Reported that in order to cope with the artificial intelligence competition, a number of Chinese Internet giants are scrambling to order chips from the global chip giant Nvidia。These Internet companies have booked nearly $5 billion worth of chip orders this year and next.。

Baidu, ByteDance, Tencent and Alibaba have placed orders worth $1 billion from Nvidia for about 100,000 A800 chips to be delivered this year, according to people familiar with the matter。In addition, $4 billion worth of chips have been ordered for delivery next year.。

The A800 is a version of the chip launched by Nvidia for Chinese companies under U.S. government chip export controls, and is a "low-profile version" of the A100 chip.。The A100 chip is a cutting-edge circular processor (GPU) used in the data center, and its data transfer speed is much faster than the A800。

Since the beginning of this year, as the competition for artificial intelligence has intensified, no matter which direction artificial intelligence develops, chips are needed to lay the "foundation."。At present, NVIDIA's GPU has become the world's largest technology companies in the most popular commodity, it can provide computing power for the development of large models。

Now there are a number of companies in China are developing their own large models, due to concerns that the U.S. chip control will be upgraded to lead to A800 supply cuts, China's Internet customer companies have no choice but to compete to hoard A800 chips, because no one wants to fall behind in the artificial intelligence race。

Baidu is the first company in China to publish a self-developed model.。Baidu's big model "Wen Xin Yi Yan" is rapidly iterative update, must have a chip to support its model training。A Baidu employee said, "If we lack chips, we can't train any large language models."。"

According to people familiar with the matter, TikTok's parent company ByteDance now has a number of small R & D teams dedicated to developing a variety of generative artificial intelligence products, including an artificial intelligence chatbot code-named Grace, which is currently undergoing internal testing.。In May, ByteDance tested an AI chatbot called Tako on TikTok, which can recommend videos based on people's requests.。TikTok also recently filed a trademark application for Tako, indicating that the company may be preparing to promote the chatbot.。

Employees with ByteDance say the company has stockpiled at least 10,000 NVIDIA GPUs to support the development of its AI-related products.。The company also ordered nearly 70,000 A800 chips, to be delivered next year, worth about $700 million, they added.。

Alibaba, for its part, is planning to plug all of its products into its big models, including Taobao, DingTalk and Gaud Maps, among others.。

In addition to A100, Nvidia has a high-performance chip H100 is also a hot product。Nvidia's A / H100 GPU chip currently has great advantages in AI training and running machine learning, and its chip is also a key underlying hardware to support ChatGPT。Like the A100, the H100 also has its Chinese market version, the H800, which can provide computing power for large language model training, autonomous driving and scientific computing.。

Tencent Cloud released a new generation of HCC high-performance computing cluster in April, which is equipped with Nvidia H800 GPU。

Tencent Cloud's "old rival," Aliyun, also bought thousands of H800 chips from Nvidia.。It is reported that many customers of Alibaba Cloud have contacted the company for cloud services supported by these chips to help them complete the development of generative artificial intelligence.。

It is understood that at a time when artificial intelligence "detonates" the world, most Chinese Internet giants have less than a few thousand chips in stock that can be used to train large models.。And as demand grows, so does the cost of buying chips.。A dealer of Nvidia revealed: "Now the price of A800 chips in the hands of dealers has risen by more than 50%.。"

This strong demand also brings Nvidia chip "shortage"。In the first quarter of this year, supply and demand in the chip market were fairly balanced, but from early April, the market became unusually tight。Not only are cloud service providers and technology companies looking for chips, but many AI labs are also ordering chips, and Nvidia is the chip company with the most orders。

In a statement, Nvidia said: "Consumer internet companies and cloud service providers invest billions of dollars each year in data center components, often placing orders months in advance.。It is understood that Nvidia's best AI orders have been placed after 2024.。

For the global craze for Nvidia chips, CoreWeave co-founder and chief technology officer Brian Venturo (Brian Venturo) believes that this is because Nvidia's advantages in chip hardware and software constitute a double "moat."。He thinks no company is better at making chips than Nvidia.。And for startups, if they are asked to re-adjust the technology stack (tech stack) to use AMD or TPU (Google's Tensor processing unit), it may waste a lot of time, causing them to miss a good time to go public.。

He also believes that Nvidia's significant investment in the CUDA (software programming platform) ecosystem has put the company essentially 10 years ahead of its peers.。"It's not just 10 years for Nvidia, it's 10 years for their customers and developers to build on that ecosystem, and I don't think any other enterprise will be able to surpass Nvidia in the short or even medium term," he said.。