Most of China’s biggest tech and internet companies are scampering to get their hands on high-powered Nvidia chips, a crucial component for creating AI and ML systems.
As per a report by the Financial Times, Chinese companies are spending a massive $5 billion on these chips because they’re worried that the US might make it harder for them to buy these chips in the future.
China desperate for the latest AI chips
Baidu, ByteDance, Tencent, and Alibaba have placed orders totalling $1 billion to buy around 100,000 A800 processors from Nvidia. These chips are expected to be delivered this year. Additionally, these Chinese companies have also bought another $4 billion worth of graphics processing units from Nvidia, which will be delivered in 2024.
The A800 chips are a bit weaker compared to Nvidia’s most advanced A100 chips, which are used in data centres. This is because of rules set by the US government last year, which restrict what Chinese tech companies can buy. Due to these restrictions, Chinese companies can only buy A800 chips, which have slower data transfer rates than the A100.
As interest in AI has grown over the past year, Nvidia’s chips have become highly sought after by major tech companies across the world.
The new gold rush
Chinese internet companies are rushing to collect as many A800 chips as they can because they’re afraid that the US might impose even more rules on them. They’re also worried about a shortage of these chips because so many companies want them.
One employee from Baidu mentioned, “We can’t train large language models without these Nvidia chips.” The companies are working on creating their own AI models, inspired by the success of ChatGPT, which is a chatbot developed by OpenAI.
Investing immensely in computing power
ByteDance, the company behind TikTok, has many teams working on different AI projects. They’ve even tested a feature for TikTok that uses AI to generate content. Two employees with inside knowledge told Financial Times that ByteDance has already gathered over 10,000 Nvidia GPUs and ordered almost 70,000 A800 chips for next year, worth about $700 million.
Similarly, Alibaba plans to use these chips to power their large language model, which will be used across their products like online shopping and mapping tools. Meanwhile, Baidu is working on their own project similar to ChatGPT, called Ernie Bot.
Nvidia has stated that internet companies often invest a lot of money in data centre components, placing orders many months in advance.
In April, Tencent Cloud introduced a new group of servers that people can rent. These servers use a special kind of Nvidia chip called the H800 GPU. This chip is based on Nvidia’s latest H100 model, but it’s modified for use in China. It’s really good for tasks like training big language models, self-driving cars, and scientific calculations.
Alibaba Cloud has also gotten lots of H800 chips from Nvidia. Many customers have already contacted Alibaba to get cloud services that use these chips. They want to use these chips to create services similar to ChatGPT in China.
ByteDance, on the other hand, has been using both A800 and A100 Nvidia chips for its cloud computing devices. They’ve saved up a lot of these chips. They even launched a platform in June where companies can try out different services that use large language models.
Earlier this year, most big Chinese internet companies had very few of these chips for training large language models. But as demand increased, the price of these chips went up by more than 50 per cent, according to one Nvidia distributor.
from Firstpost Tech Latest News https://ift.tt/zunOMe1
No comments:
Post a Comment
please do not enter any spam link in the comment box.