Qualcomm Lands Meta CPU Deal, Targets $15B Data Center Revenue by 2029
- Qualcomm secured a multigenerational CPU agreement with Meta for its Dragonfly C1000, a 250-core Arm server chip shipping in 2028 [1]
- Two hyperscaler customers are expected to generate at least $1 billion in combined revenue within a year, with initial shipments by year-end 2026 [2]
- The Dragonfly AI300 inference accelerator delivers 4-8x performance-per-watt over GPU-based architectures on memory bandwidth metrics, sampling in 2028 [3]
- Qualcomm is acquiring AI software startup Modular for $3.92 billion in stock to challenge Nvidia's CUDA software lock-in [4]
- Qualcomm targets over $15 billion in annual data center revenue by fiscal 2029, up from a near-zero base [2]
Qualcomm used its June 2026 Investor Day to formally enter the data center silicon race, unveiling a full-stack platform headlined by the Dragonfly C1000 server CPU and the Dragonfly AI300 inference accelerator. The centerpiece announcement: Meta has signed a multigenerational deal to deploy the C1000 in its next-generation server fleet, with Meta CEO Mark Zuckerberg appearing at the event to endorse the partnership [1][2].
The company disclosed it has now secured two hyperscale customers — one confirmed as Meta, the other unnamed — that CFO Akash Palkhiwala said will generate at least $1 billion in revenue within a year, with shipments beginning by year-end 2026 [2]. Microsoft CEO Satya Nadella also endorsed Qualcomm's High Bandwidth Compute architecture during the event but stopped short of confirming a commercial deployment [2].
Qualcomm set a revenue target of more than $15 billion annually from data center by fiscal 2029 — an aggressive trajectory for a company that exited the server chip market in 2018 and is building from a near-zero base [2]. QCOM shares rose 4.8% on the day to $206.97, pushing market capitalization to $218 billion [5].
The Hardware: C1000 and AI300
The Dragonfly C1000 is a 250-plus-core server CPU built on Qualcomm's custom Oryon architecture, with sustained frequencies above 5 GHz, PCIe Gen 7 connectivity, CXL support, and full enterprise RAS capabilities [1][3]. Qualcomm claims 2x performance-per-watt versus existing server CPUs, with the chip purpose-built for agentic AI orchestration — high-throughput sequential reasoning and context switching that GPUs handle poorly [1].
The Dragonfly AI300 is a third-generation rack-level inference accelerator supporting both air and direct-liquid cooling. It integrates Qualcomm's second-generation High Bandwidth Compute (HBC) technology, a near-memory computing architecture using 3D-stacked silicon that addresses AI's data-movement bottleneck [3]. Qualcomm projects a 54x effective bandwidth improvement over its AI200 baseline and 4-8x performance-per-watt gains over GPU-based inference setups on a memory-bandwidth-per-watt-per-card basis [3].
An intermediate product, the HBC Gen 1-based AI250, delivers 133 Tbps per card — an 18x improvement over the AI200 — and is slated for commercial sampling in mid-2027. The C1000 CPU and AI300 accelerator both target commercial availability in 2028 [3].
The Meta Deal
Meta's commitment gives Qualcomm a named anchor customer — a critical validation for a company re-entering a market dominated by AMD, Intel, and Nvidia. Zuckerberg stated at the event that Meta's goal is to 'deliver personal superintelligence to everyone' and positioned Qualcomm as a component of that effort alongside Meta's in-house MTIA silicon program [2].
Specific financial terms were not disclosed, and Meta declined to detail timing, volumes, or workloads beyond confirming the C1000 will power its next-generation servers [2]. Analyst Matt Kimball noted the Meta deal provides both validation and funding for Qualcomm's broader data center roadmap, making it easier to pursue additional cloud customers [2].
The Software Play: Modular Acquisition
Qualcomm announced an all-stock acquisition of AI software startup Modular for approximately $3.92 billion — 19.2 million shares at the company's pre-announcement closing price [4]. Modular brings the MAX inference engine and the Mojo programming language, which together allow developers to write AI inference code once and deploy it optimized across CPUs, GPUs, NPUs, and custom ASICs [4].
The deal is a direct attack on Nvidia's CUDA ecosystem, which has locked developers into Nvidia hardware through proprietary tooling. Modular's approximately 150 employees include co-founders Chris Lattner, the inventor of LLVM and the Swift programming language, and Tim Davis [4]. The acquisition is expected to close in the second half of 2026, subject to regulatory approval [4].
Building the Stack: Alphawave and Custom Silicon
The Dragonfly platform is not Qualcomm's only data center bet. The company previously acquired Alphawave Semi for $2.4 billion to bolster its connectivity and compute IP, and has disclosed a custom silicon engagement with an unnamed hyperscaler, with initial chip shipments scheduled for December 2026 [6].
Qualcomm's approach is to offer a full-stack alternative: Arm-based CPUs, inference accelerators, networking silicon, custom chip design services, and now an open software layer via Modular. CEO Cristiano Amon framed it as 'a comprehensive portfolio of solutions' rather than a single-product play [2].
Competitive Context
Qualcomm's $15 billion revenue target by FY2029 would make it a material player in a data center silicon market currently dominated by Nvidia in accelerators and AMD and Intel in server CPUs. The C1000's Arm architecture positions it alongside Amazon's Graviton and Ampere Computing's Altra as an alternative to x86, while the AI300 competes directly with Nvidia's inference-focused GPU offerings.
The timeline is ambitious. The C1000 and AI300 do not ship until 2028, and Qualcomm exited the server chip market entirely in 2018 after its Centriq 2400 failed to gain traction. The Meta deal and the unnamed second hyperscaler provide near-term revenue, but the $15 billion target requires Qualcomm to win significant share in a market where Nvidia's software ecosystem remains the primary barrier to entry — precisely the problem the Modular acquisition is designed to solve [2][4].
Further sources
The stories that matter, in one email. Free — unsubscribe anytime.