Qualcomm unveils Dragonfly C1000: 2x performance per watt, targets 2028 datacenter production

Qualcomm has a new datacenter flagship, and it is arriving with numbers that feel like a direct challenge to incumbents: Dragonfly C1000 claims 2x better performance per watt and 30 percent more speed than competitors' processors. During Qualcomm’s Investor Day presentation on Wednesday, datacenter EVP and GM Tony Pialis went straight at the big question executives will be asking in conference rooms and board decks: are you too late to an arena dominated by well established players such as Nvidia?

Pialis’ answer was not subtle. “When the company turns its attention to solve a new problem, we revolutionize the solution and push our way to the forefront,” he said, adding that Qualcomm is “doing in datacenter.” Then he backed it with a formal unveiling of Dragonfly, which he described as a platform rather than just a CPU. The core of that platform is the Dragonfly compute platform, featuring the C1000 CPU that Pialis showed off, and he positioned Dragonfly as Qualcomm’s path from being dominant in mobile, PC, and automotive into servers where the battle is typically measured in power, throughput, and system-level bottlenecks.

The hardware story starts with compute. Pialis said the C1000 CPU cores are based on Qualcomm’s custom Oryon architecture and will operate at more than 5 GHz. Qualcomm is also leaning into chiplet-based design, featuring more than 250 cores. That matters because datacenter buyers do not only buy chips, they buy systems that can scale across racks without turning energy costs into a runaway line item. Qualcomm is explicitly targeting lower total cost of ownership and better performance per watt than rival platforms, and it is asking customers to believe those metrics will hold up under real workloads.

Then Qualcomm tries to win a more specific fight that has been haunting AI datacenter economics: memory bottlenecks. Pialis described a technology Qualcomm calls “High-Bandwidth Compute” (HBC). In plain terms, Qualcomm is arguing that it can reduce the classic penalty of moving data back and forth between compute and memory by integrating compute and memory more closely. Qualcomm’s pitch is that HBC combines an XPU beneath a DRAM stack, claiming SRAM-like performance advantages inside a high-bandwidth memory package, with the added goal of reducing data movement and improving performance per watt.

Qualcomm also gave a timeline and workload targeting that investors and enterprise IT planners will care about. The Dragonfly C1000 processors are expected to enter production in the second half of 2028. Qualcomm plans multiple C1000 variants aimed at agentic AI, general-purpose computing, and AI head-node workloads. But Pialis was careful to frame CPUs as only one part of the pivot. He said Qualcomm is targeting three other datacenter segments alongside new CPUs: connectivity, custom silicon designed for individual customers, and AI accelerators.

On accelerators, Qualcomm said its AI accelerators will use HBC technology to address memory bottlenecks in AI workloads, aligning the compute-plus-memory concept across the stack. On connectivity, Pialis detailed new technologies intended to support cluster scaling and reduce distance constraints inside and across datacenter clusters. He said Qualcomm wants to enable new distances in cluster-to-cluster optical connectivity up to 20 kilometers using new QAM16 coherent-lite optical modules, claiming Qualcomm can scale from millimeter technology to tens of kilometers. That is an important second-order point for operators: if your interconnect road map is credible, your systems can avoid expensive redesigns as workloads grow.

Qualcomm also described custom silicon that goes beyond “a chip.” Pialis said custom silicon would involve making bespoke chips for Qualcomm’s “highest tier of customer,” who needs someone to design and fabricate AI and cloud DC CPUs from end to end. And to avoid getting stuck at the hardware layer, Qualcomm is buying software capabilities. On Wednesday, Qualcomm announced an agreement to acquire Modular, a company that develops AI software stacks, to flesh out the software side of its Dragonfly endeavors. Qualcomm said the acquisition will give it access to hundreds of billions in new market space.

The Dragonfly pitch arrived in front of high-profile guest appearances too. Microsoft CEO Satya Nadella and Meta CEO Mark Zuckerberg made guest appearances during Investor Day. Qualcomm said Microsoft plans to use HBC-based AI accelerators, while Meta separately announced plans to deploy Dragonfly C1000 CPUs under a multi-generation agreement. For decision-makers at other cloud providers and enterprise buyers, that pairing matters: it signals that Qualcomm is pursuing deployments through both a major enterprise platform and a large consumer-facing ecosystem.

Strategically, Qualcomm is trying to avoid the trap many challengers fall into. A new CPU can look great in isolation, then lose in the ecosystem once software, networking, and memory behavior show up in procurement. Qualcomm’s Dragonfly approach, as presented by Pialis, is to tackle multiple system-level constraints at once: performance per watt, memory bottlenecks through HBC, scaling via optical modules, and software through Modular. In a datacenter market with established competitors, that combination is the bet. And for executives watching their own roadmap, the question is no longer just whether Qualcomm can ship silicon in 2028. It is whether Qualcomm can make its whole platform story coherent enough that buyers see a reduced total cost of ownership, not a new source of integration risk.