Realignment on AI in Data Centers Qualcomm Signs Agreement with Bytedance to Produce AI Chips for China

Related Vendors

dataTec AG

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Qualcomm aims to grow with AI chips in China and, according to media reports, is working with Bytedance to bypass US trade restrictions legally. At the same time, the company is strengthening its software base for AI inference and acquiring AI startup Modular for nearly $4 million.

Qualcomm is advancing its shift towards AI data centers with two strategic moves: According to media reports, the US chip company is working on custom AI chips for the Chinese market and aims to strengthen its position in the inference business through the acquisition of the AI software company Modular Technologies. Both initiatives are aimed at making Qualcomm less dependent on the smartphone business and establishing itself in the growing market for AI infrastructure.

At an investor day on June 25, Qualcomm presented a broad data center portfolio. This includes the Dragonfly C1000 CPU, HBC technology to reduce the so-called memory wall, the Dragonfly AI300 inference accelerator, connectivity solutions, and custom semiconductors. With this, Qualcomm targets not only traditional cloud providers but also customers with their own AI workloads and specific requirements for energy efficiency, bandwidth, and cost per token.

China Remains a Key Market Despite Regulation

China plays a central role in this strategy. According to the business newspaper Nikkei Asia, Qualcomm plans to offer its four data center product lines there as well, including specially tailored AI accelerators designed to comply with US export controls for advanced AI chips. The market remains of great importance to Qualcomm: in 2025, 46 percent of the company's revenue reportedly came from China, primarily through smartphone chips so far.

Qualcomm aims to reduce its reliance on the mobile business. The company is one of the world's leading providers of smartphone modem chips and Snapdragon processors but faces pressure from a weaker device market and rising memory prices. Entering the market for data center CPUs, inference accelerators, and custom ASICs is intended to open up new revenue streams.

Bytedance appears to be at the center of Qualcomm's China plans. Bloomberg reported back in May that Qualcomm had reached an agreement to supply the company, which owns the social network TikTok among other ventures, with millions of custom ASICs for AI data centers. Additionally, Qualcomm is set to assist Bytedance in bringing its own internal chip design into volume production. Reuters later reported that Qualcomm is negotiating with Bytedance over chip design services, which may also involve technologies from the acquired connectivity specialist Alphawave Semi.

Details about the planned components are only partially known so far. According to Reuters, the focus is partly on video processors, with mass production potentially targeted by the end of the year. At the same time, sources emphasized that the talks would not necessarily lead to a finalized chip design or production. Qualcomm and Bytedance reportedly did not comment on the reports.

ASICs Below Export Control Limits

The regulatory framework is strategically crucial. The Qualcomm ASICs intended for Bytedance are designed to remain within the current performance limits of US export controls. Bytedance's own chip design is also set to be developed in a way that ensures it does not exceed these thresholds. This approach makes the project economically significant but politically more secure than the direct sale of advanced AI GPUs to China.

The regulatory risks remain significant. Nikkei points out that Qualcomm is likely to face similar challenges as Nvidia and other chip providers. The US government has tightened export rules for AI chips; reportedly, the US Department of Commerce requires licenses for Chinese companies headquartered in China, even if they operate overseas. For US corporations, China remains a large but increasingly complex market.

For Bytedance, such a deal would be a way to secure AI computing power beyond Nvidia. The company requires accelerators for inference and agentic AI applications while operating in an environment of increasing scrutiny from both Washington and Beijing. Qualcomm could thereby emerge as an additional Western provider alongside Nvidia and the internal chip programs of major hyperscalers.

Qualcomm ties these plans to ambitious revenue targets. According to Reuters, the company expects around 15 billion US dollars in revenue from data center chips by 2029. For the fiscal year 2027, CFO Akash Palkhiwala has already projected 5 billion US dollars, including approximately 1 billion US dollars from new custom silicon projects.

High-NA EUV Lithography

ASML Reportedly Delivers Chip Machinery to China

Software as Another Pillar of the AI Strategy

In parallel with its hardware offensive, Qualcomm is acquiring the AI infrastructure company Modular Technologies. The transaction is expected to close in the second half of 2026, subject to customary closing conditions and regulatory approvals. Qualcomm initially did not disclose financial details; however, Reuters calculations and media reports estimate the predominantly stock-based transaction to be valued at around 3.9 to 4 billion US dollars.

Modular is developing an open, AI-native software platform designed to efficiently run models across different hardware architectures. Supported architectures include CPUs, GPUs, NPUs, custom ASICs, and other accelerators. Developers will not need to rewrite models for each target platform but can deploy them across various systems using a shared software foundation.

For Qualcomm, this is strategically important because success in the data center depends not solely on hardware. Inference costs, energy efficiency, and portability are increasingly determined at the software level. Modular is expected to enhance Qualcomm's ability to offer an optimized AI compute layer from edge devices to data centers, providing customers with an alternative to highly proprietary ecosystems.

Qualcomm CEO Cristiano Amon envisions the industry heading toward disaggregated, cross-manufacturer architectures that require a more open software foundation. With Modular, Qualcomm is complementing its hardware roadmap with a developer and platform proposition: the company aims not only to supply chips for AI inference but also to create an ecosystem that efficiently integrates models, software, and heterogeneous accelerators. 

Model-Based System Development

Renesas Acquires Embedded Software Specialist Pictorus