With the rapid development of edge computing and agentic AI, enabling physical robots to "think" in the real world has become a key focus for various manufacturers. However, in the past, performing complex vision, language, and motion control often relied on bulky, power-consuming, and expensive GPUs. Now, Intel is using its latest integrated CPU, GPU, and NPU...Core Ultra Series 3 processorThis successfully disrupted the traditional architecture.
Including Sensory AI, which created the "Ella" robot barista, many of the world's leading robot developers have abandoned GPU designs and fully switched to the more cost-effective single Intel chip architecture. This technology will also shine at COMPUTEX 2026 in June this year.
Say goodbye to cost pain points: Integrated chips achieve truly commercial ROI
To enable a robot to successfully accept orders, calculate various beverage recipes, and move its robotic arm safely, the traditional system architecture typically consisted of an Intel CPU paired with a high-power GPU for acceleration.
Keith Tan, founder and CEO of Sensory AI, points out that the biggest fatal flaw of this traditional architecture is "cost." For a coffee shop owner who wants to squeeze a profit from a $5 latte, the cost of a single graphics card may even exceed the total cost of the entire robot system and other hardware.
To address the issue of excessively high total cost of ownership (TCO), Sensory AI decided to completely migrate the underlying architecture of the Ella barista to the Intel Core Ultra Series 3 computing platform. By integrating the CPU, high-performance built-in GPU, and NPU dedicated to continuous visual AI computing onto a single chip, the robot no longer needs to rely on massive, gaming-grade computing units to perform inference tasks on the terminal. This not only significantly reduces system heat dissipation and maintenance costs but also makes large-scale commercial deployment of physical robots possible.
COMPUTEX 2026 Live Demonstration: One Chip, Multiple Uses for Agent Collaboration
At the upcoming COMPUTEX 2026 Taipei International Computer Show, the Ella robot, fully equipped with Intel architecture, will be showcased. It will not only achieve a high production capacity of 200 drinks per hour without connecting to a remote server, but will also demonstrate for the first time its heterogeneous computing capabilities, simultaneously running three dedicated "AI agents."
• Avatar Agent:Responsible for handling interactions and communication with customers.
• Ella Agent (Operating Agent):Responsible for learning and analyzing the business operation models of the store level.
• Guardian Agent:Responsible for performing high-level reasoning regarding the system's health status.
In the event of an emergency such as a stuck cup, the Guardian Agent will immediately deduce a repair solution and hand it over to the system to instruct a robotic arm to resolve the problem. Simultaneously, the Avatar Agent will proactively reassure the customer. These three agents operate in parallel on the most suitable computing units within the Intel chip, demonstrating extremely high edge computing efficiency.
Global Ecosystem Response: From Industrial Automation to Healthcare
In addition to Sensory AI, startups and robotics pioneers around the world are also actively incorporating the Intel Core Ultra Series 3 computing platform into their next-generation products:
• Trossen Robotics (USA):They focus on robotic arms for automation in the food service and manufacturing industries. Their Chief Solutions Architect, Marc Dostie, stated that Intel chips offer extremely high CPU performance and data throughput comparable to existing competitors (such as NVIDIA Jetson systems), coupled with the broad developer base of the x86 architecture, making it a highly attractive development platform.
• Circulus (South Korea):They are dedicated to developing next-generation physical AI and humanoid robots (such as the social companion robot Pibo). They utilize Intel chips to achieve zero-latency, privacy-first edge AI computing, ensuring data security and autonomous operation even when the robot is completely offline.
• Oversonic Robotics (Italy):RoBee, a humanoid and centaur robot designed specifically for manufacturing and healthcare settings (guiding patients through rehabilitation), currently only uses a dedicated GPU during the "training" phase. The actual inference work deployed on the terminal devices is handled entirely by the Intel Core Ultra Series 3 computing platform built into the robot, performing real-time facial recognition, gesture understanding, and environmental analysis.
Analysis: Moving Edge AI from "Demonstration" to "Practical Application"
Intel's strategy of promoting the application of Core Ultra Series 3 (codename Panther Lake) in the field of robotics shows that the battlefield of AI hardware is extending from high-power-consuming GPUs in cloud training to terminal devices that place more emphasis on power efficiency and deployment cost.
For most commercial robots that have already completed model training in the laboratory, what they need in the real world is not monstrous GPUs with excessive computing power, but a single-chip solution that can stably handle vision, voice, and electromechanical control, while maintaining a reasonable cost and preventing overheating. Intel is trying to prove to the industry that the future robotics industry is not just about pursuing infinitely amplified computing power, but about being smart and lightweight enough to integrate into real-world business operations.
