The blockade of AI computing power has intensified! U.S. plans to limit China's use of Amazon, Microsoft cloud services

Source: Big Data Digest

The Wall Street Journal reported on July 4 that the U.S. government will restrict Chinese companies from using cloud computing services from U.S. manufacturers. As for the purpose, the website speculates: to plug the loopholes in chip control and attack China’s advanced technology.

The chip control mentioned above means that a few months ago, the White House asked Nvidia, AMD, etc. to stop exporting to China (including Hong Kong, China) two top computing chips for artificial intelligence work.

According to a Reuters analysis, the proposed restrictions are seen as a means to close a “significant loophole” in U.S. export control policy to prevent Chinese artificial intelligence companies from using U.S. cloud services to bypass its export control rules.

** In other words, this will be a complete blockade of China’s AI computing power. **

After all, while the U.S. government has restricted chipmakers like Nvidia from selling certain types of chips to China, it hasn’t stopped cloud providers like Microsoft and AWS from offering managed services. At the time, Chinese companies could still train complex AI large models by renting resources.

If the regulations restricting cloud services are officially implemented, cloud service providers such as Amazon and Microsoft in the United States must seek permission from the US government before providing cloud computing services using advanced artificial intelligence chips to Chinese customers.

The news is still being verified. The well-known media Insider said that it is consulting with the White House, the Ministry of Commerce, Nvidia, Microsoft and Amazon, but has not received a reply yet.

Blocking the computing power of artificial intelligence training

The development of AI requires the three horse-drawn chariots of computing power, data, and algorithms to work at the same time. If one of these three factors is missing, the development of AI will be hindered. Especially in the era of large models, algorithms are relatively mature, and AI technologies like Transformer can be obtained from academic papers, and rich data is the advantage of Chinese AI. Therefore, the United States cannot block China’s AI development from the data and algorithm levels.

Computing power is the necessary foundation for building a large-scale model ecology. At present, there is still a gap between China’s ability to independently develop computing power chips and foreign countries.

The core of the cloud is to provide massive computing power through the cloud. The cloud is the natural soil for data processing. Its large-scale parallel and distributed computing capabilities can help reduce the computing cost of AI.

An excellent computing power base has efficiency advantages in the training and reasoning of large models, and the cloud platform is a “bridge” between large models and computing power, which can realize the reasonable allocation of resources for different models and hardware, and achieve the goal of hardware and software. The optimal combination can greatly improve the efficiency of the training model.

From this perspective, owning cloud computing is equivalent to AI computing power, so as to obtain admission tickets for AI large models and generative AI applications.

Source: Zhiding Think Tank

From the above figure, we can see that the increasingly large amount of training has higher and higher requirements for the computing power of cloud services.

If you are still unaware of the role of cloud computing, think about the role of Microsoft as the exclusive cloud service provider in the development of GPT-4 by OpenAI.

The Potential of Cloud Computing in China

Regarding the importance of cloud computing to artificial intelligence, an article in 2019 mentioned:

Innovative technologies such as artificial intelligence have begun to integrate into the field of cloud computing.

In 2017, many domestic cloud vendors, including Tencent Cloud, Alibaba Cloud, Baidu Cloud, Kingsoft Cloud, JD Cloud, etc., successively released their own deep learning platforms.

At the same time, with the breakthrough of big data and artificial intelligence in recent years, the “artificial intelligence + big data + cloud computing” model called “ABC” will play a decisive role in the development of enterprises, industries, the whole country and the world. effect.

At the same time, this article also sorted out the current development status of the cloud computing industry at that time. A few years later, we look back and the analysis at that time is still not outdated today:

The current global cloud computing market is highly concentrated. Since 2016, based on the perspective of cloud business revenue and growth rate, Amazon Cloud (AWS), Microsoft Cloud (Azure), and Alibaba Cloud (Alibaba Cloud) have formed the first camp in the global cloud computing market, and the “3A” of the cloud computing market The pattern is emerging.

Most of the service provider companies with a large share are from the United States. As the “pioneer” of cloud computing, the United States has a clear advantage in the market.

In terms of share alone, it seems that some Chinese manufacturers are not far behind. For example, in 2016, Alibaba Cloud “squeezed out” Google Cloud and successfully entered the top three in the global public cloud market share, which was regarded as a landmark event by the industry.

In addition, at the 2022 Yunqi Conference, Zhang Jianfeng, president of Alibaba Cloud Intelligence, announced that Alibaba Pingtouge’s self-developed cloud-native processor Yitian 710 has been deployed on a large scale, becoming China’s first self-developed CPU that has been deployed on a large scale on the cloud.

It seems that China’s cloud computing has huge potential to be tapped.

Reference source:

1. Guo Chaoxian, Hu Yumeng. Development Situation and Comparison of Chinese and Foreign Cloud Computing Industry [J] .Economics and Management, 2019,33(02):86-92.

Cloud

2. AI large-scale model melee, is China’s cloud service ready? [N] . Computer News, 2023-05-08(011).DOI: 10.28184/n.cnki.ndina.2023.000343.

3. Lv Qian. Hotspots of running water, iron-clad problems, cloud computing card slots, and large-scale model competition [N] . First Financial Daily, 2023-04-13 (A10). DOI: 10.28207/n.cnki.ndycj.2023.001637.

4. Zhang Sainan. Paving the way for large-scale model entrance cloud computing manufacturers set off a new revolution [N] . 21st Century Business Herald, 2023-06-21(009). DOI: 10.28723/n.cnki.nsjbd.2023.002345.