Qualcomm has announced its return to the data center CPU market, unveiling plans to develop custom processors that seamlessly integrate with Nvidia’s AI chips. This strategic move is facilitated by Nvidia’s new NVLink Fusion technology, which enables high-speed communication between CPUs and GPUs, aiming to deliver high-performance, energy-efficient computing solutions for data centers.
A Strategic Collaboration: Qualcomm and Nvidia
At the Computex 2025 trade show in Taipei, Nvidia introduced NVLink Fusion, a platform designed to allow third-party CPUs and AI chips to integrate with its technology. Qualcomm is among the initial partners, alongside companies like Fujitsu, MediaTek, and Marvell. This collaboration marks a significant shift in Nvidia’s strategy, opening its previously proprietary AI server platform to other chipmakers.
Technical Advancements: NVLink Fusion
NVLink Fusion represents a significant technological breakthrough, allowing Qualcomm’s custom CPUs to directly connect to Nvidia GPUs with dramatically improved performance. Unveiled at Computex 2025, this technology delivers up to 1.8 TB/s of bandwidth per GPU—14 times faster than PCIe Gen5—enabling high-performance, energy-efficient computing specifically optimized for AI workloads.
Reviving Past Initiatives
Qualcomm’s renewed focus on data center CPUs follows its previous efforts in the 2010s, which included collaborations with companies like Meta Platforms. These earlier initiatives were halted due to cost and legal challenges. The current strategy is bolstered by Qualcomm’s acquisition of a team of former Apple chip designers in 2021, leading to renewed discussions with Meta and a confirmed agreement with Saudi AI company Humain to co-develop a custom data center CPU.
Implications for the AI and Data Center Markets
Qualcomm’s entry into the data center CPU market, in partnership with Nvidia, could significantly impact the AI infrastructure landscape. By offering CPUs that integrate seamlessly with Nvidia’s GPUs, Qualcomm aims to provide solutions that are both high-performing and energy-efficient. This move also reflects a broader trend of companies seeking to develop custom silicon to meet the specific demands of AI workloads. Investor’s Business Daily
