Intel and Qualcomm unveil AI inference hardware at Computex 2026

At Computex 2026 Intel detailed its Crescent Island GPU, built on the Xe3P architecture for AI inference. The card uses up to 480 GB of LPDDR5X memory instead of HBM, targets high memory capacity over bandwidth, and runs at a 350 W TDP suitable for air‑cooled servers. Sampling is planned for the second half of 2026, with revenue expected from 2027. Intel also confirmed the 2027 launch of its next‑generation Xeon processors, Diamond Rapids, on the 18A‑P node with up to 192 cores and PCIe 6.0 support.

Qualcomm announced a new data‑center brand, Dragonfly, covering server CPUs, AI accelerators and custom ASICs. The AI200 accelerator is slated for 2026, the AI250 for 2027, and a first ASIC is expected by year‑end, rumored to target ByteDance. The brand signals Qualcomm’s intent to compete across the full compute continuum. Qualcomm’s shares fell about 9 % on the news before partially recovering the next day.

Other exhibitors highlighted the broader AI hardware shift. MINIX introduced AI‑focused PCs built on NVIDIA, Intel and AMD platforms, while memory and storage makers Biwin and Klevv showcased higher‑capacity RAM and faster PCIe Gen5 SSDs aimed at meeting the growing data demands of AI, gaming and content creation. The shows underscored how AI is reshaping the entire PC hardware ecosystem.