Why China’s plans for artificial intelligence are directly threatened by the US ban on Nvidia chips in the technology conflict

The most potent artificial intelligence systems in China may be affected by US export limits on Nvidia data centre GPUs because they are dependent on such processors.
While carmaker Xpeng claims it has years of reserves, industry sources anticipate businesses to hoard Nvidia chips before the ban takes effect next year.

 

Alibaba Cloud’s Sinian computing platform broke Google’s record by recognising 1.078 million images per second in offline scenarios, and China’s largest server manufacturer Inspur’s NF5488A5 model has been hailed as a world-class product in medical image segmentation, speech recognition, and natural language proficiency. ByteDance’s enterprise cloud service, Volcengine, has reduced the training period for an image-recognition, artificial intelligence (AI) model from 5 days to 3 days
However, Nvidia Corp., the Santa Clara-based GPU juggernaut that has played a crucial role in propelling China’s advances in AI, data analysis, and computing power, is responsible for all of these accomplishments made by China’s most potent technology giants.

There is currently no immediate replacement for the Nvidia GPUs that train AI models for autonomous driving, semantic analysis, image recognition, weather variables, and big data analysis, according to industry insiders and tech analysts. As a result, China’s AI, cloud computing, and smart vehicle sectors have been rattled by the US government’s abrupt decision last month to restrict Nvidia from selling its two most advanced chips, the A100 and the upcoming H100, to clients in China.

 

It is still difficult to determine the entire impact Washington’s most recent ban on semiconductor technology will have on China’s downstream businesses. Nvidia has stated that it is making efforts to lessen the impact on Chinese customers, such as switching out sophisticated GPUs for products that are not required to obtain US licence.

 

Other Chinese GPU manufacturers have voiced anticipation that the restriction will offer them an opportunity to win over former Nvidia customers now that the US-made chips are prohibited.

However, it is easy to see the psychological effects of this focused US action on Chinese investors and authorities.
At a crucial conference this month, President Xi Jinping emphasised that China must accelerate the development of technology essential to national security and establish a top-down “new whole country system” to pursue innovations. Semiconductors or the most recent GPU restrictions imposed by the US were not included in the summary of Xi’s remarks.
He Xiaopeng, the founder of Xpeng, one of China’s largest manufacturers of electric vehicles, commented on his social media accounts that while the Nvidia chip ban would be bad news for cloud-based autonomous driving training, Xpeng would not be significantly impacted because it has already invested in the cutting-edge chips it requires in sufficient quantities for the years to come.

 

Dealers predicted that Chinese customers will compete for Nvidia chips.
The second-largest distributor of telecoms hardware and cloud services in China, VSTECS Holdings, said all of its Chinese customers would begin stockpiling the impacted chips before the ban really takes effect, according to a sales manager stationed in Shanghai. According to Nvidia, the company can keep shipping AI processors from its Hong Kong factory until September 2023.

 

However, some perceive an opportunity to increase efforts in local GPU development as a result of US restrictions on Nvidia graphics processing unit shipments to China. Lau Ka-kuen, as an example.

Alibaba Cloud’s Sinian computing platform broke Google’s record by recognising 1.078 million images per second in offline scenarios, and China’s largest server manufacturer Inspur’s NF5488A5 model has been hailed as a world-class product in medical image segmentation, speech recognition, and natural language proficiency. ByteDance’s enterprise cloud service, Volcengine, has reduced the training period for an image-recognition, artificial intelligence (AI) model from 5 days to 3 days
However, Nvidia Corp., the Santa Clara-based GPU juggernaut that has played a crucial role in propelling China’s advances in AI, data analysis, and computing power, is responsible for all of these accomplishments made by China’s most potent technology giants.

There is currently no immediate replacement for the Nvidia GPUs that train AI models for autonomous driving, semantic analysis, image recognition, weather variables, and big data analysis, according to industry insiders and tech analysts. As a result, China’s AI, cloud computing, and smart vehicle sectors have been rattled by the US government’s abrupt decision last month to restrict Nvidia from selling its two most advanced chips, the A100 and the upcoming H100, to clients in China.

 

Santa Clara, California’s Nvidia headquarters as of May 25, 2022. Image: AFP
It is still difficult to determine the entire impact Washington’s most recent ban on semiconductor technology will have on China’s downstream businesses. Nvidia has stated that it is making efforts to lessen the impact on Chinese customers, such as switching out sophisticated GPUs for products that are not required to obtain US licence.

 

Other Chinese GPU manufacturers have voiced anticipation that the restriction will offer them an opportunity to win over former Nvidia customers now that the US-made chips are prohibited.

However, it is easy to see the psychological effects of this focused US action on Chinese investors and authorities.
At a crucial conference this month, President Xi Jinping emphasised that China must accelerate the development of technology essential to national security and establish a top-down “new whole country system” to pursue innovations. Semiconductors or the most recent GPU restrictions imposed by the US were not included in the summary of Xi’s remarks.
He Xiaopeng, the founder of Xpeng, one of China’s largest manufacturers of electric vehicles, commented on his social media accounts that while the Nvidia chip ban would be bad news for cloud-based autonomous driving training, Xpeng would not be significantly impacted because it has already invested in the cutting-edge chips it requires in sufficient quantities for the years to come.

Chinese attacks on the US Chips Act are increasing, but Beijing has limited defences in place
31 Aug 2022

 

Dealers predicted that Chinese customers will compete for Nvidia chips.
The second-largest distributor of telecoms hardware and cloud services in China, VSTECS Holdings, said all of its Chinese customers would begin stockpiling the impacted chips before the ban really takes effect, according to a sales manager stationed in Shanghai. According to Nvidia, the company can keep shipping AI processors from its Hong Kong factory until September 2023.
SATURDAY
China Tech Insider Newsletter
You agree to receive promotional emails from SCMP by submitting. Tick here if you don’t want these.
You accept our T&C and Privacy Policy by registering.

The manager, who requested anonymity because he was not authorised to communicate with the media, stated that “the majority of cloud service providers in China employ Nvidia chips in their GPU-based solutions.” “GPU-based cloud resources will be in short supply and the impact will be significant,” he continued. “This gear cannot be delivered to China.
The majority of China’s cloud service providers, including Alibaba Cloud (owned by Alibaba Group Holding), Tencent Cloud (owned by Tencent Holdings), and Baidu Cloud, power their cloud infrastructure with CPUs from other manufacturers, such Intel and AMD. Central processing units (CPUs) using Intel’s x86 or Arm-based architecture are used by many cloud servers to handle the computationally intensive tasks.

 

However, the TSMC-made H100, which is anticipated to be ready for sale in the second part of this year, and Nvidia’s A100, which was introduced in 2020, have advantages in managing massive, parallel threads of computing.

Nvidia’s GPUs have thousands more cores than AMD and Intel’s multi-core CPUs, which makes them perfect for deep learning and machine learning AI models.

One of Nvidia’s fastest-growing segments, the data centre business—which includes sales of the two cutting-edge chips—reported US$3.8 billion in sales for the fiscal quarter that ended in July 2022, representing a significant year-over-year gain of 61%.

According to Nvidia’s founder and CEO Jensen Huang, a Taiwanese American businessman who is well-known in China, the nation has grown into a “very significant market” as China transforms into a worldwide AI powerhouse with about one million companies claiming to be involved in AI.

Nvidia’s market capitalization is currently approximately three times that of Intel, in part due to China’s demand for its chips. The maximum third-quarter sales losses from the most recent US government restriction are projected at around US$400 million, according to Nvidia, which accounts for about a quarter of its overall sales in gross terms from China.
On the other hand, if Huawei Technologies Co. is any indication, China’s pain might be far worse.

US trade sanctions on the Chinese telecoms giant have stifled their profitable smartphone industry by cutting off their access to cutting-edge technology. Even while Huawei has long advocated for 5G network equipment, its most recent Mate 50 smartphone models do not even offer 5G features.

 

Although it is technically possible to substitute strong GPUs for some tasks with a mix of less powerful ones, the difference can be significant, and Chinese customers have found the A100’s capabilities to be appealing.

For example, to support GPU-based cloud offerings, Alibaba Cloud and Baidu Cloud both integrated A100 processors into sizable portions of their cloud infrastructure. The South China Morning Post is owned by Alibaba.

Nvidia’s A100 chip has reportedly also been purchased by some of China’s most strategically significant research institutions, including Tsinghua University and the Institute of Computing Technology at the Chinese Academy of Sciences, according to a recent Reuters report based on government tender documents (CAS).

According to Chinese news site Jiemian, Inspur, the nation’s top server manufacturer, told analysts that it is actively conversing with Nvidia about the current scenario. Inspur has highlighted the integration of Nvidia’s A100 GPU in its products.

 

It has reaffirmed the claim made by Nvidia CEO Huang that the US business is seeking for alternatives for Chinese customers. Huang has added, however, that alternatives “may be worse for client workloads that need the full capability of the A100.”

Due to the sensitivity of the subject, an Nvidia manager in China who wished to remain anonymous claimed that Chinese AI firms would end up spending more time training their AI models, which would slow down China’s advancement in AI. Despite having access to enormous amounts of data from its 1.4 billion citizens, China still uses powerful chips to enhance its algorithms.

The latest US prohibition, which builds upon earlier limitations, is anticipated to hurt the industry’s weak spot because domestic GPU manufacturers in China have not yet produced any goods that can compete with Nvidia.

The Economic Observer last week cited Lu Jianping, chief technology officer at Shanghai-based Iluvator, one of the nation’s top GPU manufacturers, as claiming that the company’s GPU still lagged well behind Nvidia’s products. Lu, a former employee of Nvidia and Samsung Electronics, claimed that his company’s goal is to capture 5% of the market, compared to Nvidia’s 95% share.

Some domestic investors, however, are optimistic that the restriction may hasten China’s development of its own GPUs in keeping with Beijing’s frequently cited ambition for self-sufficiency in key technology industries.

 

Biren Technology, a second Chinese GPU startup formed in 2019 by Wall Street veteran Zhang Wen, is now a favourite among investors due to its ambitious plans for the GPU market.

According to PitchBook, it has raised nearly US$633 million in its first 18 months. Last month, it introduced a BR100 GPU that it says can set new records for computational power. The market has not yet tested the product.

The US embargo may give investors new incentives to invest in the GPU market, according to Wang Yu, a manager at Zhuhai-based Hengqin Financial Investment, a fund that focuses on semiconductor and hardware. However, his company still sees certain overinvestment risks.

Biren will have a slim chance of unseating Nvidia. Chinese GPU start-ups cannot compete with Nvidia and AMD, according to Strategy Analytics senior analyst Sravan Kundojjala. Even an experienced GPU competitor like AMD finds it difficult to compete with Nvidia in data centres.

Unified device architecture, a parallel programming methodology that Nvidia started developing in 2006, is one of the company’s primary advantages. According to Kundojjala, it has been widely used by developers as a simple method to accelerate computationally expensive applications.

Therefore, companies who wish to compete with Nvidia must build a complete software ecosystem, which is very challenging.
However, some are prepared to wager on China’s long-term capacity to overcome competitors like Nvidia.
An executive at a Shanghai-based GPU start-up received a call from an existing investor who wanted to invest more money two days after Nvidia announced the impending US government export restriction, according to a source familiar with the case.