It’s been a little over a year since Qualcomm first announced a new strategy to enter the feeding frenzy around the exploding AI data centers opportunity. Now, with key acquisitions and hyperscale customers, Qualcomm is positioned to enter the data center market with full rack-scale solutions that include new CPUs, AI accelerators, and networking solutions that could produce over a billion in revenue in less than a year.
Rebuilding Qualcomm
Under CEO Cristiano Amon, Qualcomm has been on a journey to expand its potential market beyond handsets, with huge success in automotive and wearable tech and growing opportunities in industrial applications, including robotics. However, none matches the scale of the opportunity presented by its expansion into the data center. In May 2025, Qualcomm announced Humain as its first data center customer and forthcoming custom CPUs based on the company’s Arm-compatible Oryon architecture. Just a few weeks later, Qualcomm announced the acquisition of Alphawave Semi for $2.4 billion. And yesterday (June 24), Qualcomm announced the $3.9 billion acquisition of Modular, a software company focused on allowing AI workloads to run efficiently across different hardware platforms.
Qualcomm already had a custom CPU architecture, an AI accelerator, and expertise in low-power semiconductor design and AI software. But Alphawave provides a foundation in high-speed interconnects and networking, advanced packaging, and custom silicon services that are critical to Qualcomm’s rack-scale strategy. In many ways, it’s similar to the value of Mellanox to Nvidia and Pensando to AMD. With high-speed Serdes, PCIs, and CXL interconnects, UCIe chiplet technology, custom ASIC design expertise, and advanced networking fabrics, Alphawave enables AI accelerator clusters, chiplet-based server CPUs, high-speed memory connectivity, custom ASICs, and complete rack-scale solutions built around Qualcomm’s core technology and expertise. Furthermore, Modular will help create a hardware agnostic data center environment, allowing for the use of different hardware platforms, including Qualcomm, collectively and efficiently for data center-scale solutions.
Data center acceleration
There have been rumors that Qualcomm is in talks with ByteDance. In addition, earlier this month at Computex, Qualcomm announced a new brand for its data center solutions, called Dragonfly, and teased that it would announce its first hyperscale customer at the company’s investor day. While the customer was not officially announced, they did indicate that the company currently has two global-scale hyperscale customers. In addition, the other companies represented at the event may provide some clues. Amazon and Microsoft were represented when discussing data center solutions, and Meta and Google when discussing consumer platforms.
View All Rather than focusing on the customers, yesterday’s investor presentation was focused on the rollout of the data center strategy. Qualcomm has a four-part strategy through 2028 that will eventually lead to an annual cadence of new products. The first is the introduction of its connectivity portfolio brought over from the Alphawave acquisition, including electrical and optical interconnects up to 800G solutions today and 1.6T by the end of 2026, custom silicon solutions to lead partners beginning in 2027, and the introduction of a new family of server C1000 CPUs in 2028. The new C1000 CPUs will feature the new High-Bandwidth Compute (HBC) memory architecture, with up to 250 Oryon cores operating at up to 5GHz, beyond the frequency range of the competing standard Arm and x86 CPU cores. Qualcomm did indicate that it is working with lead customers for all of the new products.
A key feature of the new C1000 CPUs and the AI250 accelerator is the HBC memory architecture, which includes in-package LPDDR memory stacks atop a full XPU accelerator. According to Qualcomm, this will provide higher performance efficiency and memory bandwidth than even the forthcoming HBM4e, which will feature customization on the base die. Qualcomm believes that the HBC memory architecture, combined with the Oryon CPU and Hexagon AI accelerator architectures, will give the company both a performance and power-efficiency advantage in a segment where every watt matters, or as Qualcomm put it, the cost per token matters.
Unlike its competitors, however, Qualcomm is not targeting the entire data center or AI stack. The initial Qualcomm platform is aimed at AI decoding. Qualcomm will leverage the work at Modular to ensure that all the platforms in a heterogeneous data center work together efficiently. While Qualcomm is initially focused on AI decode, this is not to say that Qualcomm does not have ambitions to be in other parts of the AI data center. Given the company’s aggressive management, there is likely more coming through both organic growth and acquisitions.
Key takeaways
In all, the new announcements and strategy put Qualcomm in direct competition with the likes of AMD, Intel, and Nvidia for the rack-level solutions; Astera Labs, Broadcom, Credo, and Marvell for networking; and Arm, Broadcom, Marvell, MediaTek, and possibly AMD and Intel for custom silicon. With a growing data center technology stack, Qualcomm seems well-positioned to compete, despite being the newest player in the game.
If the company’s projections are correct, the data center segment will add $5 billion to Qualcomm’s revenue in 2027 and $15 billion in 2029. Combined with success in automotive, industrial, and other segments, Qualcomm raised its non-handset revenue from $22 billion in 2029 to $40 billion. This comes at a time when the company is expecting a rapid decrease in Apple modem revenue. Taken together, the data center combined with auto and IoT, Qualcomm’s potential TAM is likely to increase by over a trillion dollars over the next decade.
Read also:
What Comes Next for Arduino After Qualcomm’s Acquisition?
EE Times spoke to Guneet Bedi, SVP of Arduino, who was unequivocal in saying that Arduino’s governance structure had remained intact even after the acquisition.