Qualcomm announced a full data center platform called Dragonfly at its AI Summit on June 24, 2026, marking the company’s biggest push yet into the server chip market dominated by Nvidia. The portfolio covers AI inference accelerators, custom silicon designs, and rack-scale systems built for what Qualcomm calls the “agentic AI era.”
What the Dragonfly Portfolio Includes
The Dragonfly lineup spans three product categories. The Dragonfly AI100 is a dedicated inference accelerator designed for running large language models at scale. The Dragonfly Custom line offers hyperscalers the option to design their own AI chips using Qualcomm’s IP blocks. And the Dragonfly Rack systems bundle networking, cooling, and compute into pre-configured units that can be deployed directly into existing data centers.
Qualcomm’s pitch centers on power efficiency. The company claims the AI100 delivers 3x better performance-per-watt compared to Nvidia’s H200 for inference workloads. For companies running billions of API calls per day, that efficiency gap translates to millions of dollars in annual electricity costs.
$15 Billion Revenue Forecast by 2029
The revenue projections are what sent Qualcomm’s stock up 15% on the day of the announcement. CEO Cristiano Amon told analysts that the company expects $15 billion in annual data center chip sales by 2029, nearly double its previous guidance of $8 billion. The revised forecast reflects design wins at several major cloud providers, though Qualcomm did not name specific customers.
Qualcomm has been quietly building its server chip business since acquiring Nuvia in 2021. The Nuvia team, which included former Apple silicon engineers, developed the custom CPU cores that now power Qualcomm’s laptop and server processors. Dragonfly represents the next step: a full-stack platform combining Qualcomm’s custom CPUs, GPUs, and AI accelerators with proprietary networking fabric.
How It Compares to Nvidia and AMD
Nvidia still holds roughly 80% of the AI accelerator market. AMD’s MI300X has gained traction as a cheaper alternative for some workloads. Intel’s Gaudi accelerators have struggled to gain market share despite aggressive pricing.
Qualcomm’s advantage is different. Rather than competing head-to-head on raw training performance (where Nvidia’s CUDA ecosystem is hard to beat), Dragonfly targets inference. As companies deploy trained models at scale, the bottleneck shifts from GPU compute to efficient, high-volume serving. That is the gap Qualcomm is trying to fill.
Arm-based processors have historically underperformed x86 chips in single-threaded server workloads, but AI inference is a different workload profile. It relies heavily on matrix math and memory bandwidth, areas where Qualcomm’s custom silicon can compete. The company also benefits from Arm’s lower licensing costs, which helps keep per-unit pricing below Nvidia’s data center GPUs.
Custom Silicon for Hyperscalers
The Dragonfly Custom program is particularly interesting. It gives cloud providers like Microsoft Azure, Google Cloud, or Amazon AWS the ability to specify custom configurations: different ratios of CPU cores to AI accelerators, specialized memory hierarchies, or custom interconnects. Qualcomm handles the manufacturing through TSMC, and the hyperscaler gets a chip tailored to its specific workload patterns.
This model mirrors what Arm has done with its compute subsystem licensing, but at a higher level of integration. Qualcomm is offering complete rack designs, not just chip IP.
Frequently Asked Questions
What is the Qualcomm Dragonfly chip?
Dragonfly is Qualcomm’s data center platform for AI inference, consisting of the AI100 accelerator, custom silicon options, and pre-configured rack systems.
How does Dragonfly compare to Nvidia H200?
Qualcomm claims 3x better performance-per-watt for inference workloads, though Nvidia maintains an advantage in raw compute and its CUDA software ecosystem.
When will Dragonfly chips be available?
Qualcomm says the AI100 will ship to select cloud providers in Q4 2026, with broader availability planned for early 2027.
Why is Qualcomm entering the data center market now?
The explosion of AI inference demand has created a market for efficient, high-volume serving chips, where Qualcomm believes its Arm-based architecture has a power efficiency advantage over x86 alternatives.
What companies are expected to use Dragonfly?
Qualcomm has not named specific customers, but the $15 billion revenue forecast suggests design wins at multiple major cloud providers. Arm-based server chips are already used by AWS (Graviton) and Microsoft (Cobalt).
