Agents are taking over the next phase of AI. When a task no longer requires human step-by-step instructions, but is autonomously decomposed, planned, tool-invoked, and result-checked by AI, the structure of computing power demand changes fundamentally. The decomposition and scheduling tasks must be handled by the CPU, while the GPU is only responsible for invoking models for inference. In data centers, the CPU-to-GPU ratio is shifting from 1:4 or even 1:8 closer to 1:1.
AMD (Advanced Micro Devices) CEO Dr. Lisa Su regards AI as the most important technology in the past 50 years, and AMD is redefining its role in the AI era with a completely new narrative logic. In 2026, AMD's intensive product releases include: the world's first 2nm GPU Instinct MI455X, the sixth-generation EPYC Venice server CPU, the rack-scale AI system Helios, and the ROCm.AI developer platform. This is more like a clear signal sent by AMD: the computing power landscape of AI is being redrawn, and AMD aims to be the pen holder of this new map.
Recently, in an exclusive interview with Semiconductor Industry Insights, Ji Zhaohui, Vice President of Marketing, AMD Greater China, pointed to a clearer profile of edge AI: "Agent tasks require executing code, querying databases, invoking tools, and coordinating a large number of parallel steps, which is forming a new category of AI workloads."
01. From AI PC to Agent Computer: AMD Redefines Edge AI
"An Agent Computer is not an upgraded version of an AI PC, but a completely new computing category," Ji Zhaohui defined AMD's new direction strongly promoted this year in the interview. In March 2026, AMD officially named the Agent Computer category, positioning it as a new species in the AI era, an edge device designed specifically for agent operation, rather than a traditional computer designed for human-computer interaction.
The difference between the two seems subtle but is fundamental: traditional PCs are built around keyboards, mice, and displays, with the core being humans operating machines; whereas Agent Computers are built around computing power scale, memory bandwidth, and concurrency capabilities, with the core being machines working autonomously. Users only need to issue commands via WeChat or voice, and the Agent Computer can autonomously decompose tasks, invoke tools, and complete execution. Ji Zhaohui summarized the essence of this shift in one sentence: "Agents are the key to turning AI from a chatbot into a productive force."
The hardware foundation supporting the landing of this new category is the AMD Ryzen AI Max+ 395 processor, featuring a 16-core, 32-thread CPU, 128GB of unified memory, and 96GB that can be allocated to the GPU as dedicated VRAM. In May this year, AMD further launched the Ryzen AI Halo developer platform, which can smoothly run large models with up to 200 billion parameters; the recently launched Ryzen AI Max PRO 400 series is among the world's first x86 client inference platforms capable of locally running 300 billion parameter models. Ji Zhaohui further explained: "The task of traditional computers is to assist humans in working, but the future enterprise form will undergo fundamental changes. We will see an enterprise with approximately 1,000 employees and 10,000 agents working together."
In such a scenario, where to run agents becomes a key issue: the cloud is certainly one of the options, but in application scenarios that require compliance, are sensitive to data security and sovereignty, and have demands for personalization, the importance of edge AI stands out. Based on this, Ji Zhaohui judges that the market size of Agent Computers may even surpass that of traditional PCs.
02. From Dialogue to Productivity: The China Speed of Agent Implementation
If hardware is AMD's trump card, then application implementation is its exam question. 2026 is regarded by the industry as the first year of AI agents, but the public's perception of what agents actually look like and what they can do remains vague. In the interview, Ji Zhaohui shared multiple real local cases in China, outlining a picture of a "one-person company" becoming a reality.
In the medical field, AMD's ecosystem partner EnviXion AI developed an edge doctor assistant based on the Ryzen AI Max+ 395 platform: after locally deploying the GPT-OSS-120B large model, it was fed with textbooks from Harvard Medical School and 30 years of clinical experience from chief physicians in China's top-tier hospitals, forming a multi-agent consultation platform. This system scored over 90 in relevant exams. Deploying it in community hospitals or school clinics is equivalent to equipping primary care doctors with a chief physician-level real-time consultation partner;
In the investment research field, the intelligent investment advisory system of AI startup "Creative Evolution" AInvestor has 12 built-in AI roles, responsible for research, background checks, financial analysis, report writing, and other tasks. One person becomes an investment research team. It is currently being applied and verified in institutions such as Zhongguancun Capital and Yanzhi Fund. Ji Zhaohui called this model the prototype of a "one-person company."
In the industrial field, CNC metal processing practitioners used to suffer losses from taking orders because they could not accurately calculate the process and labor costs of different machine tools. Now, by feeding all the parameters and process data of CNC machine tools into the Agent Computer, once a job is accepted and the drawings are loaded, the machine automatically calculates a precise quote; in the bidding field, the intelligent bidding solution developed by AMD ecosystem partner Cyber IoT based on the Ryzen AI Max+ 395 platform covers the entire process from bid information collection, qualification review, bid document decomposition, bid document writing, to submission and reuse. The traditional model, which originally required 3 people to work intensively for a week, is instantly enhanced to be completed by just one person in half a day.
The common feature of these cases is that they all run on the edge, solve specific problems in real business scenarios, and are being replicated at scale. Based on the Ryzen AI Max+ 395 platform, solutions covering more than 20 vertical industries such as finance and taxation, healthcare, and education have been launched, with multiple scenarios entering implementation pilots.
Behind this set of data is the unique vitality of China's AI application ecosystem: the proportion of Chinese developers among global AI developers is extremely high, and the speed at which Chinese enterprises transform AI into productivity far exceeds other markets. Ji Zhaohui specifically mentioned that the Chinese market is the first in the world to implement Agent Computers at scale, "China's AI developers are very fast in developing AI applications, and Chinese enterprises are very fast in adopting AI. These two points are very prominent."
03. Ecosystem Building and In-Depth Cooperation with Universities
Besides the industry, there is another main thread in the story of AMD's ecosystem building in China, which is the deep integration with universities. Ji Zhaohui mentioned in the interview that AMD has participated in the China Graduate Electronic Design Competition for two consecutive years. In 2026, the competition topics will be upgraded to two major directions: first, edge AI agents and innovative applications in vertical industries based on the Ryzen AI Max+ 395 processor; second, intelligent computing empowering scientific research projects based on the software ecosystem and the Radeon platform.
The newly added bonus points this year clearly lean towards "AI empowering key industry workflows." This is not only a technical competition but also an industry connection. In addition to the competition, AMD is also promoting the Spring Rain Plan, building AI practical training classrooms in schools, jointly developing courses, and establishing university AI innovative application demonstration stations. In June this year, the "AI+ Innovative Application and Practical Training Center" was established at Xidian University, and in July, it was also launched at Beijing Information Science and Technology University.
Ji Zhaohui stated that in the past, doing an AI practical training project and running a large model with hundreds of billions of parameters required hardware costs of several hundred thousand CNY. The Agent Computer lowers this threshold to tens of thousands of CNY, allowing schools to carry out AI practical training at one-tenth of the cost. This is a hardcore path to promoting the universalization of AI education.
04. Thirty Years of Deep Cultivation: Empowering Developers in the Chinese Market
AMD's positioning in the global landscape is undergoing a critical change. In her speech in Shanghai in May this year, Dr. Lisa Su frankly stated: "China has the most dynamic AI ecosystem in the world and is a place that truly understands open innovation." AMD has been deeply cultivated in China for over 30 years. But the changes go beyond this. In the past, Dr. Lisa Su's visits to China were mainly to visit customers. This year, the AMD AI Developer Day landed in China for the first time, focusing on developers.
From "visiting customers" to "empowering developers," this is a leap in strategic positioning. Ji Zhaohui mentioned a detail in the interview: China's top developers spend hundreds of thousands of CNY a year on API calls for complex coding tasks and agent tasks. If data center-class foundation models can be run locally through edge devices, costs can be significantly saved. AMD is launching a public AI developer cloud for Chinese AI developers, which will be open for free. Behind this shift from hardware sales to platform empowerment is a deep understanding of the particularities of the Chinese market: here, there are not only the richest application scenarios and the fastest iteration speed in the world but also the most powerful hardware industry cluster globally.
It is precisely this integrated software and hardware industry ecosystem that makes AMD regard China as the main battlefield for edge AI implementation. According to IDC's forecast, by 2027, the usage of AI agents by the Global 2000 enterprises will increase tenfold. Dr. Lisa Su also predicted that the AI accelerator market size will reach 1.4 trillion USD by 2030. From enduring pressure and lying low with a market capitalization of less than 3 billion USD, to a market capitalization approaching 700 billion USD, and from an industry follower to one of the core players in global AI infrastructure, AMD has completed a thorough self-reinvention in just over a decade.
In China, the story of this reinvention is far from over. As Agent Computers begin to enter thousands of industries, as AMD brings open source into university laboratories, and as every developer can embrace AI at one-tenth of the cost, AMD is paving a path to the next decade on this most dynamic land in the world. In the booming Chinese AI ecosystem, we see more innovative forces and more strivers co-existing with AI!