- Qualcomm Dragonfly AI200 AI accelerator rack is the first of multiple releases planned by the chip designer as it aims to score wins in the data center segment
- The upcoming Dragonfly AI250 accelerator leverages its proprietary High Bandwidth Compute (HBC) to offer a theoretical 18x the amount of bandwidth of its sibling
- Qualcomm’s push comes amid an increasingly lucrative datacenter market grappling with memory shortages
It is no secret that the modern AI server ecosystem is dominated by Nvidia in most countries, even as China increasingly leans towards Huawei as its own home-grown provider for similar solutions.
Qualcomm may not be one of the first companies that come to mind when you think about AI data centers or the chips housed inside them, with many investors feeling it has missed the boat altogether in the server segment.
Qualcomm’s recent Investor Day 2026 event was a reminder that it is not only still in the game but also has ambitions to carve out a large piece of an ever-increasing pie by taking a different route than most of its HBM-leveraging competitors.
An alternate ecosystem to Nvidia’s industry standards?
Much of Qualcomm’s Investor Day event focused on its plans to become a sizable player in the AI data center market, which is currently dominated by OEMs deploying a mix of Nvidia and AMD accelerators alongside custom silicon (ASIC) offerings from Google, Meta, Microsoft, and even Amazon‘s AWS.
It aims to do so by differentiating itself from the competition, relying on its own area of expertise to carve out an edge: efficient Low-Power Double Data Rate (LPDDR) memory stacked in a 3D array above its AI accelerators to drive the next generation of AI inference workloads.
The near-memory compute architecture isn’t exactly a new play in a market teeming with similar approaches, but the numbers are hard to argue with when it comes to Qualcomm’s offerings.
Qualcomm’s upcoming Dragonfly AI200 rack delivers 43 TB of LPDDR5X capacity and 414 TB/s of memory bandwidth per rack, built from accelerator cards each carrying 768 GB of LPDDR5X, which makes it an interesting offering, but much of the focus that hyperscalers will have will be on its Dragonfly AI250 sibling that incorporates High Bandwidth Compute (HBC) under the hood.
While it offers the same memory capacity per rack, its ability to leverage memory at up to 18x its sibling’s bandwidth yields a theoretical peak memory bandwidth of up to 7.4 PB/s per rack, a far cry from the AI200’s 0.4 PB/s.
The Dragonfly is positioned as an inference-centric accelerator for a reason; however, HBM is still better suited to certain tasks, such as training models rather than inference, making it the memory of choice for Nvidia’s Blackwell and upcoming Rubin GPUs, as well as AMD’s Instinct offerings.
With that being said, Qualcomm’s solution is intriguing, even if the numbers are for specific use cases and its ability to court Hyperscaler giants such as Microsoft and Meta tends to indicate that it has a potential win, at least on paper, as AI datacenters continue to increase focus on inference-centric solutions to deploy their increasingly complex models to wider audiences.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
