Qualcomm developing export-control-compliant custom AI chips for Chinese market to navigate US restrictions.
The data center accelerator market is becoming increasingly crowded, with major chipmakers entering the segment, including in heavily regulated China. According to Nikkei, citing Qualcomm CEO Cristiano Amon, the company is working to bring its full suite of four data center product lines to the country, including customized AI accelerators designed to comply with U.S. export restrictions on advanced AI chips.
On June 25, Qualcomm unveiled a comprehensive data center portfolio spanning the Dragonfly C1000 CPU, High Bandwidth Compute (HBC) technology designed to address the memory wall with lower energy per token, the Dragonfly AI300 inference accelerator, advanced connectivity solutions, and custom silicon offerings. This announcement builds on the company's broader strategy to capture China's growing AI infrastructure demand.
Qualcomm's approach is reinforced by a deal already secured with ByteDance, as reported by Bloomberg in late May, to supply custom AI data center chips structured to comply with existing U.S. export-control thresholds. This arrangement underscores Qualcomm's effort to tap AI demand while remaining within regulatory limits.
Nevertheless, Qualcomm faces similar regulatory challenges to those confronting NVIDIA and other chipmakers in the China market. In June, the Trump administration tightened AI chip export rules, with the U.S. Commerce Department requiring licenses for Chinese companies headquartered in China, even when operating overseas.
China's significance to Qualcomm cannot be overstated. The country remained Qualcomm's largest market in 2025, accounting for 46% of total revenue, primarily driven by smartphone chips.
Looking ahead, Qualcomm is ambitious about its AI chip business outlook. According to Reuters, the company's data center chip business is expected to generate US$15 billion in revenue by 2029. CFO Akash Palkhiwala said that data center-related revenue is expected to reach US$5 billion in fiscal 2027, including roughly US$1 billion from newly secured custom silicon customers.
This growth is underpinned by Qualcomm's High Bandwidth Compute architecture, which relies on lower-cost memory technologies commonly used in smartphones and notebooks rather than conventional GPU-centric systems built around expensive HBM. According to TechNews, Qualcomm expects its first-generation HBC-based AI250 accelerator to launch in mid-2027. The company has outlined a follow-on roadmap with second-generation HBC (HBC Gen2) planned for 2028, to be introduced alongside the AI300 accelerator. Compared with the AI200, the new solution is expected to deliver up to a 54-fold increase in effective bandwidth while achieving up to 7x higher bandwidth-per-watt efficiency versus HBM.
Qualcomm has already secured commitments from major customers. Microsoft and Meta will adopt its newly launched AI chips, while two additional hyperscale customers have signed on for custom silicon projects.