Like other Chinese entities, TikTok owner ByteDance cannot purchase the most powerful Nvidia GPUs and install them in its data centers in China. However, the company found that it could still use Nvidia GPUs that were physically located in cloud data centers in other countries. The company reportedly aims to expand use of such GPUs next year and is spending $7 billion on Nvidia GPUs information, Citing its own sources. ByteDance denies the report.
According to the report, the Byte Bounce Project has invested more than $20 billion in artificial intelligence infrastructure, including $7 billion in accessing advanced Nvidia GPUs in the cloud, data centers and even undersea cables. The United States bans ByteDance from purchasing Nvidia GPUs and using U.S. cloud services. However, it does not prevent ByteBounce from accessing cloud services elsewhere, such as in the Middle East or Asian countries. As a result, ByteDance can use U.S. processors while technically complying with U.S. sanctions on China’s artificial intelligence and high-performance computing fields.
ByteDance denied the report. However, if the report is true and ByteDance invests $7 billion in cloud access to Nvidia GPUs, it would become one of the largest consumers of AI hardware in the world.
Nvidia’s H100 GPU can be accessed whenever and wherever needed. In the United States, long-term use of the H100 GPU costs $1.33 per hour. Pricing in other countries should be more or less comparable, and at around $1.30 an hour, ByteDance will be able to rent a cluster of 614,682 H100 GPUs for $7 billion in 2025, 24/7/ Operating 365 hours a day.
We’re not sure if there are nearly 615,000 H100 GPUs available for rent in the Middle East and Asia, nor are we sure that ByteDance needs that many processors for its training and inference workloads since its AI projects are quite limited. For example, it has Doubao AI chatbot with 51 million active users, which is considered its largest project. Therefore, the company will either spend less on renting AI infrastructure, plan to significantly expand its AI projects and therefore require more AI capabilities, or plan to continue purchasing Nvidia’s stripped-down H20 HGX and B20 GPUs to use in its own In addition to renting processors from cloud providers, the system operates a data center in China. For example, with ByteDance reportedly spending over $2 billion to purchase over 200,000 Nvidia H20 GPUs in 2024, it’s unlikely that the company will stop buying its own hardware and go back to relying entirely on cloud providers in other countries.
It is worth noting that byte bounce also Reportedly working with Broadcom to develop its own artificial intelligence processor to reduce dependence on Nvidia. Rumor has it that the company will develop two processors: one for training and another for inference. The chips are expected to be manufactured by TSMC using its N4/N5 process technology and will enter mass production in 2026. Stopping TSMC (TSMC) from shipping high-performance GPUs to Chinese entities), in-house processors will bring greater cost-effectiveness to the company.