China’s race to build a self-sufficient artificial intelligence ecosystem may be entering a new phase, and ByteDance appears to be right at the center of it. The company behind TikTok and AI chatbot Doubao is reportedly exploring major deals with domestic chipmakers as it looks to reduce reliance on foreign technology amid tightening U.S. export restrictions.
According to sources familiar with the discussions, ByteDance is in talks to purchase at least 50,000 AI chips from Shanghai-based Iluvatar CoreX. At the same time, the tech giant is also considering deploying AI processors developed by Baidu’s chip division, Kunlunxin. If these plans move forward, they would mark another major shift in China’s rapidly evolving AI hardware landscape, where local players are increasingly replacing once-dominant overseas suppliers.
ByteDance Could Add a Third Major Domestic GPU Partner
Until recently, companies building advanced AI systems in China heavily depended on Nvidia hardware. But a combination of U.S. export controls and Beijing’s push for technological independence has accelerated the search for local alternatives. ByteDance has already emerged as one of China’s largest buyers of domestic AI infrastructure, and a potential agreement with Iluvatar CoreX would further deepen that strategy.
Sources said Iluvatar CoreX could supply at least 50,000 chips to ByteDance this year if a deal is finalized. Most of these processors are expected to handle inference workloads rather than model training. In simple terms, inference refers to the process of answering user requests after an AI model has already been trained. Every time users interact with Doubao, ask questions, generate text, or receive recommendations, inference chips are doing much of the heavy lifting behind the scenes.
If the agreement materializes, Iluvatar CoreX would become ByteDance’s third major domestic GPU supplier after Huawei and Cambricon. The move would highlight how China’s largest internet companies are diversifying their chip partnerships instead of depending on a single vendor.
The discussions remain ongoing, and the final structure of any agreement could still change, according to the sources.
Why Inference Chips Are Becoming the New Battleground
Much of the public attention around AI hardware focuses on training giant models using the most powerful processors available. However, industry experts increasingly believe inference is where the next wave of demand will emerge. Once an AI system is built, it needs enormous computing resources to serve millions of users efficiently and at lower costs.
ByteDance’s growing investment in inference infrastructure appears closely tied to the expansion of Doubao, its flagship AI assistant. As the chatbot attracts a larger user base, the company requires a steady supply of processors capable of handling everyday interactions without sacrificing speed or reliability.
Iluvatar CoreX’s product lineup is designed with this distinction in mind. The company’s Tiangai series targets AI training tasks, while its Zhikai chips focus on inference operations. According to brokerage estimates, the Zhikai chips carry an average selling price of around 12,000 yuan, or roughly $1,775 per unit.
The emphasis on inference also reflects a broader trend across the industry. Companies are increasingly prioritizing cost-efficient deployment over simply chasing the most powerful training hardware available.
China’s Homegrown Chipmakers Are Closing the Gap
The reported negotiations underline how quickly China’s domestic AI chip ecosystem has matured. Just a few years ago, Nvidia dominated China’s accelerator market with little meaningful competition. Today, the picture looks very different.
Chinese GPU and AI chip manufacturers accounted for nearly 41% of China’s AI accelerator server market last year, steadily eroding Nvidia’s once-commanding position in one of its most valuable international markets. Nvidia CEO Jensen Huang has previously acknowledged that the company’s share in China has effectively fallen to zero under current restrictions.
At the same time, local suppliers have been scaling up both production and commercial reach. Baidu’s Kunlunxin chips have already found customers among major technology firms, with Tencent reportedly using the processors. ByteDance’s interest in Kunlunxin suggests confidence in deploying multiple domestic architectures depending on workload requirements.
Beijing has also actively encouraged the adoption of locally developed semiconductors as part of its broader push for technological self-reliance. Government procurement programs, policy support, and rising demand from Chinese cloud providers have created opportunities for startups that previously struggled to compete with global giants.
Iluvatar CoreX Faces a Defining Commercial Moment
For Iluvatar CoreX, securing ByteDance as a customer could become the company’s biggest commercial breakthrough to date. The Shanghai-based startup has traditionally relied heavily on government procurement contracts and public-sector demand. Winning business from one of China’s most aggressive investors in AI infrastructure would significantly strengthen its credibility within the private sector.
The company, which debuted on the Hong Kong stock exchange in January, reported revenue of 1 billion yuan in 2025. Around 90% of that income came from GPU sales, reflecting surging interest in domestic AI hardware solutions.
Analysts expect that momentum to continue. A Huatai Securities research note projects Iluvatar CoreX’s annual revenue could climb to 3.04 billion yuan this year. Total shipments are forecast to jump by 139%, surpassing 100,000 chips for the first time.
If ByteDance ultimately signs on the dotted line, the agreement would represent more than just another procurement deal. It would signal that China’s largest internet companies increasingly trust homegrown AI processors to power consumer-facing services at scale.
The outcome of these discussions remains uncertain for now, and none of the companies involved have publicly commented on the reported negotiations. But whether the deal proceeds exactly as described or evolves into something different, one thing is becoming increasingly clear: China’s AI future is being built with more domestic silicon than ever before.
