Beyond cloud: The future of in-store physical AI and on-device experiences
At this year’s Rakuten AI Optimism, one of the most exciting, future-looking discussions centered on the intersection of AI, robotics and hardware for physical AI and on-device AI experiences. Far from being a sci-fi fantasy, agentic AI avatars, humanoid robots, and AI devices are emerging as practical solutions to real-world challenges in retail and offline environments.
AI-powered store agents and robots
One of the retail industry’s most persistent challenges is the “ROPO” dynamic: Research Online, Purchase Offline. While 80% of shoppers in Japan research products online, a staggering 90% still finalize their purchases in physical stores. For years, these two worlds have operated in silos. Kosuke Kobayashi, AI & Data Division Program Manager at Rakuten Group, opened the panel on physical AI with a vision for a truly unified AI-powered experience.
“Historically, these two environments have operated as total strangers,” Kobayashi explained while opening the panel. “Digital gives you deep intelligence, but zero physical presence. Physical stores give you human trust, but the moment a shopper enters the aisle, that rich digital context resets to zero. That disconnect is a massive opportunity.”
To explore how Edge and physical AI can bridge this gap, Kobayashi was joined by industry leaders Ran Hu, Director of Business Development, NVIDIA; Mei May Soo, Chief AI Global Solutions Specialist, Dell Technologies; and Natalie Mao, SVP of AI and Product, Rakuten Group.
As the speakers noted, the ultimate goal of this technology is not to automate the human element out of the store, but to enhance it. “At Rakuten, we believe AI should augment human creativity,” Mao noted. The key question: “How can business leaders leverage Edge and physical AI to handle the heavy lifting, so shop floor teams are freed to deliver true omotenashi (hospitality that exceeds expectation) and build lasting trust with customers?”
The event brought these concepts to life with live demonstrations, including a conversational shopping experience via a Rakuten Shop AI Kiosk powered by Dell and NVIDIA. During the demo, Soo interacted with an agentic AI avatar on a touchscreen to plan a hiking trip in Japan. The agent demonstrated an impressive ability to listen, adapt to abrupt changes in user instructions, and maintain privacy and governance standards for a secure, local experience.
As Japan faces a shrinking workforce, robots are increasingly viewed as a vital support system. NVIDIA’s Hu invited the Unitree G1 Humanoid Robot to the stage to showcase how AI-powered robotics can serve as a friendly store clerk greeting customers, navigating the floor, and even assisting with product retrieval.
By the close of the session, the consensus was clear: AI-powered interfaces and robots hold great potential to empower human experience.
Hybrid AI: Bringing on-device intelligence to users anytime, anywhere
Another area experiencing a rapid shift is the realm of AI PCs. Here, AI tasks are migrating from exclusively cloud to a mix of desktop and cloud, effectively solving the challenges of rising token costs and connectivity. Taku Okoshi, Executive Officer and Director of the AI Services Supervisory Department at Rakuten Group, was joined onstage by Toru Matsuura, Director of Japan Personal Systems at HP Japan, and Jungo Kasai, Co-founder and CTO of Kotoba Technologies, Inc., to discuss why on-device AI is a game-changing development for both companies and users.
“By running AI directly on your machine rather than through the cloud, users gain faster, more secure, and more private experiences,” Okoshi said, highlighting the recent launch of Rakuten AI for Desktop on HP PCs in Japan. In this pioneering project, the development team overcame significant technical challenges to integrate the Rakuten AI 7B LLM into HP PCs.
Toru Matsuura shared how HP is embedding AI directly into their hardware, focusing on “always-connected” functionality and enhanced security features like Sure View. “The industry is poised for rapid growth,” Matsuura commented. “HP projects that AI PC adoption will skyrocket from 5% today to 50% by 2030, fueled by the development of efficient Small Language Models (SLMs).”
Kasai from Kotoba Technologies then used an HP AI PC onstage for a live demonstration of his company’s on-device, real-time transcription voice AI. Participants at the event could see the speed and accuracy of on-device AI and how LLM-powered devices can help create more natural, inclusive human-computer interactions, even offline.
The panelists concluded with the message that hybrid AI, combining on-device and cloud capabilities, offers new opportunities for cost efficiency while making user experiences more personal and accessible. As high-performance hardware, optimized AI models and software converge, the vision of “AI at your fingertips” is quickly becoming our new reality.
