ByteDance is accelerating its transition away from American silicon, engaging in talks to purchase at least 50,000 AI chips from domestic supplier Daysci. The move comes as the Chinese technology giant dramatically increases its capital expenditure on AI infrastructure, while Nvidia’s market share in China has effectively fallen to zero amid tightening US export controls and Beijing’s push for technological self-reliance.
The chips involved in the negotiation are primarily Daysci’s Zhikai series of Cloud Inference GPUs, designed for inference rather than model training. If the deal is finalized, Daysci will become ByteDance’s third-largest domestic GPU supplier, trailing only Huawei and Cambricon, according to a 36Kr report.
A Three-Pronged Domestic Supply Strategy
ByteDance’s computing power supply chain strategy is becoming increasingly clear. The company is utilizing Huawei’s Ascend processors and Cambricon’s high-end training cards for the computationally intensive task of training large language models, while introducing Daysci’s Zhikai series for the less demanding but higher-volume inference workloads. This parallel approach of “walking on two legs” is rapidly becoming the standard configuration for Chinese internet giants navigating the complex geopolitical landscape.
The pressure to secure reliable computing power is immense. ByteDance’s AI intelligent assistant, Doubao, reached 345 million monthly active users as of March 2026. The daily average token call volume for the Doubao large model has exceeded 120 trillion, representing a thousand-fold increase since its initial launch. To support this massive scale, ByteDance has reportedly increased its 2026 budget for AI infrastructure capital expenditure by 25 percent to 200 billion yuan (approximately $27.5 billion), with some reports suggesting the cap could reach $70 billion.
(Related: Five AI Models, Zero Nvidia: China’s Labs Are Proving Domestic Chips Can Train Frontier AI)
The Daysci deal also illustrates the diversification logic driving ByteDance’s procurement strategy. Concentrating too much of its chip supply with a single domestic vendor, even Huawei, whose Ascend series has made the most progress, creates a different kind of supply chain risk. By spreading orders across Huawei, Cambricon, and now Daysci, ByteDance is building redundancy into its hardware stack. The company is also reportedly in discussions with a broader group of tier-two chipmakers, including Biren, MetaX, Iluvatar CoreX, Moore Threads, and Enflame, suggesting that the final supply chain will be deliberately fragmented across multiple vendors to avoid single points of failure.
The Rise of China’s Grade I AI Chips
The shift toward domestic suppliers is not merely a reaction to US sanctions; it is also driven by the improving quality of Chinese silicon and direct government intervention. China’s Cyberspace Administration (CAC) has reportedly ordered major tech firms, including Alibaba and ByteDance, to halt purchases of Nvidia’s latest AI chips.
In May 2026, the Chinese government established an AI chip category in its security and reliability assessment framework. Nine domestic chipmakers received the highest Grade I rating, including Huawei HiSilicon, Alibaba’s T-Head, Biren Technology, Hygon Information, Daysci, Muxi, and Moore Threads. Biren Technology, which was listed on the Hong Kong Stock Exchange in January 2026 and saw its market capitalization exceed HK$100 billion, has already completed “Day-0” adaptation for major domestic models, including Zhipu’s GLM-5.2 and DeepSeek’s full series. The Grade I designation matters not just as a quality signal but as a procurement guide: government-affiliated entities and state-owned enterprises are increasingly expected to prioritize Grade I chips in their purchasing decisions, creating a captive demand base that accelerates the commercial viability of domestic suppliers and funds the next generation of R&D investment.
The trajectory of China’s domestic chip ecosystem over the past two years has been faster than most outside observers anticipated. The conventional wisdom in 2023 was that China was five to seven years behind the frontier in AI chip design and manufacturing, and that export controls would lock in that gap. What has actually happened is more complicated: the gap in raw compute performance per chip remains significant, but Chinese companies have compensated by optimizing software, innovating in model architecture, and leveraging the sheer scale of deployment. ByteDance’s experience with Huawei’s Ascend chips, initially frustrating, now reportedly workable for most inference tasks, is emblematic of a broader pattern in which necessity has driven a level of hardware-software co-optimization that would not have occurred if Nvidia chips had remained freely available. The Daysci deal is another data point in a story moving faster than the export control architects in Washington expected.
For Nvidia, the collapse of its China market share represents the permanent loss of what was once its second-largest revenue market. The company has tried to navigate export controls by designing downgraded chips, the H20 and the L20, specifically calibrated to comply with US restrictions while remaining competitive in China. Those efforts have largely failed: each successive round of controls has closed the loopholes that the previous downgraded chip was designed to exploit, and Chinese customers have increasingly concluded that investing in domestic alternatives is more strategically reliable than depending on a supplier whose product availability is subject to Washington’s geopolitical decisions. ByteDance’s pivot to Daysci, Cambricon, and Huawei is not a temporary workaround; it is a permanent restructuring of its supply chain that Nvidia is unlikely to reverse.