Chinese tech firm ByteDance, the parent company of TikTok, is reportedly planning to spend 100 billion yuan on Nvidia's H200 AI GPUs in 2026, which translates to around $14 billion USD. This planned purchase is set to add to the growing stockpile of Nvidia GPUs the company already owns.
Despite numerous setbacks, ByteDance spent 85 billion yuan on Nvidia chips throughout 2025, marking a significant increase for next year, South China Morning Post reports. ByteDance is one of the biggest local tech players in the region, rivaling Tencent when it comes to AI operations. It was recently evaluated at a market cap of $500 billion, so this spending is part of its massive AI budget for 2026.
China is barred from receiving Nvidia's current-gen Blackwell products, but the Trump administration recently allowed the Hopper-based H200 GPUs to be sold to Beijing. Following this, the Chinese government even held meetings with its tech companies, including ByteDance, to carefully assess their inventory demands to permit these transactions.
The situation is complex, as China itself doesn't allow foreign silicon in its data centers, and the country is pushing to achieve technological self-reliance. Companies across the board are gunning to replace Western solutions with homegrown substitutes. ByteDance is part of that race, developing two custom AI GPUs with Broadcom and TSMC, which are believed to debut in 2026.
Nvidia H200 GPUs — the ones ByteDance will buy if allowed to (Image credit: Nvidia)
Around a year ago, ByteDance started renting cloud compute from other countries to evade U.S. sanctions that would prevent it from building servers within China. TikTok, the company's flagship IP, is essentially a massive inference engine with powerful AI needed for everything from curating algorithms (via TikTok's 'For You' page) to running ads and moderating content.
The company's own GPUs are conspicuously missing from the conversation. One would imagine that if mass production was originally planned for 2026, ByteDance wouldn't need to go so hard on its Nvidia investment, but Nvidia's GPUs are critical for AI training workloads, while its homegrown chips may be used for inference.
Therefore, even though custom silicon is purportedly being developed by the company for next year, ByteDance still needs chips for training; after all, it runs Doubao — the most popular AI chatbot in China.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.