Hong Kong Cyberport is becoming a "springboard" for Chinese AI to go global.

Author | Zhou YaSource | Tech Traveler

In the southern part of Hong Kong, China, there is a well-known landmark, which is the famous local technology park—"Cyberport". Zhipu entered Cyberport in 2025 and rang the opening bell on the Hong Kong Stock Exchange a year later, becoming the "world's first large-scale model stock"; Yunji Technology's overseas revenue increased by 92.1% in just one year after its establishment in Cyberport; the "CARES Copilot" surgical video large-scale model trained by the Hong Kong Innovation Research Institute of the Chinese Academy of Sciences using Cyberport's computing power has become the most downloaded in the world. These events contain two key words: listing and going global. According to the "Global Unicorn List" released by the Hurun Research Institute in June this year, the Guangdong-Hong Kong-Macao Greater Bay Area has 80 unicorn companies, an increase of 8 from last year, accounting for 21% of the national total. Many of these 80 unicorns are already doing the same thing: using Hong Kong as a springboard for globalization. The HK$62 billion initial funding from Hong Kong Investment Management Company, known as the Hong Kong version of Temasek, has been largely allocated. The list of companies invested in includes SenseTime, Bestar Biotech, Jieyue Xingchen, PPIO, and Insil Intelligence—all mainland AI unicorns, and all poised for IPOs on the Hong Kong Stock Exchange. The capital logic of IPOs is relatively easy to understand, but going global is different. Dr. Huo Luming, Director of Business Development at Cyberport, believes there are many misunderstandings. "Going global isn't just about buying a plane ticket," Dr. Huo stated frankly in an interview with TechWalker during WAIC 2026 (World Artificial Intelligence Conference). She added that she has seen too many companies that fly to Hong Kong with products that have proven successful in mainland China, only to pack their bags and return a year later, claiming they can't find business. This statement highlights a proposition that is being re-evaluated: Chinese technology companies going global need to relearn how to do business overseas, and this time, most companies are choosing not to "go to sea" alone, but rather to "board the ship" as a team. Cyberport might just be that ship. For a long time, Hong Kong has been described as a "stepping stone" for mainland companies going global. This metaphor emphasizes distance but easily obscures the real work that takes place. Companies don't simply "jump" from Hong Kong to overseas; they need to readjust their product logic, delivery methods, data governance, contract systems, and market communication. Huo Luming pointed out that although Hong Kong and the mainland are geographically close, their laws, cultures, and market practices are different: "The rules and regulations in Hong Kong are very different from those on the mainland. It's not that if you can operate on the mainland, you can definitely operate in Hong Kong. But one advantage is that if you can operate in Hong Kong, you might be able to operate in Southeast Asia." Cyberport, for a long time, has not only catered to individual overseas expansion needs, but also to the overall expansion of industrial clusters. Dr. Zheng Songyan, CEO of Cyberport, told us, "Zhongguancun, Suzhou, Jiangsu, Guangxi, and other places have already established overseas expansion bases at Cyberport. Some large enterprises entering Hong Kong do not just register one overseas company, but hope to bring their upstream and downstream supply chains overseas as well." In simple terms, Cyberport is becoming an "overseas office" for multiple mainland cities and industrial clusters. Landing on the ground is just one part of the process. In the capital sector, Cyberport connects to a network of over 200 financing partners, introducing investors and assisting some companies in preparing for IPOs. Simultaneously, Cyberport organizes trips for companies to Southeast Asian markets, bringing local customers and investment institutions back to Hong Kong.

Cyberport CEO Zheng Songyan

Cyberport has participated in WAIC for four consecutive years. This year, it brought six companies (Firetech, Ocean Technology, Laiye Technology, Soundwave Technology, Zhifa Digital Technology, and 221b Limited) to the World Expo exhibition hall. Zheng Songyan said that participating in the exhibition is not just about brand exposure. They will take participating companies to visit peers, compare their own products with those of mainland and overseas technology companies, and also take the opportunity to understand what the market really needs. He used a very simple statement: "First let people know you, and then you can take the next step." Looking closely at the Cyberport ship, you will find that its decks are not the same, with different Chinese AI companies occupying different positions on the ship. The first deck is called computing power. "Five years ago, Hong Kong had no computing power," recalled Professor Liu Hongbin, Director of the Artificial Intelligence and Robotics Innovation Center at the Hong Kong Innovation Research Institute of the Chinese Academy of Sciences. The reason for this can be traced back to an old story in Hong Kong's AI industry—"strong in research, weak in commercialization." Hong Kong's "strong in research" refers to its strong basic research capabilities; however, its "weak in commercialization" has historical reasons, as Hong Kong historically lacked a mature science and technology innovation industry, creating a gap between university research and industry needs. Liu Hongbin believes that for transformation to become stronger, industry demand must guide universities' choices in research directions. This requires industry talent, and industry talent needs enterprises. Over the past five years, this framework has been gradually being completed. Liu Hongbin observed two changes: First, Cyberport and Hong Kong Science Park have attracted a large number of enterprises, leading to a concentration of industry talent; second, the situation changed with the establishment of the Cyberport AI Supercomputing Centre. The Cyberport AI Supercomputing Centre began operation in 2024 and currently provides 3000 PFLOPS of computing power. The Hong Kong government has also established a HK$3 billion, three-year "AI Funding Scheme" to support universities, research institutions, and enterprises in using the supercomputing centre. According to data disclosed in the 2026-2027 fiscal budget, the program has approved approximately 30 R&D applications, covering fields such as large-scale models, biomedicine, and new materials. Dr. Huo Luming stated that the program could cover up to 70% to 90% of a company's AI development costs. 3000 PFLOPS isn't a huge figure, and Huo Luming didn't shy away from this point. In the interview, she frankly stated, "Compared to supercomputing and intelligent computing clusters in mainland China, this scale may not be outstanding, but within Hong Kong's local research ecosystem, it has already changed the way some institutions obtain computing power." Huo Luming, Director of Business Development at Cyberport Healthcare is one of the most challenging and representative testing grounds for technology transfer. Hong Kong's healthcare system emphasizes safety and compliance, and the barriers to entry for new technologies into hospitals are very high. The Artificial Intelligence and Robotics Innovation Center at the Hong Kong Innovation Research Institute of the Chinese Academy of Sciences, where Liu Hongbin works, is a direct user. The center has received support from the "Artificial Intelligence Funding Scheme" for two consecutive years, focusing its research on the integration of AI and healthcare. Using this computing power, the center trained and released the surgical video model "CARES Copilot," achieving the highest download volume among similar models globally. Traditional medical device giants such as Intuitive Surgical are testing it, and some companies have begun to negotiate commercial licensing. The second layer is called standards. Inspur Cloud took a different path. How could a company with a sufficiently mature business restart its internationalization from Hong Kong? Inspur Cloud Group serves over 120 countries and regions globally, but truly implementing its AI technology overseas only began in 2023. Yin Ping, Assistant General Manager of Inspur Cloud Information Technology Co., Ltd. and Chairman of Hong Kong ICI Cloud Service Limited, told TechWalker that Inspur Cloud carefully considered its choice of starting point for internationalization, ultimately concluding: "The first stop must be Hong Kong." She explained three specific reasons for Inspur Cloud's choice of Hong Kong. First, Hong Kong's mature and standardized international legal system is the first system that mainland companies must face when going global. Secondly, Hong Kong boasts a vibrant science and technology innovation environment and attracts top talent returning from overseas. Thirdly, Hong Kong's status as a hub of international standards allows Inspur Cloud to access the forefront of global artificial intelligence standards. These three reasons, in a sense, summarize the core of the "Hong Kong path." Based on these three reasons, Inspur Cloud's internationalization is divided into three steps: First, adapting to laws and regulations; second, refining products using Hong Kong's science and technology innovation environment; and third, promoting products to Southeast Asia, the Middle East, and Europe through Hong Kong's free market. Specifically, in 2023, Inspur Cloud first established joint laboratories with Hong Kong universities to incubate technologies. Officially launched in Cyberport in 2024, Inspur Cloud incubated 26 top international technologies within a year. In 2025, it launched the Hong Kong Smart Cloud Platform in conjunction with Cyberport and released the "AI Factory". Yin Ping revealed to TechWalker that "after launching in Cyberport, Inspur Cloud also gained two unexpected benefits": first, it completed many international incubations, obtaining international patents, papers, and technology publications; second, these international achievements "feed back to the mainland," helping to upgrade existing products. This is a direction rarely mentioned in previous overseas expansion stories; globalization is bilateral, involving both "going out" and "coming in". According to Yin Ping, "We need to take another step forward in international AI standards and industry standards, not just to make our technology conform to international standards, but to launch something even higher from within those standards." Simply put, what Inspur Cloud is doing in Hong Kong is not just exporting products, but also "exporting standards." The third layer is called "scenario." Yunji Technology is a case in point; it focuses on specific questions like "whether people in this region are willing to have robots deliver things to them." This company already has mature delivery robot products in the mainland hotel market and went public in Hong Kong in October 2024, becoming the "first service robot stock." In an interview, Xie Yunpeng, Chief Development Officer of CloudMinds Technology, told TechWalker that their robots have served over 40,000 hotels worldwide, performing over 750 million services in the past year, and their products have entered more than 20 countries. However, the challenges posed by the Hong Kong market are not simply whether the robots can autonomously ride elevators, avoid obstacles, and deliver goods, but rather a different set of issues: whether hotel staff and guests will accept the robots, how to handle fire safety and property regulations, who is responsible for equipment malfunctions, and whether a service process designed for mainland hotels can adapt to the management methods of overseas hotels. "Going global is not about finding overseas demands to suit our products, but about how we proactively integrate into overseas markets," he said. Therefore, CloudMinds established a joint laboratory at Hong Kong Polytechnic University, leveraging the latter's expertise in hotel management and the international hospitality industry to gain a deeper understanding of overseas customer needs. Simultaneously, it entered Cyberport, establishing case studies in government, hotel, and public service scenarios. In just one year after its Cyberport launch, CloudMinds accumulated case studies in over 20 countries, with overseas revenue increasing by 92.1% year-on-year in 2025. This figure cannot be simply attributed to Hong Kong, but it illustrates a value offered by the Hong Kong market: companies can complete a quasi-overseas validation process at a relatively controllable cost, in an environment closer to overseas cultural paradigms. The three decks—computing power, standards, and scenarios—correspond to the different ways companies choose to board the ship based on their own positions. The Chinese Academy of Sciences holds computing power vouchers, Inspur Cloud holds standards tickets, and CloudMinds holds scenario access passes. Zheng Songyan summarizes Cyberport's role with two terms: "super connector" and "super value-added agent." The former connects companies, and the latter empowers them. Combined, it's a recurring theme he emphasizes: "joining forces to go to sea." Cyberport takes companies out into the world together, rather than letting them explore on their own.

Cyberport

A complete link

The three decks are in place, but this is just the beginning. The Hong Kong 2026-2027 Budget proposes that the Hong Kong Artificial Intelligence Research Institute (AIRI) will commence operations in the second half of 2026, with tasks including promoting AI research and development, commercialization of research results, and providing advice on AI governance and regulatory frameworks. The budget also proposes advancing the Sha Ling data infrastructure cluster, promoting embodied intelligence applications, and launching a HK$10 billion Innovation and Technology Industry Guidance Fund to invest in strategic industries such as AI, robotics, and life sciences. These policy combinations share a key phrase: "AI governance." Hong Kong is attempting to connect research, computing power, applications, capital, and regulation. Zheng Songyan believes that "AI governance must proceed in parallel." He says, "When an enterprise or organization adopts AI, all management frameworks and measures must be in place simultaneously." One reason is that "many corporate management teams underestimate the capabilities of AI. In fact, AI can do much more and is developing very rapidly." Zheng Songyan frankly admits that the consequence of underestimating this is that once an enterprise begins to rely on AI, if the model malfunctions, services cannot be interrupted. Therefore, this is where "governance" comes in. Currently, Cyberport is already implementing "AI governance," and Zheng Songyan has listed some practical measures. Cyberport has brought together over 30 cybersecurity companies, many of which are listed companies, to form the "Proxy-Based AI Security Alliance." Simultaneously, it collaborates with several Hong Kong universities and research institutions to study AI governance frameworks and standards, and has developed risk management and detection tools for large-scale AI models. Furthermore, it assists the Hong Kong Monetary Authority in building a "Generative AI Sandbox," allowing Hong Kong banks to test various large-scale AI models within this sandbox to understand which are usable and which have limitations. Cyberport also continuously hosts a series of events on AI+Data, AI+Security, AI+Governance, AI+Network Governance, and AI Ethics, ensuring comprehensive implementation of governance. This may be another added value that Cyberport can provide: prioritizing compliance, risk, and responsibility planning, and facing the challenges of entering the international market alongside businesses—a significant test indeed.