Skip to content
News Product launch

Qualcomm unveils Dragonfly C1000 data center CPU with Meta as first customer and acquires Modular for $3.92 billion

· by Pondero Newsdesk

The short version

Qualcomm announced the Dragonfly C1000, a 250-core server CPU targeting agentic AI workloads, with Meta signed as its anchor customer and a $3.92 billion deal to acquire AI software company Modular.

Qualcomm unveils Dragonfly C1000 data center CPU with Meta as first customer and acquires Modular for $3.92 billion

At its 2026 Investor Day on June 24, Qualcomm disclosed a 250-core server CPU built for agentic AI, signed Meta as its anchor data center customer, and agreed to buy AI software company Modular in a deal that Reuters valued at $3.92 billion. Qualcomm shares rose 15% in extended trading after the company also nearly doubled its non-handset revenue target for fiscal 2029 to $40 billion, up from a prior forecast of $22 billion.

What

The Dragonfly C1000 is a multi-chiplet data center CPU built around Qualcomm's custom Oryon cores. Per StorageReview's Investor Day coverage, the chip carries more than 250 cores, runs above 5 GHz, and supports PCIe Gen 7 at over 2 TB/s alongside CXL memory disaggregation. The company claims more than 2x better performance per watt than current competitive server CPUs, though that figure is per Qualcomm's own published specifications and has not been independently verified. Commercial availability is targeted for 2028.

Qualcomm positions the C1000 around three use cases: an agentic CPU optimized for high-throughput orchestration and low-latency interactive AI, a general-purpose server CPU targeting performance-per-TCO, and an AI head-node CPU designed to maximize accelerator utilization over high-speed interconnects.

Meta signed a multi-generation agreement to use the C1000 in its next-generation server fleet, per the same StorageReview report. More than 35 ecosystem partners also expressed support, including Arista, Lenovo, Micron Technology, Samsung SDS, SK Hynix America, Supermicro, and VAST Data.

The Modular acquisition is the software half of the data center push. Modular's platform lets AI applications run across CPU, GPU, NPU, and custom ASIC architectures without requiring developers to rewrite code for each target. CEO Cristiano Amon characterized it as equivalent to Nvidia's CUDA software stack, a characterization that is per Qualcomm and is not an independently validated comparison. TechEchelon reported the deal was valued by Reuters at $3.92 billion; Qualcomm did not disclose financial terms directly. The deal is expected to close in the second half of 2026, subject to regulatory approvals.

Why it matters

CPU workloads for AI agents are the opening that Qualcomm is targeting. Inference on large GPU clusters is expensive and power-hungry, and agentic pipelines spend considerable compute on orchestration, context management, and sequential reasoning, tasks that scale-out CPU farms can handle at lower cost per token. CFO Akash Palkhiwala noted at the event that "there really isn't enough supply, and multiple players are needed" in the CPU market. If Qualcomm executes on its 2028 shipment timeline, Meta will serve as the public validation that determines whether the C1000 attracts further hyperscaler orders.

The Modular acquisition is the piece most relevant to software teams. Modular's stack removes the need to re-optimize models each time the target silicon changes. If the acquisition closes and the integration holds, it becomes a practical alternative to Nvidia's CUDA lock-in for workloads that operators want to run on non-GPU silicon, including on Qualcomm's own Dragonfly inference accelerators. Qualcomm also announced the Dragonfly AI300 at the event, a third-generation rack-scale inference accelerator targeting 4x to 8x higher memory bandwidth per watt compared to existing GPU architectures, per Qualcomm's own projections, with commercial sampling expected in 2028.

AI tool operators and ML platform engineers watching this should note the 2028 window. Nothing ships in the near term, but the Modular deal is actionable sooner: Modular's existing developer tools are available now, and Qualcomm plans to make Modular's AI components accessible through the Hugging Face ecosystem.

What to watch next

The Modular acquisition needs regulatory clearance to close, with a target of H2 2026. That timeline is the first gate. The second is whether any hyperscaler beyond Meta signs on to the C1000 before the 2028 shipment date, since Qualcomm disclosed it secured two custom silicon deals with unnamed hyperscalers. Qualcomm set a data center revenue target of $15 billion for fiscal 2029, and how far short of that it runs will be the measure of whether the bet paid off.

Sources