Computex 2026: AI Goes Physical with New CPUs, Agents, and Data Center Chips

Qualcomm President and CEO Cristiano Amon used the opening keynote to declare 2026 the “Year of the Agent.” His thesis is that AI is evolving from passive chat-style prompts into autonomous systems that can interpret intent, break down tasks, coordinate data, and execute actions across phones, PCs, cars, and robots . In Amon’s framing, the smartphone is no longer the center of digital life—AI agents are .

Dragonfly was the keynote’s biggest hardware reveal: a new data center AI inference chip brand that marks Qualcomm’s formal entry into the server market . Dragonfly targets AI inference workloads across the company's envisioned “Computing Continuum,” a unified architecture that dynamically balances workloads from milliwatt-class wearables to megawatt-scale data centers . While Amon confirmed that Qualcomm is already working with hyperscalers and global partners on deployments, he withheld detailed specifications and roadmaps, telling the audience those would come at Qualcomm’s Investor Day on June 24 . The move gives Qualcomm a three-brand compute portfolio—Snapdragon for consumers, Dragonwing for industrial IoT, and now Dragonfly for the data center—allowing the company to compete across the full agentic AI infrastructure stack .

Intel bets on a CPU renaissance with Xeon 6+ and rack-scale AI

Intel’s keynote, led by CEO Lip-Bu Tan and data center chief Kevork Kevorkian, built its entire narrative around the argument that agentic AI is driving a CPU renaissance. The pitch: as AI inference and agent orchestration workloads explode, CPUs will handle a growing share of the work that GPUs alone can’t cost-effectively scale .

Xeon 6+ (Clearwater Forest): At the center of this strategy is the new Xeon 6+ processor, Intel’s first data-center CPU built on the Intel 18A process node. It packs 288 Efficient-cores and 576 MB of L3 cache, targeting cloud-native, scale-out workloads with maximum compute density and efficiency per watt . Intel claims a single rack of Xeon 6+ processors can simultaneously support up to 150,000 AI agents .

Rack-scale AI infrastructure: Intel unveiled a new rack-scale reference architecture developed with SambaNova Systems—using SambaNova SN-50 Reconfigurable Dataflow Units (RDUs)—and Foxconn, which will handle manufacturing and integration . The design pairs Intel Xeon processors with SambaNova’s RDUs to accelerate inference and agentic workloads. A single liquid-cooled 32U rack can deliver 36,864 cores at roughly 100 kW of rack power, targeting hyperscalers and large enterprises . Intel also announced a CPU-dense variant for workloads that skip GPU-style acceleration entirely .

800 Series Ethernet and Crescent Island GPU: Intel expanded its networking portfolio with the new E835 controllers and teased Crescent Island, its next-generation GPU for AI acceleration, signaling a broader attempt to compete at the full system level, not just the socket .

Edge AI goes industrial: DFI, ADLINK, and Advantech

Taiwan’s industrial computing leaders used Computex to prove that edge AI hardware is ready for real-world physical deployments. The common thread: purpose-built systems for vision, robotics, and manufacturing inference.

DFI showcased industrial-grade Edge AI platforms based on NVIDIA Jetson Orin modules, including the X6-ORN-GMSL for compact embedded vision, the X6X-ORN for outdoor environments, and the X6a-AGX for high-bandwidth multi-camera systems—all built to get computer vision into the field faster . DFI also partnered with Mobilint on a compact wall-mounted edge AI server for smart city applications, demonstrated with real-time weapon detection using Vision AI and Vision Language Models .

ADLINK exhibited under the theme “Leading Edge AI Computing” with seven application zones covering humanoid robots, unmanned vehicles, smart manufacturing, and 3D medical imaging . The standout was Moby, a general-purpose humanoid robot from Noble Machines powered by an ADLINK edge AI platform running on NVIDIA Jetson Thor, showing real-time sensor fusion and full-body control .

Advantech made edge AI its headline growth driver, announcing next-generation integrated edge AI platforms delivering at least 100 TOPS of computing performance . The company presented a three-tier architecture—Edge AI Embedded modules, AI Appliance systems, and HPC platforms—as building blocks for manufacturing, logistics, and retail AI .

Note on MINISFORUM: Searches did not return direct coverage of MINISFORUM announcements at Computex 2026. While the company is a known maker of compact edge and mini-PCs, specific products or news from this year’s show were not confirmed in available sources.

The big theme: Physical AI is the next infrastructure buildout

The keynotes and exhibitor halls at Computex 2026 made the industry’s next priority unmistakable: AI is moving off the screen and into the physical world. Organizers described the shift as a transformation “beyond the race for computing power” into real-world robotics, autonomous mobility, and intelligent industry . Intel’s rack-scale blueprints and Xeon 6+ processor, Qualcomm’s end-to-end computing continuum and Dragonfly chip brand, and the flood of edge AI systems from DFI, ADLINK, and Advantech are all pieces of the same puzzle—building the infrastructure for AI that acts, not just converses.