NVIDIA, Qualcomm, Intel and other major international manufacturers gathered at the Yiding Forum! Exploring new trends in edge AI and ecosystem collaboration

As the core AI brand of Innodisk Group, Aetina is one of the key players in promoting the implementation of AI edge implementation. Aetina is also an Intel Gold Partner, Qualcomm IoT Partner, and NVIDIA Elite Partner. It is responsible for transforming cloud AI capabilities into implementable edge solutions, and continues to deepen its industrial influence through cooperation.

In this forum, NVIDIA Senior Solution Architect Hong Yujia described that robots can be robotic arms, humanoid robots, or even self-driving cars, and manufacturing will be the first stage where the next generation of "Physical AI" will be implemented. The development of a robot requires three types of computers, namely "Training Computer", "Simulation Computer" and "Deployment Computer". The robot is trained and learned through a large amount of data, tested and trained in a simulated environment, and finally executed from the computing module installed inside the robot.

Currently, Aetina International has also officially launched Preview Kits equipped with the NVIDIA Jetson Thor platform in response to market demand, helping developers around the world to preemptively deploy next-generation robots and physical AI applications, and integrating AI computing performance, real-time control and low-latency sensing into a single platform to accelerate the implementation of smart edge computing.

Qualcomm Senior Director of Engineering, Evgeni Gousev said that in the past, the habit was human input and device response. Under the trend of AI agents, AI will serve as a new user interface (UI) and become a bridge between humans and devices. Since AI agents are autonomous in some situations and think about problems from a system perspective, they may be regarded as a kind of "quasi-autonomous driving."

Evgeni Gousev also mentioned that the current AI agent is considered L1 and L2 levels, but he believes it will develop faster than "Advanced Driving Assistance System" (ADAS). Gousev emphasized that Qualcomm has demonstrated its leading advantages in the field of self-driving and has promoted these achievements to the edge AI side, providing users with complete agent, multi-agent and fully autonomous services, and has in-depth cooperation with Innodisk and Aetina to achieve true "AI edge intelligence."

Du Weiyang, general manager of the Intel Ecosystem Development Organization, revealed that he and Innodisk are currently cooperating in many vertical fields such as smart cities, industrial automation, transportation and retail. They design system development based on Intel chips and jointly design the Arrow Island reference platform (Reference kit) for ecological partners to test in advance. This not only demonstrates the close cooperation model between the two parties, but also forms a complete core platform and peripheral product portfolio, which further becomes an important foundation for building smart city and other vertical market solutions, providing support in all aspects of the ecosystem.

Fabrizio Del Maffeo, CEO of Axelera AI, observed that today’s small AI models are more efficient and device performance is constantly improving, which means more models can be run on a single device. To popularize edge computing on a large scale, price and efficiency are one of the keys. Axelera AI uses unique technology to integrate computing into memory, and jointly designs development versions with partners such as Innodisk to simultaneously provide a complete and powerful software stack, which can effectively improve the effectiveness of edge AI introduction..

In addition to closely cooperating with major international manufacturers and building a strong ecosystem, in-depth understanding of customer needs is also one of the advantages that Innodisk Group continues to gain popularity. Su Weiquan, director of Aetina International Software Division, pointed out that "computing platform, data and diversified algorithms are the three key factors that make up an AI system." Many customers do not know exactly how much computing power they need or which processor they should choose.

"Aetina's mission is to help customers put the right computing power in the right place." Su Weiquan said that Aetina assists customers in starting from thermal design, size restrictions and application characteristics to ensure that every watt of power invested by users can maximize value. At the same time, the company also provides three core values: multi-platform hardware planning, intermediary support, solution integration and ecological co-creation to ensure that every bit of computing power is fully released.

First of all, in terms of hardware architecture, Aetina has a product line covering low-power to high-performance servers, providing deeply customized solutions such as system design and cooling solutions, and providing NVLink or PCIe Switch solutions for multi-GPU requirements. It supports SDK integration at the middleware layer (Middleware) and provides a complete software stack to help customers create a smooth and complete development environment. It also provides accurate performance verification and integrates a large number of peripheral devices to enable AI It can be seen, heard, and transmitted; finally, in terms of solution integration and ecological co-creation, whether it is edge deployment, device management, backup and restoration, Aetina also has relevant solutions, and the goal is to enable customers' AI to be implemented smoothly.

Su Weiquan also mentioned that Aetina International provides diversified and customized solutions. Together with the group's resources, there are corresponding product solutions available from the smallest modules and edge systems to the entire edge server. Aetina hopes to establish a unique AI ecosystem network to help system integrators and developers overcome the pain points and challenges of AI implementation.

Huang Yukai, vice president of Dingxin Digital Manufacturing Business Group, believes that future enterprises will be built on the concept of "one model, two cores, and three intelligences." This means that a model will include the "physical world" and the "digital world". The physical world contains smart devices such as personnel, production equipment, and logistics devices. The latter interacts with the physical world through information and algorithms. Among the three types of intelligence, the "commander" is responsible for understanding task requirements and judging the specific division of labor, which is handled by the "expert agent" and the "clone agent." The expert agent has corporate knowledge, decision-making processes and experience, and provides judgment and strategy, while the clone agent is responsible for actual execution. Through this multi-smart platform, AI agents can connect the entire enterprise to realize decision-making and execution automation of smart enterprises.

Huang Yukai said that he has cooperated with Qualcomm, Innodisk, and Aetina to release an "AI Order Assistant" that integrates the technical advantages of the three parties. This product has realized automated procurement and order management within the company, and can be preprocessed in advance, increasing operating efficiency by up to 96%, making AI agents gradually become a reality.

Chen Zhijie, business director of Guangyun Machinery, believes that when it comes to the physical AI level, the industry will face three expensive problems, namely high data collection costs, high training time and engineer costs, and high robot prices. Digital twins can complete model training without consuming expensive resources, truly achieving the "no three expensive" effect. At present, Guangyun and Innodisk Group have also deepened their cooperation to provide an edge computing platform with diversified computing power and flexible deployment, as well as the integration of AI robots and digital twin infrastructure to support the closed-loop operation of smart logistics.

At the end of the event, Wu Zhiqing, senior director of Innodisk's Intelligent Peripheral Application Division, shared that in addition to computing power, AI also requires four important core technologies: "data collection, storage, I/O, and AI software." This is a "smart loop". The first three are hardware foundations, and the last is the integration of software and hardware. Only when the four loops are completed and verified can it be truly deployed in industry application scenarios.

When it comes to the integration stage of software and hardware, Wu Zhiqing said that if the AI agent wants to move towards autonomous learning, one of the keys is to support the MCP protocol (Model Context Protocol), which is a standard language and specification that will enable different language models to be connected to each other in the future. At present, Innodisk has completed the key infrastructure of "MCP Standardized Platform" to help APIs be used across platforms and operating systems, making software and hardware integration more complete and laying the foundation for agent AI applications.

Wu Zhiqing also said that Innodisk Group’s blueprint for developing agent AI is not just a hardware combination, but a complete solution integrating software and hardware. Innodisk hopes to use these solutions to make up for the problems encountered by customers, lay a solid infrastructure and scalability, so that customers from all walks of life can focus on their own business and create real business value.