Qualcomm has unveiled an ambitious roadmap to enter the AI data center market, introducing its new Dragonfly portfolio of CPUs, AI accelerators, connectivity technologies, and custom silicon solutions designed for the emerging era of agentic AI. The announcement marks Qualcomm’s most comprehensive push into data center infrastructure to date, extending its presence beyond smartphones, PCs, and edge computing into one of the fastest-growing segments of the technology industry. The company also revealed a multi-year, multi-generation partnership with Meta, which plans to deploy Qualcomm’s upcoming Dragonfly C1000 data center CPU in its next-generation server fleet.
Dragonfly CPU Targets Agentic AI Workloads
At the center of Qualcomm’s new strategy is the Dragonfly CPU portfolio, purpose-built for agentic AI orchestration, general-purpose cloud workloads, and AI head-node processing. The platform features custom Qualcomm Oryon CPU cores optimized for frequencies exceeding 5 GHz and is designed to deliver high throughput alongside strong per-core performance. Qualcomm said the architecture uses a chiplet-based design with more than 250 cores, enabling large-scale deployment across hyperscale environments. The company estimates the platform could deliver more than twice the performance per watt of competing server CPU offerings based on current specifications. Qualcomm also emphasized lower total cost of ownership (TCO), citing improved infrastructure utilization and reduced operating expenses. To support next-generation AI infrastructure, the CPUs will offer more than 2 TB/s of PCIe Gen 7 connectivity, CXL support for memory disaggregation, and compatibility with both air-cooled and liquid-cooled deployments. Commercial availability is expected in 2028.
Qualcomm Introduces High Bandwidth Compute Technology
Alongside the CPU roadmap, Qualcomm introduced its new High Bandwidth Compute (HBC) architecture, a near-memory computing technology designed to address one of AI’s biggest challenges: data movement. The company said HBC combines compute and memory in a 3D-stacked silicon design, enabling significantly higher bandwidth and efficiency than traditional memory architectures. According to Qualcomm, HBC Gen 1, integrated into the upcoming AI250 accelerator, will deliver 133 TB/s of memory bandwidth per card. Future HBC Gen 2 technology, planned for the AI300 platform, is expected to further increase bandwidth while improving energy efficiency. Qualcomm claims HBC can deliver six times higher bandwidth per watt compared to conventional high-bandwidth memory technologies and significantly improve memory capacity efficiency for AI inference workloads. Commercial sampling of HBC Gen 1 is scheduled for mid-2027.
AI300 Platform Aims at Large-Scale AI Inference
Qualcomm also unveiled the Dragonfly AI300 platform, its third-generation rack-scale AI inference system. The AI300 platform integrates HBC Gen 2 technology and is designed for large language models, multimodal AI applications, and agentic AI deployments. The company expects the platform to offer industry-leading memory capacity and bandwidth for high-throughput, low-latency inference. Qualcomm estimates AI300 could deliver four to eight times better performance per watt than existing GPU-based architectures in memory-intensive workloads. The platform will support scaling through Ultra Accelerator Link (UALink), Ethernet Scale-Up Networking (ESUN), and both optical and copper interconnect technologies. Commercial sampling is expected in 2028.
Meta Partnership Signals Hyperscale Validation
A key part of Qualcomm’s announcement is its new partnership with Meta. Under the agreement, Qualcomm’s Dragonfly C1000 CPU is expected to power portions of Meta’s future server infrastructure. The deal provides Qualcomm with a major hyperscale customer as it attempts to establish itself in a market traditionally dominated by Intel, AMD, and increasingly Nvidia. The collaboration highlights growing demand for power-efficient server processors as AI workloads expand across hyperscale data centers.
Building a Full AI Infrastructure Ecosystem
Beyond processors and accelerators, Qualcomm outlined plans to provide custom silicon solutions tailored to hyperscaler and cloud infrastructure requirements. The company said it will offer end-to-end co-design capabilities covering silicon, systems, packaging, and software integration. Qualcomm also introduced a connectivity portfolio spanning die-to-die links, optical networking, copper interconnects, and campus-scale infrastructure supporting 800G and 1.6T networking. More than 35 technology companies have joined Qualcomm’s ecosystem initiative, including Foxconn, Lenovo, Quanta, Supermicro, Samsung SDS, Micron, SK hynix, Arista Networks, and GIGABYTE.
Qualcomm Expands Beyond Mobile Into AI Infrastructure
The Dragonfly announcement represents Qualcomm’s largest effort yet to become a significant player in AI infrastructure. As demand for AI inference, agentic computing, and hyperscale data center capacity accelerates, Qualcomm is positioning itself with a full-stack portfolio that spans CPUs, AI accelerators, memory architectures, networking, and custom silicon. While commercial deployments remain several years away, the partnership with Meta and the launch of Dragonfly indicate Qualcomm intends to compete directly in the rapidly expanding AI data center market. With commercial sampling beginning in 2027 and product availability targeted for 2028, Qualcomm is laying the groundwork for a long-term role in next-generation AI infrastructure.
