TikTok parent ByteDance plans to spend $7B on cloud-based GPUs this year to fuel its AI ambitions

TikTok’s parent company Bitbeat Ltd. The company reportedly plans to circumvent strict U.S. restrictions on the export of advanced computer chips to China.

The company reportedly plans to spend $7 billion to purchase Nvidia’s most powerful graphics processing units to drive the development of cutting-edge artificial intelligence models. It is not seeking to purchase any chips, but simply rents them through data centers located outside mainland China.

ByteDance’s plans reported informationwhich cited a number of anonymous sources familiar with the matter. Sources said that ByteDance founder Zhang Yiming is personally negotiating with data center operators in Southeast Asia and the Middle East in an attempt to obtain Nvidia’s data center access rights. Next generation Blackwell GPUexpected to become widely available later this year.

In conversations with these chip suppliers, Zhang reportedly said his company’s AI investments would dwarf the total spending by these chip suppliers. all its competitorsincluding Alibaba Cloud, Tencent Holdings Co., Ltd., Baidu Inc. and Huawei Technologies Co., Ltd.

Byte bouncing needs a fix as Chinese companies Purchase prohibited Out of concerns about national security, they chose advanced processors from Western companies. The United States was convinced that China would use these chips to develop more advanced weapons systems, so it took many steps Prevent Chinese companies from obtaining them.

This has forced China’s tech giants to turn to leasing chips. ByteDance is believed to be using data centers located outside of China to leverage Nvidia’s previous generation products Hopper AI GPUare not allowed to be exported to their home countries.

However, the U.S. government could still derail the bit-hopping plans. Last year, the Wall Street Journal reported that U.S. officials were studying There are more rules This will limit the ability of Chinese companies to rent chips from U.S. cloud providers.

Under the proposed rules, these companies would be required to report critical information about their customers to the U.S. government and ensure that Chinese companies are not able to access this information. In exchange, they will be allowed to provide artificial intelligence capabilities through global data centers without any permission. Other cloud providers must compete for licenses to obtain a limited number of high-end chips in each country.

While some Chinese companies are engaging in a cat-and-mouse game with the U.S. government in an attempt to get around U.S. chip export restrictions, others are taking the approach of doing more with less. For example, Chinese AI startup DeepSeek recently announced a new open-source large-scale language model that it says can compete with OpenAI’s GPT-4o, despite being trained only on Nvidia’s downgraded H800 chips, which are allowed in China. Sale.

DeepSeek reportedly trained it DeepSeek-R1 model The total cost is only US$5.6 million, which is much lower than the cost of its competitors to develop similar models.

Image source: SiliconANGLE/Meta AI