China’s racing to beat U.S. chip curbs. How its supply chain stacks up

With the US that China reduces to buy advanced semiconductors who are used in the development of artificial intelligence, Beijing lays hope for domestic alternatives such as Huawei.

The task has been made more challenging by the fact that US CURBS not only hinders the access from China to the world’s most advanced chips, but also limiting that the use of technology is of vital importance for creating an AI chip -eco system.

These limitations include the entire semiconductor value -chain, ranging from design and production equipment that is used to produce AI chips into supporting elements such as memory chips.

Beijing you have mobilized dozens of billions of dollars to try to fill those gaps, but although it is to “brutal” in Sum Breaking Throughs, you still have a long way to go, according to experts.

“The US export checks on advanced NVIDIA AI chips have the Chinese industry to develop alternatives, while it is also more difficult for domestic companies to do this,” said Paul Triolo, partner and senior vice-president for China at DGA-Albright Stonebridge Group consultancy.

This is how China comes up with the rest of the world in four important segments that are needed to build AI chips.

NVIDIA is gifted as the world’s leading AI chip company, but is important to understand that today it does not produce the physical chips that are used for AI training and computing.

On the contrary, the company designs AI chips, or more precise graphic processing units. Orders of the patented GPU designs from the company are the sent to Chip Founding – manufacturers who specialize in the mass production of the semiconductor products of other companies.

While American competitors such as AMD and Broadcom offer different alternatives, GPU designs from Nvidia are collected on a large scale as the industry stand. The demand for Nvidia chips is so strong that Chinese customers have continued to buy one of the company’s chips where they can get their handles.

Butidia is struggling with the tightening restrictions of Washington. The company that was assessed in April had tapered to sell its H20 processor to Chinese customers.

Nvidia’s H20 was a less advanced version of his H100 processor, specially designed to bypass earlier export controls. Nevertheless, experts say, it was still more advanced than everything inland was available. But China hopes to change that.

In response to limitations, more Chinese semiconductor players have entered the AI ​​processorena. They have recorded a wide range of starters, such as Enflame technology and Biren technology, who want to absorb billions of dollars in the GPU requirement of Nvidia.

But no Chinese company seems closer to offering a real alternative to Nvidia than Huawei’s chip design arm, Hysilicon.

Huawei’s most advanced GPU in mass production is the Ascend 910B. The next generation ASCend 910C was reported to start massive shipments in May, although no updates have been created.

Dylan Patel, founder, CEO and chief analyst at Semi -Saysis, told CNBC that while the Ascend chips remain behind Nvidia, they show that Huawei has made considerable progress.

“Comb with Nvidia’s export-registered chips, the performance gap between Huawei and the H20 is less than a full generation. Huawei is not far behind the products that Nvidia is allowed to sell in China,” said Patel.

I added that the 910B ran two years behind Nvidia from last year, while the Ascend 910c lags only a year.

But although that suggests that the Chinese GPU design options have made the great progress, design is just one aspect that stands in the way of creating a competitive AI chip -eco system.

To produce his GPUs, Nvidia trusts TSMC, the world’s largest contract chip, which produces the most advanced chips in the world.

TSMC meets the American chip checks and is also excluded from taking chip orders from companies on the black list of American trade. Huawei was placed on the list in 2019.

That has led to the Chinese chip design such as Huawei for the Lost Chip Founding List -List -List, of which the most larest smic is.

Smic is far behind TSMC-IT’s Offorfial known as an aable to produce chips of 7 nanometers, for which less advanced technology is required than the production of 3-nanometer from TSMC. Smaller nanometer dimensions lead to larger chip processing and efflicy.

There are signs that Smic has made your progress. The company is suspected that it lags behind 5-nanometer 5G chip for Huawei’s Mate 60 Pro, which had founded confidence in US chip controls in 2023. However, the company has a long way to go before it can produce advanced GPUs in a cost -efficient way.

According to independent chip and technology analyst Ray Wang, the well-known operational task of Smic is overshadowed by TSMCs.

“Huawei is a good chip design company, but they are still without good domestic chip makers,” Wang said, and noted that Huawei is reported that the reported on his own manufacturing options.

But the lack of important production equipment stands in the way of both companies.

The ability of Smic to meet Huawei’s GPU requirements is limited by the family problem of export controls, but in this case from the Netherlands.

Although the Netherlands may not have prominent semiconductor designers or manufacturers, it is home to the home ASMLThe world’s leading supplier of advanced chipping equipment – machines that use light or electrical rays to transfer complex patterns to silicon waffles, creating the base of microchips.

In accordance with the American export controls, the country has agreed to block the sale of ASML’s most advanced Ultraviolet (EUV) bodily machines. The tools are crucial to make advanced GPUs to scale and cost effective.

According to Jeff Koch, an analyst at Semianalysis, EUV is the most signed barrier for Chinese advanced chip production. “They have the most other tooling available, but lithography limits their ability to scale up to 3nm and under process junctions,” I have AWD CNBC.

Smic has found methods to work on Lichography restrictions using ASML’s less advanced deep ultraviolet lithography systems, which have relatively relatively relatively fewer limitations.

Because of this “brute force” is the production of chips at 7 Nm, but the yields are not good, and the strategy is likely that it is limit, Koch said, adding that “with the current yields it seems as if Smic cannot produce enough household accelerators to meet the demand.”

Sicarrier Technologies, a Chinese company that works on lithography technology, you reported that you are linked to Huawei.

But imitating existing lithraphy tools can take years, if not decades, to reach, Koch said. Insentad, China will probably pursue other technologies and different lithography techniques to push innovation ratho than imitation, he added.

Although GPUs are often identified as the most critical componts in AI comport, these far from the only one. To operate AI training and computer, GPUs must collaborate with memory chips, which are Uble to store data in a wider “chipset”.

In AI applications, a specific type of memory that HBM is known has become the industrial stand. South Korea’s SK Hynix Has taken industrial management in HBM. Other companies in the field are located, among others, Samsung and the US Micron.

“High bandwidth memory in this phase of AI progression has Séntial for training and implementing AI models,” said Analyst Wang.

Just like with the Netherlands, South Korea works with chip restrictions guided by the US and started combining fresh curbs in December with saving certain HBM memory chips to China.

In response to this is the Chinese memory chip manufacturer Changxin Memory Technologies, or CXMT, in collaboration with Chip packaging and test company Tongfu Micro-Electronics, in the early stages of producing HBM, according to a Reuters report.

According to Wang, CXMT expect three to four years behind the global leaders in HBM development, although it is confronted with large roadbliners, including export controls on chipping equipment.

Semianysis Estate in Prinil that CXMT stayed away from the disasters of a reasonable volume for a year.

Chinese foundry Wuhan Xinxin Semiconductor Manufacturing is reported that building a factory produces HBM waffles. A report from SCMP said that Huawei Technologies collaborated with the company in producing HBM chips, although the companies did not confirm the partnership.

Huawei You have read on HBM stocks of suppliers such as Samsung for use in their ASCend 910C AI -Processor, Semianysis said in a report in April, and noted that although the chip was designed in its own country, it still reads on foreign products that were obtained before or in the eastern restrictions.

“Whether it is HBM from Samsung, waffles from TSMC or equipment from America, the Netherlands and Japan, there is a great dependence on foreign industry,” Semianalysis said.