October 16, 2026 — ByteDance is in advanced talks with MediaTek to purchase custom data center chips, according to people familiar with the matter. The chips would be application-specific integrated circuits, or ASICs, tailored to ByteDance’s platform needs. If the deal goes through, ByteDance would likely become MediaTek’s second-largest ASIC customer after Google.
Neither side has announced a final agreement, and details such as chip specifications, volume, pricing and delivery schedule have not been disclosed.
For MediaTek, the potential order would be more than just another customer. The company has long been known for smartphone chips, but it is now trying to build data center ASICs into a second growth engine. MediaTek expects AI chip sales to reach about $2 billion in 2026 and aims to capture around 15 percent of the roughly $80 billion data center AI chip market in 2027. Winning an internet company of ByteDance’s scale would help it reduce reliance on mobile chips and strengthen its position against Qualcomm and Broadcom in the data center market.
For ByteDance, the move is mainly about diversifying its AI computing supply. Due to U.S. export controls, Chinese tech companies cannot reliably purchase Nvidia’s most advanced AI chips, while domestic suppliers such as Huawei cannot yet meet all demand. ByteDance earlier reached a similar arrangement with Qualcomm to buy millions of AI chips. Partnering with MediaTek would give it another source of custom compute capacity beyond general-purpose GPUs.
The bigger backdrop is spending. ByteDance is pushing AI infrastructure investment to a new level and is considering raising its 2026 capital expenditure to as much as $70 billion, more than double the previous year. If business conditions allow, it could increase spending to $100 billion in 2027. To support the plan, ByteDance secured a $30 billion loan in September.
If finalized, the deal would further show that custom ASICs are moving from a backup option to a standard part of large tech companies’ infrastructure. They may not fully replace GPUs for large-model training, but they offer clear cost and efficiency advantages for inference, video processing, recommendation systems and other workloads. For ByteDance, it is about expanding capacity and reducing dependency. For MediaTek, it is a key step from smartphone chips into the data center market.
