Author: Frank, PANews
At most tech conferences, the most common question is "Who released what?" But at mu Shanghai AI WEEK in May 2026, PANews heard a more practical question: As AI makes it increasingly easier to build product prototypes, where has the real difficulty of starting a business shifted?
What makes this event special is that it's not like a standard conference, but more like a temporary developer space. There are very few booths, very few company presentations, and the topics are not fixed. A large number of overseas developers fly from Argentina, Silicon Valley, Japan, or Southeast Asia to Shanghai, just to connect with Chinese developers, model companies, investors, and the local ecosystem within a month.
The event venue wasn't set up like a traditional hotel conference room, but rather a hybrid space comprised of open workspaces, tiered seating, beanbag chairs, and a temporary projector. Some people sat at their desks coding, others gathered around carpets and square cushions to listen to presentations, and still others leaned against corners to continue refining their products on their laptops. Colorful mu Shanghai flags hung on the walls, and a world map with the question "Who am I? What shaped me?" covered in sticky notes and connecting lines, resembling an identity network being collaboratively filled out by the participants.
PANews's on-site discussions with several organizers, project representatives, investors, and model company personnel revealed that AI entrepreneurship is entering a new phase. If the first stage of AI entrepreneurship was "who can connect to the model and create a product faster," then the second stage is "who can find real-world scenarios, acquire users, build a community, and survive for a sufficiently long period." Models are like water, electricity, and gas; what's truly scarce now is no longer just the ability to connect the pipes, but who can find those who need the water most.
A deep social experiment for global developers
What makes mu Shanghai truly unique is its organizational structure. In an interview with PANews, founder Sun mentioned that mu didn't originate in China, but rather spread through pop-up cities and startup communities in places like Thailand, Argentina, Africa, and Japan. Compared to traditional two- or three-day conferences, it emphasizes a group of people entering the same city to co-create, exchange ideas, live together, and build relationships over a period of about a month.
This format naturally imbues the event with a strong community element. According to Sun, approximately 2,000 people registered for mu Shanghai, with over 800 ultimately selected. The participants were also quite diverse: approximately 20% were from China, 18% from other Asian regions such as Japan, South Korea, and India, 16% from Southeast Asia, 10% from Latin America, 10% from the United States, and 11% from Europe, with about 6% from Africa. In terms of industry background, AI professionals comprised about 40%, Web3 professionals about 20% to 30%, and there were also participants from hardware, biotechnology, investment, and other diverse groups.
In an interview, Sun explained the appeal of this type of event: "After leaving university, people rarely have that kind of deep relationship anymore. It's also difficult to form this kind of connection in the workplace and in big cities, so I think it's very valuable." In his view, mu is not trying to replicate the instantaneous flow of traffic in traditional conferences, but rather a relationship density that is closer to that of universities, communities, and communal living.
The actual event environment was indeed closer to this. The main stage wasn't always the center of the space; the subtitle screens next to the projection screen, makeshift display stands, and computers scattered throughout formed the everyday backdrop of the event. In a presentation about user experience, the audience wasn't seated neatly in chairs, but rather dispersed among low cushions, the floor, and open workstations. The speaker presented at the front, while those in the audience listened, took notes, replied to messages, or continued working on their own projects. This slightly relaxed atmosphere was actually closer to how a developer community truly operates.
The significance of these figures lies not in the scale of the event itself, but in the organizational logic it demonstrates, which differs from traditional exhibitions. Traditional conferences often connect brands and users, or companies and clients; mu Shanghai, on the other hand, is more like connecting Chinese and international developer cultures. The event featured large model roundtables, hackathons, co-creation activities, language learning, community sharing, and impromptu discussions. Feng Wen, product manager of MiniMax, mentioned in an on-site exchange that the atmosphere here is not just about "sharing AI on stage," but also includes cultural exchange, developer co-creation, and community participation.
The emergence of a large number of Web3 practitioners has also made this connection more complex. What the Web3 industry has accumulated over the past few years is not just on-chain assets and speculative narratives, but also a set of methodologies for community mobilization, global collaboration, social media dissemination, and developer organization. As AI startups shift from focusing on model implementation to focusing on user reach, this methodology has become valuable again.
From "how to do it" to "who to sell it to": AI startups are entering a more complex phase.
PANews' most striking observation at the event was that AI entrepreneurs are no longer so excited about "whether they can create a product." Multimodal models, code generation tools, agent frameworks, and automated workflows are rapidly lowering the barrier to product prototyping. A small tool that previously required collaboration between designers, engineers, and operations staff can now be built into an initial version by a few people in a few nights using AI-powered coding tools.
More recent data further illustrates this shift in barriers to entry. JetBrains' AI Pulse survey in January 2026 shows that 90% of professional developers routinely use at least one AI tool in their work, and 74% have adopted AI tools specifically designed for developers. For entrepreneurs, the ability to "make it" is becoming a more common skill, rather than a natural barrier.
However, the real problems only began after the product was developed. An entrepreneur named Nathan told PANews that he's creating a product to help AI entrepreneurs find their entrepreneurial direction. His logic is that AI can now expand its information gathering scope and distill the judgment and taste of serial entrepreneurs into a set of rules, which can then be used by AI to discover business opportunity signals. But this product also reveals a larger reality: as product development becomes easier, the question of "what to do" becomes the scarcer one.
Nathan told PANews, "It's already quick to create something new with the help of AI coding tools. The real key is whether the direction is worth pursuing." His product essentially productizes the act of "finding direction" itself. This small case reflects a new trend in AI entrepreneurship: when execution is amplified by AI, judgment becomes a scarce asset.
During the roundtable discussion on "Innovative Practices and Path Exploration in the AI Consumer Ecosystem" hosted by PANews, several guests expressed similar views: AI has indeed made rapid prototyping, demonstration samples, and initial launches easier, but the truly difficult aspects of entrepreneurship have not disappeared. Customer acquisition, commercialization, community engagement, user education, and interpersonal connections still require teams to possess more comprehensive capabilities.
In other words, AI lowers the development barrier, not the startup barrier. In the past, the first hurdle in product competition was "whether it could be made." Now that this barrier has been significantly lowered, the real selection process has shifted to distribution, application scenarios, and commercialization. One interviewee summarized it as: making tools isn't difficult now; the difficulty lies in getting the product, its intellectual property, and its value seen by more people.
This is a common dilemma faced by many AI tools. The more tools there are, the harder it is for users to choose; the more powerful the model, the easier it is for individual functions to be swallowed up by the next model update. For entrepreneurs, a product that seems viable today may lose its presence in six months due to improvements in the underlying model's capabilities. Therefore, the real question is not "whether to do AI," but whether it is possible to find a specific scenario that the model cannot completely eliminate in the short term.
The use of AI is rapidly becoming more widespread, but there are still gaps between using it as a tool and realizing its stable value, including scenarios, processes, governance, and organizational capabilities.
Web3 users flock to AI, not just to follow the trend.
From a purely narrative perspective, the influx of Web3 professionals into AI may seem like just another shift in trending topics. However, at mu Shanghai, there are more practical reasons behind this shift.
On the one hand, the wealth effect, capital dividends, and technological dividends of the crypto industry are diminishing, and many practitioners are beginning to look for new technological directions; on the other hand, AI applications happen to require the capabilities that the Web3 industry is most familiar with: community, global dissemination, developer relations, and social media distribution.
A veteran Web3 practitioner stated frankly at the event that the crypto industry has been around for 10 years, and the era of capital and cognitive dividends is largely over. Now, it's time to move towards new technological directions. He advised entrepreneurs to gradually shift their businesses, personal brands, and asset allocation towards AI, rather than continuing to pour a lot of energy into cryptocurrencies. This assessment may not represent all Web3 practitioners, but it certainly reflects the true mindset of some in the audience.
His statement was straightforward: "I believe AI is worth the long-term investment. By investment, I mean not just using the tools, but gradually shifting your career, personal brand, and asset allocation towards AI." His personal choice is to transform into an AI blogger, using an action camera to find teams making AI products and filming vlogs.
This kind of judgment may not represent all Web3 practitioners, but it is enough to illustrate the atmosphere at the scene: AI is no longer just an optional track, but is becoming a direction for some Web3 practitioners to reconfigure their time, assets and professional identity.
XerpaAI, an AI-driven social media assistant, set up a booth at the event. During an interview, a staff member stated, "We are a pure AI project, technically not closely related to Web3. However, from the user's perspective, we will definitely reach Web3 users. For example, the X AI assistant will serve a segment of Web3 users with operational needs." This statement aptly represents the current ambiguous relationship between AI applications and the Web3 community: the product may not be Web3, but users, dissemination, and early-stage needs often inextricably linked to Web3.
During the on-site exchange, representatives from model companies also mentioned that the user groups of AI and Web3 are becoming increasingly difficult to separate, as many heavy users of AI tools originally have a Web3 background. Especially in scenarios like Hong Kong and Shanghai, AI and Web3 often share the same group of high-frequency attendees, early users, and community dissemination nodes. For them, it doesn't matter whether community members are Web3 users or not; as long as the topic is AI, everyone's goal is the same.
From this perspective, Web3's entry into AI is not merely a "transition." Web3 brings not just on-chain technology itself, but a methodology for bringing global developers together around a project, fostering continuous discussion, and encouraging them to contribute their attention. For current AI applications, this capability may be more difficult to replicate than a short-term feature.
Hardware, supply chain and China's foundation
Compared to the anxiety about whether AI software applications will be "eaten up" by models, the discussions at the event regarding AI hardware, embodied intelligence, and the Chinese supply chain seemed more certain. Several interviewees mentioned that once AI enters the real world, hardware, robotics, embodied intelligence, and multi-sensory interaction will see greater opportunities. In the consumer AI roundtable hosted by PANews, Feng Wen, product manager of the MiniMax open platform, also predicted that smart hardware, robotics, and embodied intelligence will reach a significant turning point in the next three to five years, with AI no longer confined to software interfaces but entering the real physical world.
Beyond the conference venue, the robotics field is also gaining attention. A parcel sorting competition held by overseas robotics manufacturer Figur on May 18th, pitting humans against robots, sparked widespread online discussion. Even though humans won by a narrow margin within 10 hours, it's clear that robots emerged victorious over longer periods. Stanford's HAI 2026 AI Index also shows that the accuracy of AI agents in real-world computer task tests like OSWorld has increased from approximately 12% to 66.3%. Autonomous driving is also beginning to see large-scale deployment, with China's Apollo Go completing 11 million fully driverless trips.
AI entering the real world through hardware, robots, and edge deployments is no longer just a distant narrative.
This is precisely where China's ecosystem's unique advantage lies. Sun repeatedly mentioned in the interview that China possesses a near-complete supply chain, encompassing hardware, AI, consumer technology, and infrastructure. For overseas entrepreneurs looking to develop AI hardware, whether it's raw materials, factories, engineers, or rapid prototyping capabilities, it's ultimately difficult to bypass China. He also revealed that many overseas entrepreneurs who came to China for this event aimed to experience and observe China's complete industrial chain firsthand.
Sun stated, "As long as you're making hardware, overseas teams will ultimately return to China to find supply chains, raw materials, engineers, and prototyping capabilities." He believes that in the next five to ten years, more international talent will come to China to find supply chains, raw materials, talent, and capital. For overseas entrepreneurs, China is not just a market, but also a set of infrastructure for bringing products to market.
A venture capitalist at the event told PANews that their main goal in participating was to see if there were any applications that leaned more towards hard technology, embodied intelligence, and world models, rather than simply consumer applications. Their logic is that if the cost of replicating software AI is decreasing, then hardware, supply chains, and real-world interactions may become barriers that are harder to overcome by simply updating models.
However, the appeal of China's AI ecosystem to overseas developers doesn't solely stem from its supply chain. The emergence of domestically developed models such as DeepSeek, Kimi, MiniMax, Zhipu, and Qianwen has led overseas developers to re-evaluate the capabilities of Chinese models. However, trust and deployment challenges remain when Chinese models are exported. Feng Wen, product manager of the MiniMax open platform, noted that Chinese models primarily gain attention and brand influence overseas through open source, but many overseas developers still worry about data, compliance, and trust issues. Even with open source models, most people may not have sufficient computing power to deploy them themselves. This has led to the emergence of an intermediary layer where US companies deploy Chinese open-source models and then provide them to overseas clients.
For overseas developers, the appeal of China's AI ecosystem no longer comes solely from cost or market size, but also from its continuously expanding model supply, engineering capabilities, and industrial transformation capabilities.
This means that the opportunities in China's AI ecosystem are not linear. Model capabilities, hardware supply chains, government execution, and the developer community all need to work together to truly bring overseas entrepreneurs in. mu Shanghai plays a role in this process more like a connector, bringing overseas developers into China.
Major model companies begin vying for developer communities
If the competition among large model companies over the past year mainly focused on parameters, rankings, and prices, then at mu Shanghai, the importance of the developer community has been brought to the forefront. Domestic large model companies don't just need more API calls; they need to make developers aware of them, trust them, and be willing to build applications around their models.
During the on-site exchange, Feng Wen mentioned that they had done a lot of work related to developers. Developer experience, event selection, guest participation, hackathons, judges, token sponsorship, etc., all need to be incorporated into the ecosystem work of the model company.
“Developers are our users, so we value the developer experience and hope to help more developers understand what we are doing,” Feng Wen said. This statement can almost be seen as a footnote to the ecosystem strategy of large domestic model companies. Models are no longer just placed on platforms waiting to be used, but are actively entering spaces where developers gather.
This isn't just a choice made by MiniMax. Attendees revealed that Zhipu has its own Yuandian Academy in Beijing, which hosts events almost weekly and is close to resources from top universities like Tsinghua and Peking University. The AIGC and AGI communities are also continuously attracting talent through fixed spaces, hackathons, hot pot gatherings, and developer nights. These spaces are becoming offline gateways for developers.
Behind this lies a larger shift: model companies are no longer content with simply "releasing the model." They need documentation, trial platforms, case studies, video tutorials, as well as communities, hackathons, and developer events to help users overcome the initial hurdles. As agent capabilities improve, user education itself is being restructured. Previously, developers had to read documentation, check error codes, and understand parameters themselves; now, agents can help users read documentation, search for solutions, select models, and automatically correct their paths.
For model companies, the real competition isn't just about the price of model usage, but about who can integrate into developers' daily workflows first. For application entrepreneurs, the real opportunity isn't just about which model to integrate, but about finding a group of early users willing to continuously use, provide feedback, and even actively spread the word.
To be needed, to be understood, to be left behind
mu Shanghai did not provide a unified answer for AI startups. Some are optimistic about hardware, some are focusing on social media growth, some are discovering startup opportunities, some are discussing cultural export and spiritual consumption, while others see it as an entry point to connect with overseas developers and local partners.
These seemingly scattered clues, however, precisely constitute the most realistic state of AI entrepreneurship today. Model capabilities continue to advance, but application forms are still searching for stable scenarios; the development threshold has decreased, but distribution and commercialization have become more critical; the Web3 craze has cooled down, but the community methods it left behind are being absorbed by AI; China's supply chain and model capabilities have become important, but overseas developers still need a credible entry point to understand China.
In the interview, Sun mentioned that mu Shanghai's long-term goal is not just to host a single event, but to create a sustainable space where people from overseas and China can meet, collaborate, and create new things in one place. In fact, mu has very few full-time employees; much of the work is driven by contributors and partners. This organizational structure is very similar to Web3 and open-source communities—decentralized, contribution-driven, and focused on networking—and therefore more likely to attract people familiar with this culture.
Of course, this model still has many uncertainties. Whether the activities can be transformed into long-term spaces, whether the community's enthusiasm can be translated into real projects, whether overseas developers will stay in the Chinese ecosystem long-term, and whether large model companies can convert developer activities into stable usage volume all remain to be seen. Communities can create encounters, but they cannot replace a closed business loop; cities can provide scenarios, but they cannot guarantee product success.
However, mu Shanghai has at least made one trend clear: AI entrepreneurship is shifting from "model worship" to "scenario competition," from "creating tools" to "being seen by users," and from single-product focus to comprehensive competition involving communities, supply chains, and cross-border collaborations. For ordinary entrepreneurs, the opportunities brought by AI are not about making everyone an easy winner, but rather about exposing more people to the same, more intense selection process earlier.
As products become increasingly easy to manufacture, the real scarcity lies in the ability to understand users, integrate into scenarios, build trust, and maintain continuous connections with people. AI will continue to reduce the production costs of tools, but it won't automatically answer the question, "Why you?" In this sense, creating a product is only the first step; being needed, understood, and retained is the more challenging second half of AI entrepreneurship.
