Qualcomm says Dragonfly C1000 server chip stays on schedule for Meta

Qualcomm's big bet on a server comeback is still on the rails. The company's CEO confirmed the Dragonfly C1000, the data center CPU built for Meta's server infrastructure, remains on schedule for delivery as planned — a small sentence with a lot riding on it.

The chip landed with an announcement on June 24, 2026, pitched at agentic AI and general-purpose server workloads. It leans on a chiplet design — multiple smaller dies stitched together instead of one giant slab of silicon — packing more than 250 cores, custom Oryon cores that can push past 5 GHz, and support for PCIe Gen 7. Qualcomm also claims the C1000 delivers more than double the performance per watt of rival server CPUs.

That efficiency number is the whole sales pitch. Qualcomm spent years squeezing power out of mobile chips, and it's now arguing that discipline travels to data centers, where electricity bills are an operational headache. The wider Dragonfly platform adds a High Bandwidth Compute architecture, with sampling expected in the first half of 2027 and full production for Meta's servers targeted for the second half of 2028.

Meta isn't a one-off buyer here. The arrangement is a multi-generation agreement, with Meta planning to use the C1000 to grow its server fleets and, according to the company, back its push toward what it calls personal superintelligence. For Qualcomm, that's the credibility play: re-entering a market you walked away from is a lot easier when the world's biggest social network is publicly attached to your roadmap.

The walkaway was in 2018, when Qualcomm sold off its Centriq server CPU business after failing to gain traction against Intel's entrenched position and a resurgent AMD Epyc line. Now the company has set an internal target of roughly $15 billion in annual data center sales by fiscal 2029 — an aggressive number for a business that spent most of the past decade at effectively zero revenue.

The battlefield looks nothing like the one Qualcomm left. AMD has taken real ground from Intel with Epyc, while Arm-based designs from Amazon's Graviton line and Ampere Computing have proved x86 isn't the only architecture that scales at volume. Qualcomm's Oryon cores are Arm-based too, putting it in the growing camp. And the power-efficiency argument lands harder in 2026 than it would have a decade ago, with data centers burning electricity at a rate that has become a genuine policy issue in multiple countries.