Inside Shanghai’s New Generation of AI Startups

Inside Shanghai’s New Generation of AI Startups

From multimodal foundation models to humanoid robots, Shanghai’s emerging technology companies are trying to turn artificial intelligence into globally competitive products—and sustainable businesses.

On the banks of the Huangpu River, Shanghai’s artificial-intelligence industry is moving beyond research demonstrations and chatbot launches. A new generation of companies is building video-generation platforms, autonomous software agents, industrial robots and humanoid machines designed to work alongside people.

The scale of the city’s ambitions is already considerable. Shanghai’s AI industry generated an estimated more than 550 billion yuan in 2025, an increase of over 30 per cent from the previous year, according to the Shanghai Municipal Commission of Economy and Informatization. The municipal government attributes part of that expansion to programmes intended to reduce the cost of computing power, data and model development for local companies.

Yet the most important change is not simply the industry’s size. Shanghai’s emerging AI sector is becoming increasingly diverse. Its leading young companies are no longer concentrated exclusively on building large language models. They are combining models with consumer software, industrial systems, video generation, scientific research and physical robots.

MiniMax takes Shanghai’s consumer AI products overseas

MiniMax is one of the clearest examples of a Shanghai startup attempting to compete internationally.

Founded in early 2022 by former SenseTime executive Yan Junjie, MiniMax develops multimodal models capable of working with text, audio, images, video and music. Its best-known products include Hailuo AI, which generates video, and Talkie, a platform through which users interact with AI-generated characters.

The company’s international reach distinguishes it from many Chinese AI developers. MiniMax reported revenue of US$79 million in 2025, an increase of 159 per cent year on year, with more than 70 per cent of sales coming from outside China. Its income came from a combination of consumer subscriptions and enterprise services.

That growth has not yet translated into overall profitability. MiniMax reported a net loss of US$1.87 billion for 2025, although a large portion was connected to changes in the value of financial instruments rather than ordinary operating expenses. The wider point remains: developing frontier models, acquiring users and operating generative-AI services require enormous amounts of capital.

In January 2026, MiniMax raised approximately HK$4.82 billion, or about US$619 million, through its Hong Kong initial public offering. The company said that much of the money would be directed towards research and development over the following five years.

MiniMax’s transition from privately funded startup to publicly traded company illustrates a possible path for Shanghai’s most successful AI businesses. Rather than remaining laboratories that depend indefinitely on venture capital, they are under increasing pressure to develop products that attract paying users at global scale.

StepFun bets on multimodal models and AI agents

Another major Shanghai company, StepFun, is pursuing a different version of the foundation-model opportunity.

Established in 2023, StepFun has focused heavily on multimodal systems capable of processing combinations of text, images, audio and video. Its product strategy has also increasingly emphasised AI agents: systems designed not merely to answer questions but to use tools, complete sequences of actions and participate in more complicated digital workflows.

At the 2025 World Artificial Intelligence Conference in Shanghai, StepFun presented Step 3, a new generation of its foundation model. The company also helped establish an alliance connecting large-model developers with Chinese AI-chip manufacturers, including Huawei, Biren Technology, Moore Threads and Enflame. The purpose was to improve compatibility between domestically developed models and computing hardware as restrictions continue to complicate Chinese companies’ access to the most advanced foreign chips.

By April 2026, StepFun was restructuring its offshore corporate arrangements as it prepared for a possible Hong Kong listing. Reuters reported that the process took place amid tighter Chinese scrutiny of corporate structures commonly used by technology companies to raise money overseas.

StepFun represents a broader shift in China’s AI market. Producing a capable general-purpose model is no longer enough to guarantee commercial success. Developers must demonstrate that their models can operate efficiently, run on available hardware and support applications for businesses, vehicles, personal devices or autonomous agents.

AgiBot brings AI into the physical world

Shanghai’s AI story is not limited to software.

In a large facility on the outskirts of the city, humanoid robots operated for as long as 17 hours a day have repeatedly practised activities such as folding clothes, making sandwiches and opening doors. The facility belongs to AgiBot, a Shanghai humanoid-robotics company founded in 2023. The repeated exercises generate data that can be used to train robots to understand and perform physical tasks.

This process addresses one of embodied AI’s greatest constraints: physical training data are much harder and more expensive to collect than the text and images used to train conventional generative models. A language model can learn from enormous existing collections of digital material. A humanoid robot must learn how force, balance, movement and objects behave in the real world.

AgiBot is attempting to develop robots for manufacturing, commercial services, education and entertainment. In 2025, it completed a strategic funding round involving investors including LG Electronics and Mirae Asset. The company did not disclose the amount raised.

By July 2026, AgiBot had formally begun the process of pursuing a Hong Kong IPO, according to a Securities Times report cited by Reuters. Reuters had previously reported that the company was considering a valuation of between HK$40 billion and HK$50 billion, although such proposed valuations are not guaranteed and may change before an offering takes place.

AgiBot’s rise shows how closely Shanghai’s AI ambitions are connected to its established strengths in manufacturing, engineering and supply-chain management. For an embodied-intelligence company, the city offers access not just to programmers, but to component producers, industrial customers, hardware engineers and factories in the wider Yangtze River Delta.

An ecosystem designed to reduce startup costs

Shanghai’s government has deliberately created infrastructure around young AI companies.

The Shanghai Foundation Model Innovation Center in Xuhui district was established as an incubator for foundation-model developers. By April 2025, it was home to more than 100 enterprises. Municipal sources later described a broader ecosystem of several hundred model-related businesses clustered around the centre and associated programmes.

The centre provides access to computing resources, shared datasets, financial services, talent support and opportunities to test products in practical scenarios. Resident companies have been offered computing packages valued at 1 million yuan, alongside more flexible pricing arrangements.

Shanghai has also introduced vouchers covering as much as 30 per cent of qualifying intelligent-computing rental costs. Other municipal measures support training-data purchases, model use and work in areas including AI chips, embodied intelligence, brain-computer interfaces and intelligent software.

For embodied-intelligence companies meeting the relevant requirements, Shanghai’s policy framework provides for computing-power support worth as much as 40 million yuan per year.

These programmes do not remove the commercial risks faced by startups, but they can reduce the cost of reaching a working prototype. They also allow the city to concentrate companies, investors, universities and public research organisations in a relatively small number of innovation districts.

The difficult search for sustainable revenue

Despite the excitement surrounding Shanghai’s AI companies, their long-term economics remain uncertain.

Model training and operation are expensive, while intense competition has pushed down the price of AI services. China’s market has experienced what industry participants have described as a “war of a hundred models”, with companies offering inexpensive or free products in the hope of attracting developers and users. Hardware restrictions, price competition and slower-than-expected adoption by some businesses all place pressure on profit margins.

The experience of MiniMax demonstrates both sides of the opportunity. Its revenue and international customer base grew rapidly, but the company remained loss-making. For the next generation of startups, technical performance will therefore be only one measure of success. They will also have to control computing costs, establish reliable revenue streams and demonstrate that customers will continue paying after the novelty of generative AI has faded.

Robotics companies face an additional challenge. Impressive demonstrations do not necessarily prove that a machine can perform safely and economically in an uncontrolled workplace. Humanoid robots must become reliable enough to justify their purchase, maintenance and supervision costs when compared with conventional automation or human labour.

Shanghai’s emerging model of AI entrepreneurship

Shanghai is developing an AI ecosystem that differs from the image of a startup sector dominated solely by consumer applications or standalone chatbots.

Its leading companies sit at the intersection of several industries. MiniMax combines foundation models with international consumer software. StepFun connects multimodal models, autonomous agents and domestic computing hardware. AgiBot is transforming AI research into physical machines intended for factories and service environments.

Behind them sits a city-supported network of incubators, computing subsidies, universities, public research institutes, manufacturers and investment organisations. This gives Shanghai an unusual combination of software talent, industrial infrastructure and access to capital.

The central question is no longer whether Shanghai can produce ambitious AI startups. It already has.

The question is whether these companies can convert technical capability, public support and investor enthusiasm into businesses that remain competitive after subsidies decline, markets consolidate and customers begin demanding measurable financial value.

The companies that survive that transition will define Shanghai’s next generation of artificial intelligence—not simply through the models they release, but through the products, machines and industries those models make possible.