconductor industry to question whether national hardware can maintain the pace demanded by the Government. SMIC co-CEO Zhao Haijun has warned that this hasty addition of capacity can leave data centers idlecomparing this situation to building highways before there is traffic. In addition, several Chinese executives in the chip sector have recognized separately that his country is five to ten years behind in the cutting edge in silicon for AI data centers. In the current situation, the decision has its logic. Another thing is whether China can comply with it. Image | Cambricon Technologies More information | Tom’s Hardware In Xataka | NVIDIA has to deal with the absolute distrust of several US legislators. Your plan in China is in danger In Xataka | The US wants to end Chinese AI chips sold abroad. And China knows how to defend itself