ByteDance Evades U.S. Sanctions with $7 Billion Nvidia Cloud Deal!

Despite restrictions placed on Chinese companies, including TikTok’s parent company ByteDance, from purchasing top-tier Nvidia GPUs for use within China, ByteDance has navigated these limitations by utilizing Nvidia GPUs hosted in international cloud data centers. According to sources cited by The Information, ByteDance plans to significantly increase its expenditure on these GPUs, potentially investing up to $7 billion next year. However, ByteDance has refuted this claim.

This ambitious financial commitment is part of a broader $20 billion strategy to enhance its AI infrastructure, which includes spending on cloud-based Nvidia GPUs, data centers, and even undersea cables. Although U.S. regulations prevent ByteDance from buying Nvidia GPUs directly or using American cloud services, these rules do not restrict the company from accessing similar services in other regions, such as the Middle East or various Asian countries. This loophole allows ByteDance to use American technology without violating U.S. sanctions targeting China’s AI and high-performance computing sectors.

If the reports are accurate and ByteDance does allocate $7 billion for Nvidia GPU cloud access, it could become one of the largest global users of AI hardware.

In the United States, leasing Nvidia’s H100 GPUs starts at $1.33 per hour for long-term commitments. Comparable rates are expected in other nations. With an hourly rate of approximately $1.3, ByteDance could afford to operate about 614,682 H100 GPUs continuously throughout 2025 with a $7 billion budget.

It remains uncertain whether such a large quantity of H100 GPUs is available for lease in the Middle East and Asia, or whether ByteDance truly requires this many processors for its AI operations, which currently seem to be fairly modest. For instance, its largest AI initiative appears to be the Doubao AI chatbot, which has 51 million active users. Thus, ByteDance might either spend less on its AI infrastructure or it might be planning substantial expansions to its AI projects, necessitating increased computational power. Alternatively, it could continue to acquire lower-tier Nvidia H20 HGX and B20 GPUs for use within its own data centers in China, while also renting cloud-based processors. Reports suggest that ByteDance spent over $2 billion on more than 200,000 Nvidia H20 GPUs in 2024, indicating it likely won’t completely shift away from purchasing hardware to solely relying on international cloud providers.

Furthermore, ByteDance is reportedly collaborating with Broadcom to develop its own AI processors, which could lessen its dependence on Nvidia. Rumors suggest that the company is working on two types of chips: one for training and one for inference. These chips are expected to be manufactured by TSMC using its N4/N5 process technologies and are projected to start mass production in 2026. While these in-house processors might not outperform Nvidia’s HGX H20 due to U.S. export controls limiting TSMC’s ability to deliver high-performance GPUs to Chinese firms, they could provide a more cost-effective solution for ByteDance.