While others are still pulling gacha cards, miHoYo has already started pulling for chips.
On October 19, chip company Mingxin Qirui announced the completion of a Series A financing of over 400 million yuan, with miHoYo appearing on the list of co-investors;
Three days later, Jiutian Ruixin disclosed that it had raised a total of 1 billion yuan in financing over the past six months, and miHoYo became one of the newly joined investors.
According to Guangzhou Daily, on the same day it invested in Mingxin Qirui, miHoYo completed industrial and commercial changes and withdrew its stake of 49% in Zero Unique Thinking, a brain-computer interface company.
Making two moves in the chip sector and divesting from the brain-computer interface business within a week, this set of actions does not look like the conventional business of a game company, but forms the most appropriate entry point for understanding miHoYo at the moment:
While it divests from speculative businesses far from its main business, it pushes deeper into the AI industrial chain; while looking for its next hit game, it tries to reshape the way games are produced.
Cai Haoyu, co-founder and major shareholder of miHoYo, used a metaphor at the 2026 annual meeting: miHoYo in the past was like a big tree, and in the future it will grow into a forest.
In this metaphor, *Genshin Impact* is undoubtedly the thickest tree. It has brought global users, abundant cash flow and the leapfrog development of the company, and also made every subsequent expansion of miHoYo face the same question: Where is the next tree?
miHoYo has proven that it can develop its second and third top-tier games, and it is also trying to use new products to fill the gap left by the decline of old products.
It is now striving to rebuild two sets of capabilities at the same time: to get rid of the single hit model on the product side, and to turn AI into reusable and affordable means of production on the technical side.
What miHoYo wants to change is not just the revenue proportion of a single product, but the growth mode of the entire company.
01 The taller *Genshin Impact* grows, the more miHoYo needs a second tree
When these two chip investments landed in October, miHoYo also just completed a key node on the game side: *Starcloth Valley* has completed its final test in September and is scheduled for public beta in the winter of 2026.
Investing in computing power upstream and testing new products downstream, these two moves together reflect how miHoYo reallocates the resources accumulated from its game business: part is used to broaden the product portfolio, and part is used to build AI and R&D capabilities.
miHoYo's vigilance against relying on a single product actually dates back to earlier than the success of *Genshin Impact*.
When applying for A-share IPO in 2017, miHoYo had already warned of the risk of relying on a single IP in its prospectus.
According to reports from media such as GameTea, in 2020, miHoYo's internal project initiation guideline *How miHoYo makes games* further proposed to "form an aircraft carrier fleet of products", and planned to reduce the revenue proportion of flagship products to one third.
But the early drafting of the concept does not mean that the structure can be changed quickly.
After the launch of *Genshin Impact*, the company's scale expanded rapidly, and the gravitational pull of its top products also increased accordingly: the most mature talents, the strongest content pipeline and the most stable cash flow naturally gather around successful projects.
Changes are also visible in the data.
Statistics from AppMagic, a mobile application data agency, show that in the 2025 global revenue ranking of two-dimensional mobile games, *Honkai: Star Rail* ranked first with 420 million US dollars, and *Genshin Impact* and *Zenless Zone Zero* also entered the top ten.
The most notable point of this list is not that the championship shifts between miHoYo's products, but that three products have entered the top ten at the same time, and the company has initially formed a product echelon.
The revenue of these products has reached a considerable scale. Data disclosed by AppMagic in July this year shows that mobile player spending of *Zenless Zone Zero* in its second year after launch was 198.9 million US dollars, down 51% year on year; its cumulative revenue in the first two years was about 17% of that of *Genshin Impact* in the same period, and about one third of that of *Honkai: Star Rail* in the same period.
Entering August, the IAP revenue of *Genshin Impact* and *Honkai: Star Rail* was about 24 million US dollars and 23 million US dollars respectively, and that of *Zenless Zone Zero* was close to 20 million US dollars.
At the same time, the external market is also increasing the urgency of this transformation.
China Game Industry Report 2025 shows that the actual sales revenue of the domestic game market increased by 7.68% year on year, while the revenue of two-dimensional mobile games decreased by 3.64% year on year to 28.281 billion yuan, declining for the second consecutive year.
In other words, miHoYo's diversification is not an icing on the cake on a naturally expanding track, but a search for new growth space when its core category underperforms the overall industry.
miHoYo has proven that new products can take over the baton, and the next thing to prove is whether new products can push the overall revenue back up when old products decline.
For a company that was once considered to be mainly driven by one super product, having three products remain in the top tier at the same time is more structurally significant than topping the list in a single quarter.
Of course, this caliber only covers App Store and Google Play, excluding PC and console revenue, and is not equivalent to miHoYo's total revenue. But it at least shows that the "forest" is no longer just a rhetoric at the annual meeting.
*Honkai Gakuen 2* launched in 2014 and *Honkai Impact 3rd* launched in 2016 are still in operation. *Tears of Themis* and *Genshin Impact* launched in 2020 cover different groups of people, followed by *Honkai: Star Rail* and *Zenless Zone Zero*.
What miHoYo really needs to solve is no longer whether it has a second successful game, but whether it can continuously create successes and reduce the impact of fluctuations of any single product on the overall situation.
From this perspective, de-singularization does not mean mechanically reducing the proportion of *Genshin Impact*.
A healthier path is to let other trees grow taller, and turn the shade of old trees into nutrients for new trees instead of blocking the sunlight.
02 Changing the growth mode: from version peak to long-term companionship
In fact, the new game *Starcloth Valley* is more worthy of observation than an ordinary new product.
This is miHoYo's first attempt at a small planet life simulation game, where players can plant, fish, cook, build homes, and interact with AI NPCs, with plans to realize interconnection across PC, iOS and Android platforms.
What miHoYo has changed is not just the theme and art style, but also the user structure, retention method and business model that miHoYo is familiar with.
In the past, miHoYo was good at "enriching content": providing new maps, new characters and new storylines through version updates, and then letting players pay for content and emotional connection.
Life simulation games rely more on daily habits and social relationships, and the competition is not only about the number of downloads in the first week, but also about whether players are still willing to come back on the 100th day.
It does not require constant creation of grand version peaks, but the formation of low-pressure, high-frequency long-term companionship. This is also miHoYo's first time to truly face the organizational challenge of social products with a large daily active user base.
However, the track will not leave much room for trial and error. Tencent has *Lili's Little People Kingdom*, NetEase has *Xinghui Youqingtian*, and Nintendo has long established user mindshare for life simulation games represented by *Animal Crossing*.
The competition in 2026 is no longer simply about who has more exquisite pictures, but about who can connect creation, interaction and retention into a closed loop.
Public information shows that the value of adding AI NPCs in *Starcloth Valley* by miHoYo should not only be measured by whether NPCs can say a few more words, but by whether it can continuously provide personalized interaction and keep the additional computing power cost within the affordable range of the business model.
Another pipeline *Honkai: Fate Spirit* is relatively closer to miHoYo's familiar capability circle.
It has completed three rounds of tests, combining spirit raising, automatic wargame and exploration, which not only continues the Honkai IP, but also tries to expand gameplay.
Looking further ahead, the company's disclosed pre-research and in-development directions also include UE5 realistic fantasy action, life MMO, realistic shooting, 3D MOBA and sandbox categories; the outside world is also paying attention to Varsapura and high-spec projects whose details have not been announced, but their release schedule remains to be confirmed by the official.
The longer product list is only the most superficial appearance of the forest. The real difficulty is whether the organization can accommodate different success standards: high-budget two-dimensional masterpieces value content density, casual social products value long-term retention, and shooting and MOBA games rely on control feel, matching mechanism and continuous operation.
Liu Wei, founder and president of miHoYo, put forward the idea that "we need to be patient and hold reverence for the market", which is not a modest statement in this context.
It means that miHoYo begins to admit that past success cannot guarantee the success of the next cross-category attempt; allowing projects to have different rhythms is itself the premise of transforming from a big-tree-type organization to a forest-type organization.
03 Behind the 100 billion yuan, miHoYo is transforming the soil for the forest
When products need to grow into different forms, capital allocation also changes accordingly.
Before 2021, miHoYo's outward investment mainly focused on the upstream and downstream of the game industry, with art, animation, dubbing and cloud game services supporting the content production system.
After *Genshin Impact* brought more abundant cash flow, its investment gradually diverged into two lines: AI hard technology and pan-ecology, and large models, embodied intelligence, XR chips, commercial aerospace, nuclear fusion and brain-computer interface have all entered its investment vision.
"Tech otakus save the world" is more than just a slogan.
Cai Haoyu once publicly described the vision of "building a virtual world that one billion people around the world are willing to live in by 2030". Those seemingly cutting-edge investments far from games can be understood as advance bets on the underlying conditions of the virtual world.
But the vision can also lead to dispersion: when rockets, nuclear fusion and brain-computer interfaces appear in the shareholder list of a game company at the same time, it is difficult to answer when they will form synergy with the cycle of the game business.
The latest one divestment and one new investment show that this dispersion is being reined in.
Judging from the changes in the asset portfolio, divesting from Zero Unique Thinking and previously reducing its stake in Oriental Space can be interpreted as the investment focus converging to the main AI line. However, miHoYo has not announced the reason for divesting from Zero Unique Thinking, which is still an external judgment based on investment actions.
miHoYo's AI layout is not only carried out in the primary market.
Public reports show that miHoYo has registered trademarks related to the "Wuyong" model; on October 22, it jointly launched and open-sourced the GameScaling series of game models with universities and scientific research institutions; in addition, Cai Haoyu is personally in charge of the basic model team.
From this we can see that miHoYo's chip investment is more suitable to be viewed in the context of "simultaneous development of self-developed models, R&D pipelines, product applications and underlying computing", rather than being understood in isolation as two financial investments.
Among them, MiniMax provides a sample of "both investing in and using the technology".
miHoYo participated in its early financing before the large model boom fully broke out. According to the calculation of Guangzhou Daily, the book return of this investment in the early stage of MiniMax's listing may exceed 24 times;
More important than the financial return is that the MiniMax model has been used for character behavior rules, NPC dialogue and R&D process exploration in *Honkai: Star Rail*. However, public reports have not clearly distinguished which of these belong to officially launched functions and which belong to internal R&D tests.
Investment here is not a financial decoration outside the main business, but a way to connect external technologies to its own content production system.
AI has also gone beyond the stage of "icing on the cake".
In traditional games, a large amount of content can be repeatedly served to players after production is completed; generative AI may generate new reasoning costs for every dialogue and every behavior response.
Traditional games copy the same content to more players, while AI games may need to pay for every interaction. As a result, the more active the users are, the longer the computing power bill may be.
miHoYo's AI companion product *BSide: Olivia Lin* launched in July 2026 was shut down less than a month after its release.
Its quick exit cannot prove a single reason for failure, but it brings to the forefront the common problem faced by AI-native products: whether user retention can cover the continuously occurring reasoning costs.
This short trial and error shows that AI products cannot be naturally established by simply connecting the model to the game, and retention, experience and cost must all meet the standards at the same time.
The technical directions of the two companies both involve improving computing efficiency, thus providing an entry point for observing miHoYo's AI cost problem.
Mingxin Qirui explores the integration of storage and computing and language processing units, while Jiutian Ruixin takes the route of neuromorphic and low-power chips. They do not directly produce games for miHoYo, but both are trying to improve computing efficiency.
For a company that hopes to integrate AI NPCs into daily gameplay and make generated content serve massive players, investing in reasoning chips is more like laying out the future cost structure in advance.
However, equity investment does not equal procurement relationship, and there is no public evidence at present to prove that the relevant chips have been connected to miHoYo's model or game business.
This also gives a more specific meaning to Liu Wei's statement that "a maximum of 100 billion yuan will be invested in the next three years".
100 billion yuan is the upper limit of investment, not the expenditure that has already occurred; the 2000 billion yuan valuation given by Hurun Global Unicorn Index 2026 is also only an estimate of the list, not the cash on the company's books.
The two cannot be directly converted into miHoYo's payment capacity, but they are enough to show the strategic intensity: if this investment plan is gradually fulfilled, its essence is to convert the cash flow accumulated from the game business into models, computing power, talents and next-generation content production capabilities.
Therefore, to judge whether miHoYo can grow into a forest, we should not only look at how many companies it has invested in and how many new products it has announced, but also see whether *Starcloth Valley* can establish a new retention model, whether *Honkai: Fate Spirit* can continue the product echelon, and whether AI can evolve from NPC dialogues in demonstrations to reusable and affordable production tools.
The big tree will still be miHoYo's most important asset, but it no longer wants to prove itself only by the height of a single tree.
The meaning of the forest has never been that every tree can survive, but that when the seasons change and the cycle of individual products fluctuates, the entire company no longer needs to rely on the prosperity and decline of the same tree.