Amid the ongoing volatility in the Hong Kong stock market, artificial intelligence unicorn companies have experienced a dual resonance of capital and technology. With the launch of a major video generation model and an unexpected surge in commercialization, their stock price has risen by more than 20% since August 31st, becoming a focal point in the industry.

The core driving force behind this strong performance stems from MiniMax's substantial breakthroughs in multimodal technology and exponential growth in commercialization. On August 31st, MiniMax officially announced the integration of H3Max768P and 480P versions into its open platform and Design system. This version was fine-tuned and optimized for inference by ecological partners fal based on MiniMax H3 open weights, achieving a revolutionary improvement in generation speed. By introducing new data on the algorithm side and using verifiable reinforcement learning to optimize instruction following and visual tension, while combining engineering-side deep adaptation of its own inference infrastructure, the time required to generate a 5-second, 768P resolution audio-visual clip has been reduced to less than 3 seconds, and a 15-second video can be completed in about 15 seconds.

This qualitative change in speed has completely broken through the traditional application boundaries of video generation, giving rise to new product forms. Since the model can quickly prepare the next segment of content while playing the previous one, overseas developers have successfully integrated it into real-time live streaming scenarios. For example, engineer Rehan Sheikh integrated it into the Twitch platform, creating a "dimensional TV" without a fixed schedule, continuously generating audio-visual content and live-streaming it in real time, which triggered nearly 5.3 million high-traffic views on social platforms. At the same time, China has also seen the commercialization of AI-generated long-form dramas without real people, with the first full AI broadcasted long drama "The Later Journey to the West" produced by Mango TV and supported by Seedance technology, officially scheduled, marking that AIGC is accelerating from single material creation to frequent content generation products.

image.png

While the technical ecosystem is rapidly advancing, MiniMax's commercial monetization has also entered a period of explosive growth. Financial data shows that in the first half of 2026, the company achieved revenue of approximately $117 million, a year-on-year increase of 283.1%, with a month-on-month growth of 81.8% in the second quarter. After entering the third quarter, the growth rate became steeper, with the token consumption in July surging to 20 times that of January, and the annual recurring revenue in August broke through the $800 million mark.

Notably, the company's revenue structure has successfully transitioned from the consumer end to the business end. Currently, enterprise and developer businesses contribute about 80% of the ARR, with the total number of enterprise customers and developers exceeding 2 million, representing a tenfold expansion from the end of 2025, demonstrating strong customer loyalty and ecological attraction.

To support future large-scale technological iteration, the company has made sufficient preparations in terms of funding and computing power. In July 2026, MiniMax raised approximately HKD 16 billion through share placements and zero-coupon convertible bonds, and its current cash reserves exceed $3 billion. Relying on the efficient training of core clusters and in-depth cooperation with cloud vendors and inference factories, the company's existing and planned computing power is fully capable of covering the training and iteration needs of flagship models such as M3.1, M3Pro, and H3.1 in the future.

According to forecasts from multiple leading institutions, the market maintains a high level of optimism about the company's future development. As AI model commercialization accelerates in the next 4 to 6 months, the M3Pro model with a parameter scale of up to 3 trillion is expected to be officially launched in September to October, further boosting the company's revenue growth. With the continuous iteration of new models and the deep penetration of high-Token consumption scenarios such as Agents, the demand for computing power on the inference side will continue to rise, which also fully demonstrates that domestic large models are accelerating into the fast lane of large-scale monetization.