Hari Investor Qualcomm: Satu CPU, Satu Teknologi Memori, Target US$40 Miliar

Text丨Bo Yang

Editor丨Xu Qingyang

On June 25, local time in the United States, Qualcomm held its 2026 Investor Day in New York.

Qualcomm announced a complete roadmap for data center AI infrastructure, unveiling the Dragonfly C1000 CPU, AI300 inference accelerator, and high-bandwidth computing (HBC) technology. It also revealed a multi-generational partnership with Meta, deepened collaboration with Hugging Face, and the acquisition of AI software company Modular.

Qualcomm's non-handset revenue target for fiscal year 2029 announced at the event

Financially, Qualcomm raised its non-handset revenue target for fiscal year 2029 to $40 billion, nearly double its previous long-term goal. Data center revenue for that fiscal year is expected to exceed $15 billion.

In after-hours trading, Qualcomm's stock rose as much as 16%.

01 Data center revenue to surpass $15 billion

Qualcomm CFO Akash Palkhiwala predicted during the event that by fiscal year 2027, Qualcomm's data center business would generate "billions" of dollars in revenue. By fiscal year 2029, annual revenue from this business is expected to exceed $15 billion.

From the company's overall revenue structure, by fiscal year 2029, non-handset revenue for the QCT (semiconductor) segment will reach $40 billion, compared to a long-term target of $22 billion set in 2024.

By fiscal year 2029, Qualcomm's handset business will account for only about one-third of QCT revenue.

The remaining portion will be split among several growth engines: automotive business revenue of $10 billion, and IoT business revenue exceeding $14 billion. IoT includes industrial, networking, and robotics ($8 billion), as well as personal AI and computing ($6 billion).

Profit guidance was also raised.

Analysts' average estimate for Qualcomm's adjusted earnings per share in fiscal year 2029 is $15.26, but Qualcomm's own target exceeds $18. This gap was the direct reason for the stock's after-hours surge.

CEO Cristiano Amon explained the growth logic by focusing on changes in AI usage. He believes AI is evolving from simple Q&A to agentic applications—models capable of autonomously executing multi-step tasks. Such workloads demand more low-power computing, an area where Qualcomm has accumulated expertise through mobile chips.

Amon also stated that AI computing is entering automotive, everyday electronic devices, and robotics, and chip demand in these areas will continue to "open up."

02 Dragonfly C1000 unveiled, Meta becomes first customer

The highlight of the hardware launch was the Dragonfly C1000, a CPU Qualcomm designed specifically for data centers.

Based on a custom-designed Oryon core, the Dragonfly C1000 adopts a multi-chiplet architecture, integrating over 250 cores running at frequencies above 5GHz. Qualcomm's performance tests show its per-watt performance is more than double that of existing server CPU competitors.

The Dragonfly C1000 supports PCIe Gen 7 and CXL connectivity, uses low-power memory technology for its memory system, and includes RAS features like built-in ECC, fault isolation, and error recovery. Cooling solutions are compatible with both air and liquid cooling, with racks compliant with the OCP ORv3 standard.

Rack configurations equipped with the Dragonfly C1000 were also announced: featuring 43TB of DRAM, with samples expected in fiscal year 2026.

Qualcomm has planned three specific directions for this CPU:

First, agentic CPUs, targeting high-throughput agent orchestration and low-latency interactive AI tasks.

Second, general-purpose CPUs, balancing two requirements: achieving optimal TCO (total cost of ownership) performance when running first-party workloads, and maximizing vCPU (virtual central processing unit) performance for third-party elastic usage.

Third, AI head node CPUs, which handle host processing with low overhead to allow XPUs to run at full capacity during generative AI computations.

What truly gave the Dragonfly C1000 weight was Meta's endorsement.

Qualcomm announced a "multi-year, multi-generational" agreement under which Meta will use the Dragonfly C1000 for its next-generation server clusters, with chips expected to go into production in the second half of 2028. Subsequent CPU iterations are also within the collaboration scope.

Qualcomm CFO Palkhiwala stated that through phone chips and other existing products, Qualcomm already does business with nearly all hyperscale companies. "This is not a newly established relationship." This implies Meta is likely not the only negotiation target, and more customers may still be in discussions.

Addressing external doubts about whether Qualcomm is too late to enter the data center market, CEO Amon responded: "When people ask if it's too late to enter the data center now, you should think about scale and execution capabilities, engineering capabilities, or operations and supply chain."

His point is that the large-scale system engineering capabilities Qualcomm accumulated during the mobile era remain effective in this market.

03 AI accelerator plus HBC to break the "memory wall"

Beyond CPUs, Qualcomm also updated its AI accelerator roadmap.

Following the previously released AI200 and AI250, the AI300 inference accelerator was introduced at this Investor Day. The three products iterate on an annual cadence.

The core logic of this platform is "decoupled rack-level AI inference." Tony Pialis, Executive Vice President and General Manager of Qualcomm's Data Center business, explained that agent workloads require collaboration between CPUs, AI accelerators, and connectivity technologies, rather than relying on a single chip. What Qualcomm is doing now is integrating compute, AI, memory, and connectivity into a unified rack-level platform.

Within this platform, memory issues are unavoidable, and Qualcomm's solution is high-bandwidth computing (HBC).

This is a technology designed to break the "memory wall." The memory wall refers to the bandwidth bottleneck of data transfer between processors and memory in AI computing. HBC uses 3D stacking silicon technology to tightly integrate computing units with memory, following a near-memory computing approach.

Qualcomm provided several data points to illustrate HBC's potential.

The AI250 equipped with HBC Gen 1 achieves an effective memory bandwidth of 133 TB/s per card, an 18x improvement over the AI200 using LPDDR5X. The AI300 with HBC Gen 2 will deliver a 54x bandwidth improvement over the AI200.

Compared to current mainstream HBM (high-bandwidth memory), HBC offers 6x the bandwidth at the same power consumption. Compared to SRAM (static random-access memory), HBC provides 200x the capacity at the same power consumption.

In other words, HBC significantly increases the amount of data processed per unit of power, directly impacting data center TCO (total cost of ownership). Commercial samples of the AI250 are expected in mid-2027, while the AI300 commercial samples will wait until 2028.

Connectivity products are Qualcomm's core competency and were not absent this time. The company offers interconnect solutions ranging from die-to-die, copper, fiber, to campus-level, supporting 800G and 1.6T rates, covering scenarios from within data centers up to 20 kilometers.

More than 35 technology ecosystem companies publicly expressed support for this roadmap, including Supermicro, Lenovo, SK Hynix, Micron, Samsung SDS, and Arista.

04 Acquiring Modular, partnering with Hugging Face

Beyond hardware, Qualcomm also made intensive moves in the software ecosystem.

First, the acquisition of AI software company Modular. The consideration is approximately $3.9 billion in Qualcomm stock, expected to close in the second half of 2026, subject to regulatory approval.

Modular's core product is an open, AI-native software stack that allows models to run on different chip architectures such as CPUs, GPUs, NPUs, and custom ASICs, without requiring developers to rewrite code for each hardware platform. Modular was co-founded by Chris Lattner and others, and its platform is seen in the industry as an open alternative to Nvidia's CUDA.

Commenting on the acquisition, Amon said that as agents expand across data centers and edge devices, the industry needs a more open and modern software foundation. Qualcomm hopes this acquisition will provide customers with genuine deployment choices in diverse computing environments.

Second, expanding collaboration with Hugging Face. The partnership includes three parts:

• Bringing Hugging Face's internal and developer workloads into data centers powered by Qualcomm's Dragonfly CPUs;

• Making over 3 million open models on the Hugging Face platform directly loadable onto Qualcomm-powered devices and data center racks, simplifying the process from experimentation to deployment;

• Developing a "Hugging Face Agent" for orchestrating AI workloads in hybrid environments spanning devices and the cloud, dynamically assigning tasks based on performance, cost, and latency requirements.

Hugging Face co-founder and CEO Clément Delangue explained: "We are making it easy for our 16 million developers to run open models anywhere, from devices in their hands to full racks in data centers."

The collaboration also includes a specific arrangement. Hugging Face will provide customers using Qualcomm-powered devices or cloud systems with access to Hugging Face PRO, which includes advanced storage, compute, and collaboration features.

This step lowers the barrier for developers to build applications using open models.

05 Automotive, robotics, China

Beyond the data center mainline, Qualcomm also updated progress on other businesses.

On the automotive side, the "automotive design win pipeline" has expanded to $65 billion, and Qualcomm raised its revenue target for fiscal year 2029 to $10 billion. The demand for automotive chips is driven by the continued penetration of ADAS and autonomous driving.

The IoT business was broken down further. Industrial, networking, and robotics are listed separately, with a target revenue of $8 billion; personal AI and computing target $6 billion. Qualcomm believes that agents will trigger a new upgrade cycle for smart connected devices. The company estimates that by 2030, the total addressable market for these businesses will reach $1.7 trillion.

Regarding the Chinese market, Amon made a brief response during the event. The U.S. government currently has regulations on exporting AI-related hardware to China, but he indicated that Qualcomm would have versions of data center chips that do not trigger export restrictions. He did not elaborate on specific plans, but this statement suggests that opportunities in the Chinese market have not been shelved.

Overall, the signals from Qualcomm's Investor Day were quite comprehensive. From C1000 to HBC, from the acquisition of Modular to the partnership with Hugging Face, from the $15 billion data center target to $18 EPS—all are verifiable milestones. Customers are secured, products have sample timelines, and financial models are provided.

The next few quarters' earnings reports will be the first round of validation for Qualcomm's roadmaps.