en.Wedoany.com Reported - On June 24, local time in the United States, U.S. chip company Qualcomm announced that Microsoft and Meta will adopt its new AI chips, and the company will also develop custom chips for two other unnamed hyperscale cloud providers. At its 2026 Investor Day, Qualcomm unveiled the Dragonfly product portfolio for data centers, including the Dragonfly C1000 CPU, High Bandwidth Compute (HBC) technology, the Dragonfly AI300 inference accelerator, and custom chip solutions, aiming to enter the AI data center infrastructure market.
Microsoft will adopt Qualcomm's new High Bandwidth Compute chip architecture. This solution, referred to by Qualcomm as High Bandwidth Compute (HBC), primarily targets memory bandwidth and data movement bottlenecks in AI inference scenarios. Reuters reported that this chip category from Qualcomm relies on relatively low-cost memory used in smartphones and laptops, rather than the high-bandwidth memory (HBM) commonly used in Nvidia's GPUs or the static random-access memory (SRAM) employed by Cerebras Systems. Tony Pialis, head of Qualcomm's data center business, stated that the value of this approach lies in its advantage between performance and cost.
Meta will adopt Qualcomm's Dragonfly C1000 CPU, specifically designed for AI data centers. According to Qualcomm's official materials, the Dragonfly C1000 CPU targets agent workloads, general computing, and AI head node tasks. It features custom Oryon CPU cores with a target frequency exceeding 5GHz and a chiplet design with over 250 cores. The product also supports PCIe Gen 7 and CXL connectivity, with commercial availability expected in 2028. Qualcomm stated that the Dragonfly C1000 is planned for use in Meta's next-generation server clusters, and the two companies have reached a multi-year, multi-generation data center CPU collaboration.
This supply plan indicates that Qualcomm is shifting its business focus from mobile chips to data center chips. Historically, Qualcomm has been known for smartphone chips, communication basebands, and low-power computing capabilities, with data centers not being its primary revenue source. As AI inference demand expands, cloud providers need to rebalance computing power, power consumption, memory bandwidth, and deployment costs. Qualcomm is attempting to introduce the low-power design expertise accumulated in the mobile chip field into server and AI infrastructure scenarios.
Qualcomm also disclosed that it has secured two unnamed hyperscale cloud customers, for whom it will develop custom chips, with related revenue expected to begin by the end of this year. Custom chip business typically requires joint design between cloud providers and chip suppliers in areas such as architecture, power consumption, memory, interconnects, software stacks, and manufacturing timelines, leading to longer development cycles. However, once large-scale deployment begins, it may form a continuous supply relationship. Reuters reported that Qualcomm is advancing three types of data center chips: central processing units, inference accelerators, and application-specific integrated circuits (ASICs), a market where companies like Broadcom and Marvell are also competing.
In its growth plan released on the same day, Qualcomm stated that it expects its data center business revenue to exceed $15 billion by fiscal year 2029, and has raised its non-mobile business revenue target from $22 billion to $40 billion. The company said that over the next three to five years, AI computing will continue to be distributed across terminals, edges, and clouds, with markets such as data centers, automotive, industrial systems, networks, and robotics entering new growth phases.
Qualcomm still faces intense competition in entering the AI data center chip market. Nvidia maintains a leading position in the AI accelerator market, while Amazon and Google are also advancing their own cloud chips. Qualcomm's recent support from Microsoft, Meta, and two unnamed hyperscale cloud customers provides early commercial validation for its data center roadmap. However, whether it can achieve scale revenue will depend on the production progress of HBC technology, the delivery timeline of the Dragonfly C1000, the adaptability of the software ecosystem, and the actual deployment scale by cloud providers.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
