ByteDance AI Chips Face Substrate Shortage
According to supply chain sources, the rapid development of artificial intelligence (AI) models in China is placing increasing pressure on computing capacity, while shortages of advanced packaging substrates are emerging as a key constraint. ByteDance is reportedly planning to spend RMB 100 billion (approximately US$14.9 billion) annually by 2030 on large-scale purchases of chip packaging substrates.
At the same time, Huawei HiSilicon, Hygon Information Technology, ByteDance, and major semiconductor packaging and testing companies are competing for substrate capacity as manufacturers accelerate expansion plans. Industry sources described ByteDance’s reported procurement target as exceptionally large, noting that even if substrate manufacturers restart or accelerate capacity expansion, existing production capacity could struggle to meet such demand.
ByteDance is expanding its access to packaging substrates to support its in-house AI chip development and the gradual transition of China’s semiconductor localization efforts toward mass production.
Market reports have also indicated that ByteDance has been in discussions with Broadcom about jointly developing proprietary AI chips and with Samsung Electronics regarding wafer foundry services and high-bandwidth memory (HBM) procurement. The reported chip production volume could reach at least 100,000 units in 2026 and potentially increase to 350,000 units later. However, ByteDance has previously stated that reports concerning its in-house chip development project were inaccurate, while Samsung Electronics declined to comment.
Substrate manufacturers said shortages of high-end substrates remain ongoing, with major customers such as NVIDIA already securing significant capacity through long-term agreements. For ByteDance, which entered the market later, obtaining priority access to capacity could be difficult unless it can commit to sufficiently large purchase volumes.
To prevent capacity from being reserved without firm demand, some substrate manufacturers have raised their cooperation requirements. These may include advance payments equivalent to as much as 50% of the value of the reserved capacity or joint investment in capacity expansion. Manufacturers are also increasingly favoring customers capable of committing to substantial order volumes, aiming to avoid a repeat of the capacity oversupply seen after the COVID-19 pandemic.
Industry sources said Unimicron Technology’s subsidiary Phoenix Precision Technology, Nan Ya PCB’s Kunshan facility, and Zhen Ding Technology’s subsidiary Sumei Technology are among the substrate manufacturers that AI computing chip companies are seeking to secure for advanced packaging capacity.
The developments highlight the growing importance of advanced packaging substrates in China’s AI semiconductor supply chain. As AI chip production moves toward larger-scale deployment, access to high-end substrate capacity could become an increasingly important factor affecting chip production schedules and overall computing infrastructure expansion.
