On September 12, the PEC 2026 AI Innovators Conference and the Third Prompt Engineering Summit convened at Zhongzhi Park in Beijing. The event was jointly organized by the PEC Alliance, Zhiding Tech, SoftBlocks, and Haisheng Group, with co-hosts including AIGCLink, the LangGPT Community, Aizhidao AI Artist Creation Alliance, Silicon Valley Innovation Alliance, AI Origin Community, Haisheng Ruida, and Zhongzhi Park. More than a hundred industry associations, research institutes, communities, and prominent organizations lent their support to the gathering.
PEC stands as the premier annual gathering for AI innovators, originating from the Prompt Engineering Summit and symbolizing Pioneers, Engineers, and Creators in the AI field. This year's conference revolved around the theme "Won Token, Won World," bringing together renowned domestic experts, scholars, industry leaders, and innovation practitioners. The event reached an audience of over 3.5 million online viewers through live streams across WeChat, Weibo, PC platforms, and collaborations with more than 50 leading tech media outlets and AI communities, while 1,600 attendees participated on-site. Coverage was further amplified by 120 media organizations.
During the opening segment, PEC co-founders including Gao Fei (PEC China initiator and CEO/Editor-in-Chief of Zhiding Tech), Liu Haifeng (PEC China initiator and CEO of SoftBlocks), Zhan Bingqiang (AIGCLink initiator), Yunzhong Jiangshu (Founder of LangGPT), and Ye Liao (Co-founder of Aizhidao AI Artist Creation Alliance) delivered welcoming remarks. Gao Fei noted that this third edition centered on Token factories and the Token economy, emphasizing that the sufficiency and quality of Tokens have become pressing topics. Liu Haifeng highlighted how the industry now demands speed, stability, and effectiveness simultaneously as applications mature, requiring software architecture to bridge hardware and software capabilities for better Token deployment. Zhan Bingqiang observed that developers have entered the deep-water phase of large model applications, taking responsibility for delivery and business outcomes. Yunzhong Jiangshu suggested that AI applications are transitioning from enhancing productivity to reshaping production relations, with the focus shifting from usage to collaboration and division of labor. Ye Liao added that AI content is moving from sporadic hits toward serialized, sustained production, potentially evolving into long-term intellectual properties.
Xin Guo, Party Secretary of Dongsheng Town in Beijing's Haidian District, addressed the gathering, affirming the town's commitment to the district committee and government's strategy of building a global AI innovation hub, actively participating in the construction of the Centennial Beijing-Zhangjiakou AI Innovation Belt. He noted that Zhongzhi Park, situated at the northern end of the innovation belt, is dedicated to becoming an accelerated zone for full-stack independent AI innovation and a platform for enterprise globalization, offering accompanied, full-cycle services.
A key highlight was the opening dialogue titled "2027: What Kind of Models Do We Need?" featuring Gao Fei alongside Wei Kai (Director of the AI Research Institute at the China Academy of Information and Communications Technology), Yao Yuan (Chief Scientist for Multimodal at ModelBest), Zhang Yu (Founding Partner of Qingzhi Capital), and Liu Ziming (Assistant Professor at Tsinghua University's School of Artificial Intelligence). Wei Kai pointed out that model capability evolution outpaces technology diffusion across industries, with benchmark scores merely representing sparse sampling of model capabilities; enterprises must deepen data reserves and governance while accelerating the construction of AI-native organizations. Yao Yuan emphasized that intelligence will coexist across cloud and edge environments, with cloud pursuing the upper limits of intelligence while edge focuses on intelligence density, predicting that more intelligence will run locally. Zhang Yu cautioned that model development outpaces application adoption, advising startups not to bind themselves to a single model but to reserve interfaces for various models, suggesting that speed in model utilization itself can build competitive advantage. Liu Ziming articulated that model development follows cycles of convergence and divergence, identifying world models and automated research as the most significant future directions.
The conference proposed annual topics centered on Token factories and the Token economy based on a year of industry observation, with over twenty AI innovation leaders participating in discussions. In the themed dialogue "Token Factory: True Capabilities and Real Challenges," panelists including Liu Song (VP of PingCAP), Xie Weiyu (Chief Engine Architect at Trendseeker Technology), Zhou Libin (CMO of Yaqi Kangyuan), and Wang Chao (Solutions Director at Capital Online) agreed that Token consumption will experience explosive growth, with inference services transitioning from selling computational power to selling outcomes. They asserted that Token factory efficiency directly determines output and revenue, making large-scale domestic inference an inevitable trend.
In the dialogue "Local AI: What Does It Take to Run Models Better Locally?", Chen Zhanling (CMO of Yinpu Computing), Ran Zhongyang (Head of AI Marketing for Greater China at AMD), Liu Jingzhi (Deputy General Manager of Edge Server Products at Inspur Information), Zhang Shaofeng (Head of Strategic Accounts for Internet at Minimax), and Li Wei (Senior Manager at Dell Technologies) explored the value of local deployment from perspectives of privacy compliance, cost, and latency. They argued that device selection should consider not just computational power but also memory capacity, bandwidth, and the synergy between CPU and GPU.
In the dialogue "AI Transformation: FDE Boundaries and the Primary Driving Force," moderated by Zou Jian (Co-founder and CTO of SoftBlocks), guests including Pan Chun (Head of the Pan Chun Innovation Studio at Postal Savings Bank of China), Zhao Lina (General Manager of Dazhihui Technology), Lan Tian (Director of Architecture Governance Center at Neusoft), Zhou Liangwei (Chief Solutions Architect at NetEase Zhiyan), Mo Xin (Founder of Zhiti Yuanyu Science and Technology), and Ye Liao (Co-founder of Aizhidao) highlighted that FDE's core lies in entering the field to identify needs, validating through prototypes, delivering results, and transforming services into reusable assets. Essentially, FDE serves as a bridge connecting model capabilities with business scenarios.
In the dialogue "AI-Native: Product Opportunities After Coding," moderated by Gao Fei, participants including Qin Wenshan (Founder of Suzhou Huazong AI Technology), Li Shizun (CEO of Jisi Technology), Ma Haoting (Google Cloud AI GTM Expert), Li Jinwei (CMO of OpenDesign), Christina Chen (Director of Enterprise Accounts for Greater China at Stripe), and Liu Geng (Founder of Elser.AI) discussed AI Coding subscription loops, AI livestreaming, and Agent commercialization. They concluded that AI products are shifting from selling seats to charging based on results and usage, with the true moat lying in the ability to integrate deeply into user workflows and deliver certain returns.
The conference also witnessed the release of three major outcomes: the "2026 Global Token Industry Panorama Map," the "2026 Global Artificial Intelligence Outlook Report," and the launch of the "FDE Acceleration Program and Capability Map Research." Gao Fei emphasized that the PEC conference aims to thoroughly address the year's most critical topics through community co-construction, enabling attendees to calibrate their judgments on AI trends for the coming year.
In the afternoon, three parallel forums continued the discussions on AI technology implementation. The AI Creation Pioneer Dialogue, led by Aizhidao, brought together AI directors, artists, and designers to exchange ideas on AI content creation, film industrialization, and super-individual practices. The FDE Hands-On Camp, spearheaded by AIGCLink, explored team collaboration, reusable asset accumulation, and delivery risk boundaries. The Harness Workshop, hosted by the LangGPT Community, demonstrated methodologies for workflow construction, tool integration, permission management, and outcome evaluation.
A special edition of "Origin TALK" featured a dedicated session titled "Going Global 2027: How AI Rewrites Enterprise Internationalization?" Additionally, the exhibition area showcased AI industry chain enterprises including Haisheng Investment, Google Cloud, Amazon Global Selling, NetEase Zhiyan, AMD, Schneider Electric, ASUS, Dell, Unity, and China Telecom, displaying the energy, chips, infrastructure, models, and applications driving AI, allowing attendees to experience firsthand the value of AI technology in industrial implementation.