The conference, themed "Token Factory, Intelligence Ignites the Future," featured three major panels: "Intelligent Computing Infrastructure," "Intelligent Computing Ecosystem," and "Intelligent Computing Industry Linkage," delving deeply into the future development of intelligent computing. As computing power becomes the "new foundational energy" of the digital economy, Tokens emerge as the core "currency" of the intelligent era. However, how do Tokens drive industrial development? How do they reshape the value of local industrial ecosystems? These cutting-edge topics urgently require clarification and answers. On May 14th, the 2026 Intelligent Computing Infrastructure Innovation Conference & Xiaoshan District Chip-Machine Community Industry Ecosystem Co-construction Exchange officially commenced in Hangzhou's Xiaoshan District. Guided by the Xiaoshan District People's Government, the conference was hosted by the Zhejiang University Computer Innovation Technology Research Institute and the Zhejiang University Shanghai Institute for Advanced Study. Co-organizers included Cyzone, YunJian Information, SuperLink, Zheyin Huijin, and the Chip-Machine Community. Centered on the theme "Token Factory, Intelligence Ignites the Future," the conference structured discussions around three key areas: "Intelligent Computing Infrastructure," "Intelligent Computing Ecosystem," and "Intelligent Computing Industry Linkage," facilitating in-depth exploration of intelligent computing's future trajectory. Hundreds of representatives from government, academia, global technology firms, domestic computing power manufacturers, investment institutions, and innovative enterprises gathered. Xu Bin, Deputy Secretary of the Party Working Committee and Director of the Management Committee of Xianghu National Tourist Resort, attended and delivered opening remarks. Lou Hualiang, Vice President of Zhejiang University Shanghai Institute for Advanced Study and Secretary of the Party Committee of Zhejiang University Software College and Ningbo International Science and Technology Innovation Center, spoke on behalf of the hosts. Xiaoshan's Opening Act: Unveiling a New Token Ecosystem Map Hangzhou's Xiaoshan District is rapidly emerging as a new hub for intelligent computing industry agglomeration. With the Chip-Machine Community as its nucleus, the local area is accelerating the construction of a complete intelligent computing industry chain covering chip design, computing power clusters, and scenario implementation. The inaugural 2026 Intelligent Computing Infrastructure Innovation Conference's choice of Xiaoshan as its venue signifies a gathering of computing power momentum and the opening of a new intelligent chapter atop this highland of the intelligent computing industry. In his address, Lou Hualiang stated that computing power is becoming the core strategic resource driving digital economy and artificial intelligence development, a vital national asset in the new era, and a shared mission for national strategy and industrial upgrading. Over the more than five years since its establishment, the Zhejiang University Shanghai Institute for Advanced Study, centered on its "computing+" core layout, has attracted over 100 high-level scholars and undertaken more than 70 national, provincial, and ministerial-level research projects. Moving forward, the Institute will continue to focus on key areas such as intelligent computing theory, high-performance algorithms, and computing power system architecture, solidifying a self-reliant and controllable intelligent computing foundation, deepening integration with industries, and promoting computing power services across all sectors. This conference serves as a feast of ideas on cutting-edge technology and an important platform for industrial collaboration and resource integration, showcasing China's innovative vitality and collaborative prospects in the field of intelligent computing infrastructure. Xu Bin expressed that artificial intelligence is not merely an industry but an era. Currently, Xiaoshan is positioning the AI industry as its core breakthrough, planning three major industrial communities and building four highlands. Relying on a solid industrial foundation and an open innovation ecosystem, it is steadily advancing on the intelligent computing track. Xianghu is committed to building Hangzhou's next trillion-yuan industrial cluster, focusing on the "Chip-Machine Community" AI industrial cluster and the "China Vision Valley" visual manufacturing industrial cluster to cultivate new quality productive forces. Xianghu sincerely invites all sectors to establish core projects, R&D headquarters, and business roots in Xianghu, pledging the most open attitude, most favorable policies, and highest quality services to empower all parties mutually and advance together, striving to build Hangzhou's computing power hardware manufacturing agglomeration. At the 2026 conference, Xianghu officially released its AI industry space plan and the "Chip-Machine Community" urban design. Teng Yizheng, Member of the Party Working Committee and Deputy Director of the Management Committee of Xianghu National Tourist Resort, introduced that the "Chip-Machine Community" follows the "one axis, two zones, multiple plots" layout along Shidai Avenue, with a core area of 1,155 acres planning 2 million square meters of industrial space. Policy-wise, the community launched the nation's first special policy for visual intelligence. Capital-wise, it established a 3 billion yuan Vision Valley Fund and a 1 billion yuan Xianghu Venture Capital Fund, supplemented by a hundred-billion-yuan National Strategic Emerging Industry Fund. In the future, the Chip-Machine Community will focus on industrial agglomeration, supporting facilities, technological empowerment, and industry-academia-research-application integration to create a benchmark AI community attractive to young talent. While computing power is rapidly expanding, the collaborative pathways between various links of the industry chain remain unclear. Who produces Tokens? Who schedules computing power? Who implements applications? The entire ecosystem lacks a clear corporate map. To address this, Nan Lixin, Founder and CEO of Cyzone, officially released the "2026 Token Factory Ecosystem Map" at the conference, categorizing and inventorying relevant players in the field, from emerging enterprises to tech giants. Data shows China's average daily Token calls surged from 100 billion in 2024 to 100 trillion by the end of 2025, reaching 140 trillion by March 2026—a growth exceeding a thousandfold in two years. Nan Lixin pointed out that although application-side demand has surpassed that of the United States, a significant gap remains between China and the U.S. in terms of intelligent computing center scale and intelligent computing power capacity. She segmented the Token Factory ecosystem into: the Energy Layer, Computing Power Layer, Model Layer, Data Layer, Scheduling & Operations Layer, Agent Application Layer, and external ecological value circles like policy. Currently, investment in the ecosystem layers is severely insufficient, with the Scheduling & Operations and Data Service layers particularly undervalued. Nan Lixin emphasized that the core of Sino-U.S. competition lies not in applications but in infrastructure. She called for collaborative efforts from government, capital, large enterprises, and entrepreneurs to improve the domestic chip and software ecosystem and seize the strategic window for AI infrastructure construction. The conference also hosted a cooperation signing ceremony between "Zhejiang University Shanghai Institute for Advanced Study and SuperLink." The Institute and SuperLink—a Xiaoshan-based computing power network infrastructure provider—signed an agreement to co-establish a "Computing Power Optimization Joint Innovation Laboratory." This aims to leverage both parties' strengths in scientific research and industrial practice to promote talent cultivation, technological innovation, and result commercialization. The laboratory will focus on key directions in computing power optimization, particularly research on SONiC switch system development, containerized architecture optimization, ecosystem integration, and commercial adaptation. Furthermore, OWS Global and Guoxun Information reached a global computing power development strategic cooperation at the conference. OWS focuses on AI infrastructure, intelligent computing power networks, and enterprise-level AI application ecosystems, providing clients with end-to-end solutions from underlying computing power to upper-layer AI business implementation. Partnering with Guoxun Information aims to further enhance capabilities across cross-border computing power efficient circulation, local computing power elastic supply, and scenario-based AI rapid deployment, providing solid support for deepening digital economy innovation and the intelligent transformation and digital upgrading of the real economy. Charting a New Blueprint for Intelligent Computing Infrastructure: Academia-Industry-Research Dialogue on Frontier Developments At the conference, numerous experts, scholars, and industry figures shared their insights and reflections on computing power. These forward-looking perspectives intersected from different dimensions, outlining the technological direction and future landscape of the intelligent computing industry. Wei Chengkun, Doctoral Supervisor at Zhejiang University's College of Computer Science and Technology and a "Hundred Talents Program" researcher, delivered a keynote report titled "Computing Power Network Convergence Trends and Frontier Technology Outlook." He systematically analyzed the infrastructure transformation driven by AI applications shifting from model training to large-scale inference Agent deployment. The report noted that inference demand is rapidly surpassing training demand. AI network design goals are shifting from the training-era AllReduce collective communication throughput to inference-era point-to-point low-latency transmission and efficient KV Cache management. Citing前沿进展 like Google TPU architecture, Ethernet inference networks, KV Cache compression, CXL memory pooling, and open interconnect standards, the report posited that training and inference infrastructures will further diverge. High-performance Ethernet is expected to become the unified foundation for inference networks, while the storage, transmission, compression, and scheduling of KV Cache will become core constraints for next-generation computing power network architecture design. In a presentation titled "Intel Empowers AI Infrastructure Reshaping," Hou Zhiqiang, Chief Scientist of Intel China's Marketing Group, pointed out that intelligent computing power is shifting from GPU-centric to CPU-centric. Computing power clusters have evolved into three parallel paradigms: GPU clusters, CPU clusters for Agentic AI, and storage clusters. Intel's Xeon 6 processors already support AI AMX acceleration engines, 52 types of RAS capabilities across six categories, DDR5 frequencies up to 6400, and MRDIMM support up to 8800. At the "New Era of Heterogeneous Computing: AMD Defines the Future of the Trillion-Dollar Computing Power Market" session, Ivan Li, AMD Global Vice President, stated that Agentic AI is driving CPUs to reclaim their core position in computing power. In past computing architectures, CPUs only handled basic scheduling, with GPUs bearing 90% of the computational load. Currently, the core proportion of CPUs in data preprocessing and task flow orchestration is continuously rising. Driven strongly by Agentic AI, the server CPU sector could achieve a 35% CAGR, with the overall market size exceeding $100 billion, marking CPUs' official return to the core seat of the computing power arena. AMD has consistently focused on open-source ecosystem布局, achieving full-dimensional openness from hardware architecture to software stack. Currently, AMD's server CPU market share has surpassed 40% and continues its rapid growth trajectory. Wang Yu, Solution Architect at Volcano Engine, shared practical applications of ArkClaw and Volcano Ark in industrial intelligence. In a keynote dialogue moderated by Fan Zhimin, Founding Partner of Zheyin Huijin, participants including Gong Yue (CEO of SuperLink), Tang Shunwu (Vice President of YunJian Information), Ma Yanping (Data Director of Xiaoshan District Data Resource Administration Bureau), and Yu Hui (Deputy Director of the High-Performance Computing Facility Center at Zhejiang Lab) engaged in an in-depth discussion on "Token Reconstructing Computing Power, Frontier of Intelligent Computing Infrastructure." Participants unanimously agreed that AI computing power is accelerating its shift from "training-centric" to "inference-centric," with Token production efficiency becoming a core metric. Gong Yue noted that as computing power cluster scales expand, network importance becomes increasingly prominent, with current bottlenecks mainly集中在 GPU interconnect and scheduling latency. The future should focus on high-bandwidth, low-latency, lossless networks to ensure each GPU can continuously and efficiently produce Tokens, while坚持全栈自研 to guarantee system stability, security, and controllability. Gong further stated that based on judgments about AI cluster network evolution, SuperLink launched the CE11-128N data center switch for large-scale intelligent computing scenarios. The product utilizes Spectrum-4 fifth-generation ASIC chips, features 128×400G full line-rate high-density ports, and is compatible with SONiC systems and ConnectX-7 and BlueField-3 ecosystems, better meeting large-scale data交换 and low-latency interconnect demands in high-density GPU clusters. Ma Yanping added that leveraging local industrial advantages, Xiaoshan will empower AI industry development from three aspects. First, coordinating data resources, linking local优势 industries like textiles and auto parts to create industry datasets and promote government-enterprise data integration and application. Second, introducing special subsidy policies for data vouchers covering data circulation, usage, and processing to助力 the "Model-Data Resonance" initiative. Third, implementing data classification and分级 management to create a "sterile" data usage environment and筑牢 data security防线 across the entire chain. Tang Shunwu proposed evaluating computing power systems using a comprehensive "Six Forces Model": computing power, storage power, transport power, power supply, liquid cooling capability, and model power. Two major directions for future computing power infrastructure are: promoting liquid cooling technology普及 and establishing unified open architectures. Currently, YunJian provides services to nearly 20 domestic GPU manufacturers. Yu Hui pointed out that domestic computing power ecosystem construction requires four stages: building computing power, utilizing computing power effectively, training models well, and model decoupling. Currently, we are in the intermediate stages. Looking ahead 3-5 years, China will encounter a critical window for constructing larger-scale clusters exceeding 10,000 cards. During this period, real large-scale business scenarios are needed to drive the domestic software stack to完成 a "unified, open, mature" leap and achieve system-level hardware-software协同 optimization, thereby truly realizing the value of hardware investment. Strengthening the Domestic Computing Power Foundation and Connecting Computing Power Networks The innovation conference also featured two breakout sessions focusing on "Computing Power Infrastructure Construction" and "Computing Power Empowering Industrial Scenarios." Wang Lei, Vice President of Ecosystem at Moore Threads, delivered a keynote speech titled "Domestic GPGPU Building Intelligent Computing Infrastructure," introducing Moore Threads' general-purpose GPGPU product practices in supporting intelligent computing facilities. Additionally, Moore Threads adheres to an open-source strategy, collaborating with numerous domestic universities on course development, summer camps, and innovation competitions, and participates in organizations like CCF to promote domestic AI ecosystem construction,致力于 building an "Android ecosystem" in the AI field. Ding Yunfan, Vice President of AI Framework Architecture at Biren Technology, in his talk "Building Domestic 10K-Card Clusters Based on Optical Interconnect GPU Super-Nodes," stated that large model implementation in the Agent era exhibits "three super" characteristics: super-large parameters, super-long context, and requiring super-nodes to address computing power challenges. Solving computing power issues must address three dimensions: hardware cluster computing power, software effective computing power, and heterogeneous aggregated computing power. Biren will continue deepening the domestic intelligent computing foundation, promoting heterogeneous协同 ecosystem construction, and providing autonomous, efficient, open solutions for Agent-era computing power infrastructure. Fu Weifeng, Founder and CEO of OWS Global, in his presentation "Computing Power Laying the Foundation for Token's Native Nature: Supercomputing Clusters Driving New AI Industry Paradigms," stated that computing power and Tokens have become core production factors in the AI industry. Fu predicted that in the next 3-5 years, model parameters might reach百万亿级别, Tokens will become genuine production factors and tradable commodities, and cross-regional computing power collaboration will become常态. He called for deep cooperation with chip manufacturers, software vendors, application developers, and academic institutions to jointly build an open, prosperous AI industry ecosystem. Zhang Renwen, Pre-sales Director at Vodafone China, in "New Paradigm for Cross-Border Computing Power: How Global Networks Support Efficient AI Computing Power Flow," stated that the logic for Chinese companies going global has shifted from early AI applications to infrastructure and computing power出海. Physical network divides, geopolitical compliance, and high overseas operational costs pose three major challenges. Coco Cheng, Vice President of Business Development (Greater China) at SUNeVision, presented "Data Center Evolution in the Computing Power Era: The Upgrade Path from IDC to AI Computing Power Hub." She指出 that 2026 will become a key inflection point for AI infrastructure investment, with AI inference surpassing training to become the main data center workload,预计 to account for two-thirds of total AI computation. This shift意味着: low latency becomes the core SLA, interconnect ecosystems become strategic assets, and computing power must migrate closer to the user side. Breaking New Ground in Intelligent Computing Infrastructure: Discussing Token and Industry Synergy Within the Intelligent Computing Ecosystem panel, how computing power transitions from technology to application became the focal point. Cai Wenkuan, Founder of Dreaming Aerospace, shared a keynote "New Paradigm for Low-Altitude Intelligence: Computing Power Integration and Aerospace Synergy in the eVTOL Era." He noted that eVTOLs are not merely aircraft but intelligent agents and edge intelligent computing terminals. He emphasized the eVTOL industry is in an early stage similar to electric vehicles, with large-scale application still requiring 10-15 years. Future competition will shift from单机性能 to system intelligence, requiring safe collision avoidance and dynamic scheduling for thousands of协同 aircraft. Qin Lan, Founder and Chairman of United Imaging Intelligence, shared the latest breakthroughs in AI and robotics technology in neurointervention. She noted that traditional neurointervention surgery属于 L4超高难度 procedures heavily reliant on top experts' experience and skill,难以实现 standardization and普及. Qin introduced that her team, with over 20 years in cerebrovascular disease, successfully developed the world's first and currently唯一 catheter shaping robot. Li Xiaofei, Founder and CEO of Shenpu Intelligent, in "Computing Power as Foundation, Intelligence-Embodiment Integration: The Evolutionary Path of Embodied Robots Driven by Large Models," stated that embodied robots directly entering homes face core technical challenges like generalization ability and interactive understanding. He believes directly relying on a "brute force for奇迹" training model has low success rates. A more feasible path is accumulating real data and闭环 capabilities from类家庭 scenarios (e.g., hotels, elderly care) and gradually切入 home scenarios. In "Agent-Driven Protein Design and New Paradigm for Biological Computation," Liu Hao, Co-founder and CTO of Tianwu Technology, introduced that Tianwu, founded in September 2021, is among China's earliest AI protein design companies. By late 2022, it had developed and applied a protein large model. In 2026, Tianwu released the world's first "conversational dry-wet结合" protein design agent—MatwingsVenus™ (Xiaowu™)—opening AI protein design capabilities to individuals. He welcomed各界 to register and experience their agent platform to jointly promote the落地 of new biological computation paradigms. Ye Sen, Co-founder and Head of Scientific Affairs at Yaosu Technology, in "Organ-on-a-Chip × AI: The New Operating System for Life Sciences,"指出 that new drug R&D requires $1 billion and a decade, with a success rate below 10%. The天然鸿沟 in translating traditional animal实验数据 to humans is a core reason for low success rates. Yaosu致力于 extending 3D Bio-Intelligence from底层 data capture to intelligent data interpretation, promoting innovation in the大健康赛道. Zhang Xiaobo, Co-founder of Songying Technology, in "Intelligent Computing Driving Physical Simulation: Building the Digital Cornerstone for Embodied Intelligence Evolution," stated that AI is accelerating from the virtual world into the物理世界. Unlike autonomous driving, embodied intelligence emphasizes强交互 with the environment (e.g., grasping,搬运), requiring higher-dimensional data—not just vision but encompassing depth, LiDAR, tactile sensors,力学 data, and other multidimensional information. Wen Deliang, Founder and CEO of Red Bear AI, in "Memory Reconstructs Intelligence, Computing Power Drives Implementation—New Pathways for Enterprise AI,"指出 that current large models have serious memory issues like knowledge forgetting, multi-Agent协同 difficulties, and poor multimodal memory coherence, directly impacting business解决率. He believes AI memory有望 become the core infrastructure for next-generation AI. Xue Haoxin, Vice President of LuChen Technology, in "Innovative Computing Power Architecture Empowers the AI Era: Large Model Inference and Industrial Computing Power Integration," stated that LuChen始终致力于 obtaining more computing power from有限硬件 through algorithm optimization, computing power architecture innovation, and hardware potential挖掘, liberating AI productivity. The company has traversed a development path of "technology → system → service." In the roundtable "Computing Power Logic Under the 'Everything AI' Investment Trend," moderated by Zhu Zan, Partner at ShiHui Law Firm, participants including Jiang Chun (Managing Partner at Pu Hua Capital), Qiao Yuting (General Manager of OPPo Xunxing Investment), Wang Meng (Partner at Junshan Capital), Wang Xin (Partner at Oriental Fortune Capital), and Zhang Qian (Founder of Skyline Technology Investment) engaged in in-depth discussion on AI investment布局, computing power logic, and evaluation criteria. Investors noted that with AI demand exploding, the world needs more "large-volume, cost-effective" computing power to achieve AI democratization, allowing more people to enjoy AI红利. Over the next five years, the infrastructure layer will still occupy the most significant value share across the entire AI industry chain. China is a major application country with the world's most comprehensive千行万业 industrial chain, offering大量 opportunities for AI落地 across all sectors. In the "Intelligent Computing Industry Ecosystem Linkage Session," over 30 AI innovative enterprises conducted one-on-one对接 with active AI investment institutions including Ivy Capital, Chizi Fund, DT Capital, Volcano Rock Capital, Leading New Frontier, Oufang Angel, Skyline Technology Investment, Tongchuang Weiye, Vertex Ventures, Leap Capital, Zhejiang International Trade Asset Management, Zhemin Investment, Zheshang Venture Capital, and智明星通. Computing power lays the foundation, industry moves toward innovation. Today's conference is a gathering of ideas and a starting point for action. In the future, we will continue using the Token Factory as a nexus to link academic wisdom, industry leaders, and capital力量,助力 computing power enterprises showcase value and对接 resources, building合作桥梁 for local governments, investment institutions, and large enterprises, and jointly推动 China's intelligent computing industry toward a new stage of协同发展.
Comments