The transition to agentic AI would demand computing upgrades from personal devices to data centers, with Intel’s silicon powering this next wave of AI workloads, Intel CEO Lip-Bu Tan said during his keynote at COMPUTEX 2026 on June 2. Intel is orchestrating these silicon upgrades across PCs, edge devices with agentic or physical AI, data centers and intelligent centers, Tan said at the Nangang Exhibition Center Hall 2 in Taipei.
Alex Katouzian, Executive Vice President and General Manager of Intel’s Client Computing and Physical Group AI, highlighted the progress across PCs. The Intel Core Ultra Series 3 introduced in April can “transform any PC into an agentic capable platform,” Katouzian said, emphasizing that it is already being used in more than 70 light, thin laptops designed by Intel’s partners. Katouzian noted Intel’s silicon would enable customers to expand into new physical AI, which is projected to be a US$25 trillion market by 2050.
Perplexity Co-Founder and CEO Aravind Srinivas demonstrated a hybrid agentic AI system running on a laptop. Srinivas showed how this hybrid system runs smaller AI models locally on the Intel Core Series 3, while outsourcing other computing to the data center to maximize efficiency and ensure privacy.
Shifting to infrastructure, Kevork Kechichian, Executive Vice President and General Manager of Intel’s Data Center Group, discussed the silicon transformation happening in data and intelligence centers. AI inference workloads are expected to become 40% of all data center power demand, Kechichian said. “For agentic AI, the CPU orchestrates the show,” he said, introducing how the Intel Xeon 6+, unveiled at COMPUTEX yesterday, can power this shift. While generative AI placed more demand on GPUs, agentic AI workloads are much more CPU intensive. Intel is working with partners to design server racks that meet these unique enterprise needs, Tan added.
Tan explained that agentic AI is causing token usage to explode, making it critical to design compute solutions optimized for token consumption. SambaNova Co-Founder and CEO Rodrigo Liang introduced the SambaRack SN50, which uses Intel’s CPUs, Nvidia’s GPUs and SambaNova’s RDUs to orchestrate disaggregated inference. The SN50 is two to three times faster than a pure GPU stack, Liang said.
Vista Equity Partners CEO Robert Smith highlighted that the shift toward agentic AI is driving immense demand for computing and reshaping silicon software. Through its partnership with Intel, Vista is introducing the first commercially available architecture for disaggregated inference to turn existing data centers into efficient, air-cooled inference hubs, Smith said.
Tan closed with examples of how Intel’s custom silicon solutions can support all kinds of companies in meeting their goals, highlighting partnerships in biomedical engineering with Echo, biomedicine with Greenstone, energy with Hitachi and industrial automation with Siemens.
