ByteDance Plans $14 Billion Nvidia H200 AI Chip Buying Spree As Computing Demand Soars

ByteDance plans to purchase 100 billion yuan ($14 billion) in AI chips in 2026, up from 85 billion yuan in 2025, with the bulk of spending directed toward Nvidia hardware, according to the South China Morning Post. The plan hinges on Beijing approving sales of Nvidia’s H200 GPUs in China. If approval is granted, Nvidia would need to scale up production of the China-tailored H200 with its manufacturing partner, TSMC (Taiwan Semiconductor Manufacturing Company).

The Trump administration recently authorized exports of H200 AI chips to China under a controlled licensing rule, marking a significant shift from prior export curbs. Despite U.S. approval for exports, Beijing has not yet formally approved purchases of H200s by Chinese firms, and reports indicate that access may be restricted or that imports of AI chips may be discouraged to protect its domestic semiconductor industry.

SCMP reports that ByteDance is planning a massive AI capex push, with H200-related spending in the neighborhood of $14 billion. This comes despite the company operating a 1,000-person internal chip design team, which has made progress on a new processor but has not yet matched Nvidia’s performance.

Demand for computing power is surging across TikTok, Douyin, its cloud unit Volcano Engine, and its large language models, driving the need for more advanced chips.

Doubao, ByteDance’s chatbot, now processes more than 50 trillion tokens daily, up from 4 trillion in late 2024, while Volcano Engine serves over 100 enterprise clients and will be a top AI cloud partner for China Central Television’s Spring Festival Gala.

In a separate report, Reuters said Chinese technology companies have shown strong interest in Nvidia’s second-most powerful AI chip and hope shipments can begin before the Lunar New Year.

Reuters also noted that Nvidia holds about 700,000 H200 AI chips in inventory, while Chinese technology firms have ordered more than 2 million units for next year, prompting Nvidia to ask TSMC to increase production.

Beijing now faces a strategic balancing act: ensuring its tech giants use best-in-class chips to compete in the AI race against the West, while simultaneously promoting the adoption of domestic alternatives, including products from Huawei Technologies’ Ascend unit, Moore Threads Technology, MetaX Integrated Circuits, and Cambricon Technologies.