On 22 to 23 September, Intel held its ‘Same Core, Far Intelligence’ technology and ecosystem conference in Suzhou, putting the two main AI deployment lines on stage: industry AI, and physical AI for robots.
Wang Jingjia, general manager of Intel China’s industry and solutions business, set the tone by citing a 22 September forecast that by 2030 about 80 per cent of token consumption will happen at the edge. Dan Rodriguez, vice president and general manager of Intel’s edge-computing group, argued AI is moving from understanding information to understanding the world and acting in it, with agentic AI and physical AI converging, driven by a closed loop of model, simulation, compute, hardware and field data.
Industry AI: keep the agent inside the fence
The trust barrier, not the technology barrier, now blocks industry AI. Firms will let an agent in but only with read-only rights. Intel’s answer is the Intel Agent Box, an edge-agent compute scheme built as ‘software-hardware integrated, local-first, hybrid-scheduled’, from a third-generation Core Ultra entry tier needing no discrete GPU, through Arc Pro B60 or B70 mid tiers, to a Xeon-plus-Arc-Pro flagship for full private deployment. It bundles storage encryption, sensitive-data identification, tiered permissions and audit trails, with a hybrid routing that splits tasks between local and cloud. The Open Edge Platform now has more than 60 supply partners and 125 optimised ISV components.
Physical AI: brain and cerebellum on one chip
Gao Song, vice president and head of physical AI in China, opened with the ancient yarn of Yan Shi’s automaton, then cited that roughly 80 per cent of the world’s robots are made in China with about 94 per cent annual growth, yet most still run fixed stations with fixed moves. Intel’s fix is a third-generation Core Ultra part fusing brain and cerebellum: the GPU block, new Xe architecture, gives up to 120 TOPS INT8 for model inference, environment understanding and planning; the NPU holds 50 TOPS INT8 always-on for sensing; the CPU keeps task-scheduling latency under 10 microseconds and whole-body motion jitter under 10 microseconds.
A joint embodied-AI dev kit with Cogo Tech and Digital China ships on the same Core Ultra with BSP, BIOS, ROS2 middleware and optimised reference code preloaded, reaching mass production in one to two months and fitting the torso space of humanoid, quadruped and wheeled robots. The humanoid ‘Xiao Ying’ this year performed real grasping under varying light on a VLA model.
Partners showed results: a PAZ Interconnect AI EDA appliance; a MAXHUB medical terminal in Shanghai’s Fourth People’s Hospital lifting record and order efficiency by 20 to 60 per cent; Sunmi retail edge; Qijuz industrial process-world-model control. The through-line is that edge AI should become reusable infrastructure, not one-off projects, with both answers sharing one processor generation and one toolchain.
Editor’s note: This is an adapted translation of the original OFweek Robotics report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-09/ART-8321200-8120-30704817.html.