MiniMax, the Shanghai-based AI startup best known for its M-series large language models and the Hailuo AI video generation tool, has disclosed growth metrics that underscore the accelerating commercial momentum of China’s second-tier AI companies. Speaking at the UBS Asian Investment Conference in Hong Kong on May 28, co-founder and president Yun Yeyi announced that the company’s global enterprise and developer client base had surpassed 1 million, a fivefold increase from approximately 200,000 clients just six months ago, according to South China Morning Post reporting.
The company’s global user base has reached 300 million, and annual recurring revenue (ARR) has more than doubled over the past two months alone. CEO Yan Junjie disclosed on an earnings call in March that ARR had reached US$150 million or more by February, driven by surging demand for generative AI products. The doubling since then would put ARR at US$300 million or above, though MiniMax has not published an updated figure.
From IPO to Hang Seng Index
MiniMax’s trajectory since its Hong Kong IPO in January 2026 has been exceptional even by the standards of China’s AI boom. Shares have surged nearly 500% since listing. The company will be added to Hong Kong’s benchmark Hang Seng Index after market close on June 5, alongside Zhipu AI, following the exchange’s quarterly review. Inclusion in the Hang Seng Index typically triggers significant institutional buying as index-tracking funds are required to hold the stock, a structural demand driver that could sustain the share price momentum.
The financial picture is more complex beneath the headline growth. MiniMax’s 2025 revenue rose nearly 159% year-on-year to US$79 million — strong growth, but modest in absolute terms for a company with a 300-million-user base. Total losses swelled 302% to US$1.87 billion, though the company noted that much of this figure reflects fair value losses on financial liabilities rather than operational cash burn. The gap between revenue and losses is characteristic of the current phase of China’s AI industry, where companies are investing heavily in model development and infrastructure ahead of monetization.
The M3 Series and Self-Evolving Models
On the technology side, MiniMax is preparing to release its M3 series, which will feature a new “sparse attention” mechanism designed to reduce computational costs. CEO Yan Junjie said the M3 series would be developed with “much greater” resources than its predecessor — a signal that MiniMax is prepared to increase its capital expenditure on model training even as it works to improve efficiency.
The company published a technical paper on Tuesday outlining the development of its M2 series, which revealed an intriguing finding: the models are showing early signs of what MiniMax describes as “self-evolution,” a phenomenon in which AI systems begin to automate parts of their own research and development process, helping to build more powerful iterations without proportional increases in human engineering effort. If this capability scales, it could significantly accelerate MiniMax’s model development cadence.
MiniMax’s current flagship, the M2.7, ranks 8th among open-weights models on the Artificial Analysis benchmark, trailing Moonshot AI, Xiaomi, DeepSeek, and Zhipu AI. The ranking reflects MiniMax’s position as a strong but not leading player in China’s intensely competitive model landscape, a position that its enterprise client growth and international user base suggest it is compensating for through distribution and product breadth rather than raw benchmark performance.
The self-evolution finding in the M2 technical paper deserves particular attention. If AI systems can automate meaningful portions of their own R&D cycle, identifying architectural improvements, generating training data, or tuning hyperparameters without proportional human engineering input, the relationship between capital expenditure and model capability changes fundamentally. Companies that achieve self-evolving model development can potentially outpace rivals who rely entirely on human researchers, regardless of the size of their engineering teams. MiniMax’s disclosure that its models are showing early signs of this capability is a signal worth watching, even if the practical impact on the M3 series remains to be demonstrated.
As EastFrontier has reported, Hong Kong IPOs hit a five-year high in Q1 2026 led by China AI tigers ZhipuAI and MiniMax, and the China AI startup funding environment shows no sign of cooling. MiniMax’s client and user growth figures add commercial substance to what has until now been a story driven largely by financial market enthusiasm.