Qualcomm Investor Day revealed High Bandwidth Compute (HBC)—stacked compute/DRAM architecture claiming 6x bandwidth-per-watt of HBM. New Dragonfly C1000 CPU; Meta announced as first data center customer.

I attended Qualcomm Investor Day 2026 in New York City, the event Qualcomm positioned as the venue for relaunching its data center offerings. Earlier today, Qualcomm announced a major acquisition of Modular to strengthen its AI story, though the company's ambitions extend beyond the data center.

Christiano Amon, Qualcomm's CEO, opened by outlining three dimensions to Qualcomm's future, with rack-scale data center infrastructure as the first priority. He highlighted Qualcomm's substantial IP footprint and noted that the company consumes over one million leading-edge wafers annually.

Tony Pialis, Qualcomm's head of data center, framed the challenge ahead: agentic AI will reshape infrastructure, requiring new compute architectures. Qualcomm outlined a strategic progression—first, qualified connectivity solutions from Alphawave; next, custom silicon; then a new AI accelerator; and finally, a new CPU for agentic AI workloads. The company identified a fundamental constraint: as models grow larger, performance becomes limited by memory footprints and bandwidth.

Qualcomm introduced High Bandwidth Compute (HBC) as a key innovation. Rather than stacking HBM over a CoWoS interposer, HBC places memory directly on top of compute tiles—approximately 768GB of LPDDR memory paired with a base die that functions as an accelerator. The company claims HBC achieves superior performance per watt by eliminating the power overhead of data movement across an interposer, while matching HBM in capacity and SRAM performance through proximity to the CPU or accelerator. Smaller HBC variants are planned for automotive and client devices.

Satya Nadella, in a recorded message, spoke to the partnership from PCs through AI agents, highlighting HBC's role in the data center. Qualcomm indicated HBC will arrive with the AI250 generation in 2027, with HBC Gen2 (AI300) to follow, integrating fabrics like ESUN and UALink.

Tim Davis, co-founder of Modular (which Qualcomm recently acquired), presented the company's software stack, designed to compete with CUDA, Dynamo, and Triton. The strategy enables the use of Modular software with heterogeneous compute environments—a critical capability given Qualcomm's plan to operate alongside other processors and accelerators.

The Qualcomm Dragonfly C1000 CPU marks a significant data center entry. This chiplet design features 250+ cores running at 5GHz, PCIe Gen7, CXL, LPDDR memory, optional HBC attachment, and enterprise RAS features. While 250-core counts will be commonplace by its 2028 launch window—matching Intel's Xeon 6+ Clearwater Forest and AMD's EPYC Turin dense configurations—the C1000's 5GHz clock speed differentiates it; competitors are not operating at these speeds. This enables Qualcomm's claim of 2x better performance per watt. The memory flexibility, including optional HBC attachment, further distinguishes the design.

Qualcomm announced that Mark Zuckerberg, speaking on behalf of Meta, has committed to a multi-generational agreement to use Qualcomm processors, marking a significant hyperscaler win for the company's Arm-based AGI CPU.

The company emphasized connectivity as a cornerstone of its data center strategy, with a portfolio spanning die-to-die connections, co-packaged optics, PAM4 electrical and optical SerDes, and QAM16 coherent optics. The roadmap positions Qualcomm to compete directly with established players like Broadcom and Marvell.

Qualcomm outlined its strategic pillars as: CPUs, AI accelerators, custom silicon (with a $115 billion TAM), and connectivity. The company presented multiple engagement models for customers seeking custom silicon solutions.

Tareq Amin, CEO of Humain, discussed his company's long-standing partnership with Qualcomm as a closing partner announcement.

In parallel, Qualcomm introduced the Arduino Ventuno Q platform, powered by Qualcomm's Dragonwing IQ-8275 processor paired with an STM32H5 microcontroller. The platform delivers 40 TOPS of performance, 16GB of RAM, 64GB eMMC storage, and integrated Wi-Fi and Bluetooth, with availability planned for Amazon later this summer.

Financially, Qualcomm significantly revised its projections. Compared to forecasts made 18 months ago for FY29, the company updated its revenue target from $22 billion to $40 billion, with the data center emerging as the primary growth driver. Qualcomm already records data center revenue through connectivity in FY26. Looking forward, two hyperscaler customers are projected to each exceed $1 billion in revenue during FY27, when AI accelerator and CPU products launch. The company targets capturing over 5% of a $1 trillion-plus TAM within 5 to 7 years. Automotive revenue is also forecast to grow substantially as Qualcomm increases silicon coverage in vehicles. CEO Christiano Amon noted that all server CPUs currently produced are selling and expressed confidence that the C1000 will prove highly competitive even if the market environment intensifies by its 2028 launch.

The Dragonfly C1000 appears positioned primarily toward a limited set of large hyperscaler customers rather than the broader enterprise segment—a deliberate departure from Qualcomm's Centriq-era approach. The dual appearance of Meta alongside both Arm and Qualcomm for server CPU announcements raises questions about Arm's future role and competition from AMD and Intel. The Modular acquisition represents a strategic reset for Qualcomm's AI story. Unlike the AI100 generation, which struggles with a challenging software stack despite shipping in new Dell laptops, the combined Modular and Qualcomm offering promises new hardware, a unified software framework, and integrated connectivity—all designed initially for heterogeneous environments alongside competing processors and accelerators.