Qualcomm Dragonfly Data Center Chip to Ship in 2027, Expected to Generate Billions in Revenue This Year
N.R. Finch
Qualcomm unveiled its Dragonfly data-center chip portfolio at Investor Day, with the first HBC-architecture chip shipping in FY2027 and the new business line expected to contribute billions of dollars starting this year — its first direct assault on Nvidia's data-center stronghold.
What is Dragonfly, and why does Qualcomm think it can challenge Nvidia?
Dragonfly spans four product lines: AI accelerators, data-center CPUs, custom silicon, and connectivity chips — Qualcomm's first complete data-center portfolio.
The core differentiator is HBC — High Bandwidth Compute, a design that tightly integrates memory and compute. Qualcomm claims six times the bandwidth per watt versus conventional GPU/HBM setups, with lower power draw and total cost of ownership.
This means → Qualcomm is not trying to out-muscle Nvidia on raw GPU performance. It is redefining the value metric around "same power, six times the bandwidth."
Microsoft and Meta endorsed it — how much does that matter?
Microsoft CEO Nadella confirmed Azure will deploy Qualcomm's HBC solution in select data centers. Meta CEO Zuckerberg called it a "multi-generational partnership" and said Meta will use Qualcomm CPUs.
This means → two hyperscalers publicly backing Qualcomm answers the "does anyone want this?" question — the data-center story is not hypothetical.
But endorsement is not volume procurement. HBC chips ship in FY2027 — at least two years separate the stage appearance from real revenue contribution.
Why is the Modular acquisition more important than the hardware?
Qualcomm acquired AI software firm Modular for roughly $4 billion in an all-stock deal. Modular's core capability: write AI code once, run it across different chip architectures.
In plain terms = Nvidia's deepest moat is not the chip itself — it is CUDA, the programming ecosystem that locks developers into Nvidia hardware. Buying Modular is Qualcomm's attempt to break that lock-in.
Modular co-founder Chris Lattner built the LLVM compiler infrastructure and Apple's Swift language. About 150 employees join Qualcomm — this is a talent-and-ecosystem acquisition, not just a technology purchase.
China — Qualcomm's biggest revenue source and its biggest risk?
China accounted for 46% of Qualcomm's FY2025 revenue, mostly from smartphone chips. Qualcomm has struck a deal with ByteDance for custom AI data-center chips, structured to stay within current U.S. export-control thresholds.
But a June Commerce Department rule now requires licenses for Chinese-owned entities operating outside China — Qualcomm's data-center expansion faces regulatory scrutiny similar to Nvidia's.
This means → Qualcomm wants to scale in data centers, yet may hit the same ceiling as Nvidia with its largest incremental customer base.
Why didn't the market buy it?
Qualcomm shares fell more than 3% on Investor Day, recovering only in after-hours trading once the presentations ended.
Bank of America analyst Vivek Arya noted that Qualcomm's current data-center revenue is "almost negligible" and the company must prove it can port consumer CPU/NPU performance to more complex data-center workloads. He warned Qualcomm is entering a "fast-growing but fiercely competitive market dominated by large incumbents."
In plain terms = Wall Street's stance: great story, but your revenue today is near zero and your opponent is Nvidia — show us production numbers first.
Where is Qualcomm's window?
CEO Amon pushed back on timing concerns: "It is never too late," because "this market moves very, very fast — if you have technology leadership, there is always a place for you."
Nvidia CEO Jensen Huang responded the same day, saying Nvidia delivers the "lowest cost per token, highest token throughput, and highest revenue" — implying performance is not just about bandwidth per watt.
This reflects a shift in data-center AI competition from single hardware metrics to a hardware + software ecosystem + scaled deployment contest. Whether HBC can ship on schedule in 2027 and deliver on the six-times-bandwidth promise at scale is the make-or-break checkpoint for Qualcomm's data-center story.
Content is for reference only, not financial advice.