ByteDance commits $14 billion to purchasing Nvidia chips for AI infrastructure expansion.
ByteDance, the parent company of TikTok and Douyin, plans to spend 100 billion Chinese yuan ($14 billion) on Nvidia's H200 AI chips in 2026, marking a significant increase from its estimated 85 billion yuan ($12 billion) budget for 2025. This expansion reflects the company's growing demand for computing power across its social media platforms, cloud services, and large language models. The spending surge comes days after the Trump administration approved exports of Nvidia's H200 chips to China under a controlled licensing framework, though Beijing has yet to formally approve purchases, leaving Chinese firms in regulatory limbo.
Nvidia's H200, built on Taiwan Semiconductor Manufacturing Company's four-nanometer process, is the company's second-most powerful AI chip, offering six times the performance of its China-tailored H20 predecessor and remaining the most advanced AI processor legally available to Chinese firms. The chip features 2,496 CUDA cores responsible for executing parallel computing tasks essential for AI and high-performance computing applications. Chinese tech giants including ByteDance, Tencent, and Alibaba have already placed orders exceeding two million H200 units for 2026. With Nvidia currently holding 700,000 chips in inventory and requesting TSMC to ramp up production, industry analysts estimate Nvidia could generate an additional $15-20 billion in revenue this year alone from Chinese buyers.
ByteDance's computing demands have skyrocketed, driven by Doubao, its AI chatbot now processing 50 trillion tokens daily—up from 4 trillion in late 2024—and Volcano Engine, its cloud division serving over 100 enterprise clients and securing an exclusive AI partnership with China Central Television's Spring Festival Gala. To support this growth, ByteDance has expanded its internal semiconductor team to 1,000 engineers, developing a lower-cost alternative to Nvidia's H20, though the company still relies heavily on Nvidia's CUDA software ecosystem, making a full transition to domestic chips challenging.
Despite U.S. approval for H200 exports, Beijing's regulatory uncertainty fuels speculation that China may restrict imports to protect domestic chipmakers like Huawei, Cambricon, and Moore Threads, or mandate hybrid procurement requiring firms to pair Nvidia orders with local alternatives. Nvidia hopes to begin shipments before Lunar New Year in mid-February 2026, but delays could disrupt timelines. Meanwhile, ByteDance is reportedly deepening ties with Huawei, ordering $5.7 billion worth of Ascend AI chips as a hedge against Nvidia supply risks.
Nvidia CEO Jensen Huang's recent visit to Beijing, where he met Chinese officials and industry leaders just days after a private meeting with U.S. President Donald Trump, highlights Nvidia's delicate balancing act between U.S. regulations and Chinese market demands. Huang is positioning Nvidia as a neutral enabler of AI progress amid escalating U.S.-China tensions. As ByteDance's $14 billion AI chip budget underscores the voracious demand for computing power in China's tech sector, the battle for AI supremacy is being fought not just in algorithms, but in semiconductor supply chains.