HoYoverse giant wheel steering

HoYoverse wants to be a forest.

2026 At the annual meeting at the beginning of the year, Cai Haoyu used a metaphor: In the past, HoYoverse was like a big tree, and the majority of head games contributed to the majority. In the future, he wants to make it a forest. The trees are big and small, but each tree has uniqueness and is the best in their respective fields.

This is not the first time that HoYoverse has sought diversification. As early as the 2017 prospectus, it stated that it is necessary to get rid of the dependence on a single IP; in 2020, the internal project guide ‘How Mihoyo Makes’ Games’ also proposed to ‘form an aircraft carrier fleet to form a product’, which should reduce the proportion of revenue of flagship products to one-third.

However, reality has gone back to the old road repeatedly. After the launch of ‘Original God’, the game has carried most of the revenue, and the concept of the ‘Fleet Group’ has been put on hold by a huge success. Since then, ‘Honkai: Xingqiong Railway’ and ‘Absolute Zone Zero’ have been launched one after another, the product line has been enriched, but the head effect has not changed.

But this time it seems different. After Cai Haoyu, Liu Wei explained with a more straightforward and humble attitude why Miha You had to do accumulation, slump, ten years, and three generations. ‘Be patient and be in awe of the market. Don’t think we can do anything, the first generation may have lower expectations.’

This may be the first time Mihayou has ever talked so seriously at the annual meeting to talk about ‘respect’. Some employees understood the overtones: ‘Actually, it is a vaccination, and they think that the next few films may not be as dazzling as they used to be.’

One talks about the forest, the other talks about awe. Two paragraphs point to the same thing: this company that has won all the way from its establishment to the present is seriously preparing for possible stepping on the pit and accumulation.

01

Snake bites its tail

In 2013, Miha You almost didn’t survive.

There are 7 people in the team, and the annual net profit is 280,000 yuan, which is allocated to each person, less than 3,400 yuan a month. Cai Haoyu, Liu Wei and Luo Yuhao gave themselves a monthly salary of 4,000 yuan, and shared a house with a monthly rent of 3,900 yuan. Their last game, ‘Honkai Gakuen’, uses a buyout system, priced at 6 yuan. Although the gameplay is bright, the paid model is completely unreasonable, and there are too many pirated copies, and no one is willing to spend money.

If the next game can’t make money, the company will be gone.

After the buyout system hit a wall, they redesigned the numerical and payment system, referring to the proven card commercialization model in the Japanese market. ‘Honkai Gakuen 2’ is redone under this new framework. After its launch in early 2014, it entered the top ten of the App Store free list and best-selling list, and the revenue that year was close to 100 million. Later, Cai Haoyu recalled that period: ‘The pursuit of business success does not seem to be idealistic, but only the combination of ideals and reality can go on.’

‘Honkai Gakuen 2’ has made the company survive, and also established a set of underlying logic that has been repeatedly verified after that: roles are core assets, paying for content around roles and paying for content.

Many peers will choose to change the skin after earning the first pot of gold, or sell the company while the valuation is high. But Mihayou put all the money earned in ‘Honkai Gakuen 2’ into the next game. In Liu Wei’s words: ‘I earn so much money entirely by luck. If we sell the company and sell the company, we will be struck by thunder.’

They won again. At the end of 2016, ‘Honkai 3’ was launched, with a turnover of more than 1 billion in the first year. The 3D action performance has reached the level that there is almost no mobile game at that time, and this game has taken a big step forward around the content operation and payment of the character.

But success has also quietly delineated the boundary. In the early days of the launch, the team tried to take another step on the basis of the previous one, and use the new way to drive the version update to expand the audience. As a result, the data did not rise but fell. The team adjusted urgently and returned to the main role and content, and the flow quickly hit a new high.

An early employee recalled that there was a review internally after that, and the conclusion was refined into a sentence: ‘It is still necessary to return to the content, not too much.’ For a long time since then, no one has easily proposed to deviate from this direction.

Every deviation is punished by the market, and every return is rewarded. A choice forced out at the juncture of life and death, under the reinforcement of repeated victories, has gradually grown into a creed. A complete set of play styles has solidified: iterating new characters, providing supporting plot experience and character development, and commercializing the character draw card – by constantly creating new characters, users can pay for the characters. It effectively answers a fundamental question: how to continuously earn revenue through stable content updates.

After ‘Honkai 3’, Mihayou decided to ALK the third time, and put the previously accumulated profits into the project code-named ‘Original God’. This time, the bet is not just about money: they have to be an open world on mobile phones, running synchronously across PC, hosting, and mobile, and synchronously distributed in multiple languages around the world. Every single order is something that other manufacturers rarely try, and Mihayou intends to do it at the same time. Anxiety is stronger than ever. A person close to Cai Haoyu recalled that the core team was very nervous at the time before the launch: ‘They themselves also doubt that they can be well enough to win every bet.’

Fortunately, the results are once again far exceeding everyone’s expectations. In September 2020, the first month of ‘Original God’ was launched with a turnover of more than 300 million US dollars, and the first-year revenue exceeded 10 billion.

Three times all in, each brings more than ten times the income transition. From 280,000 net profit to 10 billion annual income, only seven years apart. There are indeed hidden worries about only relying on one game to carry revenue, but compared to those companies in the industry that do not even plant a tree, this is a luxury trouble.

However, the fate of the company is almost always tied to a game at any time. The excess return of the original god made Mihayou no longer anxious about survival, but did not change the structure of a tree to carry the overall situation, it just made the tree grow bigger.

Miha You didn’t try to plant a few new trees. ‘Honkai: Xingqiong Railway’ and ‘Absolute Zone Zero’ were established in this context. The two games have different categories and seem to enrich the product line, but they share the same set of core formulas – celluloid two-dimensional painting style, random entries-driven character development, and card draw payment – ‘Mi’s three-piece set’ has been completely copied in three copies.

The three games are updated with a large version in 42 days, and each card pool lasts for two weeks. Even if the schedule is deliberately staggered, if the three games are played, the update rhythm is just seamlessly connected. When the Iron Collapse was launched, the interior was expected to expand the new circle, and later it was found that the original batch of players came. The same set of formula feeds three games at the same time, not three trees, but the three branches of a tree, which is competing for the nutrients in the same soil.

So, the greedy snake bit his tail.

The superposition of the revenue of the three games did not bring linear growth, and the internal friction began to show. In the past ten years, it has accelerated from 280,000 to tens of billions of revenue, and the outline of the downhill has been seen from the peak of tens of billions. The formula summed up from the line of life and death makes it more difficult to question each time it is successfully verified, until the day when the three games use it at the same time and touch the border, it has changed from the biggest asset to the constraint.

This recipe didn’t grow out of thin air. It grows from the founder’s characteristics and the early memory of the organization, and has been repeatedly strengthened by ten years of success, precipitated into the company’s process and decision-making inertia. Its roots are deeper than any specific product, and it is stuck on people.

02

person with alarm clock

The two courses with the highest grades in Cai Haoyu’s university, one is the principle of compilation, with full marks; the other is music appreciation, 99 points.

This combination is not common. The physics competition is recommended to Shanghai Jiaotong University, and the game has been recognized by the industry before graduating. The temperament of both technology and aesthetics on Mihayou products is not unrelated to the combination of the founder himself.

An early core employee described him like this: ‘He is extremely focused on what he wants to do, and he is not in that range. Heart.’ He can push down the top entrepreneurs and turn back to work overtime, but he often takes children to and from school and tells them bedtime stories.

In 2011, Cai Haoyu launched Mihoyo studio in the D32 dormitory of Shanghai Jiaotong University. The three founders live in the same dormitory building. When they are studying, they stay up all night to do projects, write code, and play games together. In their own words, they are ‘otaku sympathetic to each other’.

The division of labor was decided very early, and it was deliberately designed by Cai Haoyu. He was the first to find Liu Wei. Liu Wei felt strange: I experienced a failed entrepreneurship when I was a student, the code is unfamiliar, and I can’t help technically. But Cai Haoyu said that what he wants is people who don’t do technology. He used to be a project at Jiaotong University. ‘We want people like you who don’t understand technology to take care of us, you do what we don’t want to do, and you will match us when we have dissent.’

Liu Wei felt that this matter was very suitable for him, so they started like this. Cai Haoyu manages product direction, technology and art; Luo Yuhao manages commercialization and numerical design, and takes into account the discussion of gameplay; Liu Wei manages business, operation and foreign affairs. The three of them walked through the whole process from the dormitory to 10 billion in revenue together, but from the first day, Cai Haoyu was the one who defined the direction.

Every flagship game of Mihayou, he personally serves as the project leader. A person who used to be a strategic consultant of the company said that even if the annual income is 10 billion yuan, Cai Haoyu ‘still lives the same life as ordinary people, and is still in a state of leading troops to fight.’

For a long time, making games was his most focused thing. The way he makes games, in the words of an industry person who has been observing Mihayou for a long time, said: ‘If science students see an alarm clock, the first thing to think about is to remove it, and see how the second hand, hour hand, and movement are installed in it. After dismantling, please add it.’

Mihayou has a project approval manual, and the first requirement is to determine the ‘fire’: an excellent experience prototype that has been verified by the market, and the team’s task is to bring this fire to more popular users.

The starting point of this methodology is an early lesson. The company once made a game that is completely designed from zero to design, and the investment is not small, but the market response is flat.

Since then, Mihayou has formed a set of product rhythms of ‘finding fire and amplifying fire seeds’: first identify an experience direction with potential but not fully explored, and then push it to new heights through technological breakthroughs and content investment.

‘Honkai 3’ brings the feel of the host-level action to the mobile phone, and ‘Original God’ brings the open world experience to the whole platform. The increment of each time is not the invention category, but in the redefining the ceiling of the category.

Find the fire, dismantle the parts, and do the increment in technology and experience. To run efficiently, this method also requires a matching organization.

The above-mentioned consultants recalled that the first time they went to Mihayou, ‘Honkai 3’ was already online, and the annual income was billions. The founding team still maintained the habit of wearing a company’s cultural shirt, and did not have its own independent office.

Don’t spend extra money on decorations, and don’t spend extra time on the process. More than one employee described to us that in the early years, doing things was to move a small bench, and several people sat together to align things together.

Cai Haoyu drew a picture for this structure in the cultural manual. It is not a common tree-like level, but a staggered net. The manual calls it ‘Team of’ Teams’: Each team has its own decision-making core, and the core is equally interconnected instead of superiors and subordinates, and any individual can also communicate directly across the boundaries of the team.

The working method that is used with it draws on Netflix’s ‘Context, Not Control’: it does not rely on the layers of instructions issued by the superior to split layer by layer, but allows everyone to fully understand the background information and judge independently based on professional cognition.

When the company has dozens of people to hundreds of people, this system is extremely efficient. But it has a premise that it is not easy to see: highly dependent on the people on a few key nodes. The larger the company, the heavier the load on these nodes.

Cai Haoyu planned to push the team to thousands of people very early, and even challenge the scale of 10,000 people in the longer-term future.

During the research of ‘Original God’, the load has begun to appear. In an internal review at the organizational level, Cai Haoyu confessed his views on the company: ‘The company is not a home’, it is the carrier of realizing the vision, and there is no need to take everyone’s growth on it. ‘In theory, if the company continues to grow, we can get the best people in the world.’

Some people put forward different views on the spot. The other party believes that the success or failure of things is often in the slightest, and that millimeter depends on the state of the person, the emotions, and the depth of the connection between him and this matter. If everyone feels that they can be replaced at any time, that kind of investment that determines success or failure will not happen.

Cai Haoyu did not refute. The other party continued: ‘If you manage the company in the original way, before about 5,000 people, you can be exhausted, because you have a super human flesh computer. But if you have a few more, there will be problems in parallel.’

The process of being exhausted has actually happened. Among the core members present at the time, there was a project leader of the original god, a supercomputer at a critical node. During the period of the most stressful research and development, he went home at 1:30 in the morning, arrived at the company at 4:30, and spent four or five hours to deduce everything that was about to happen the next day.

Of the more than a dozen of the most core people present, almost no one knew that he worked in this way.

‘If you can’t wake up everyone’s sense of security and meaning, 5,000 people will definitely collapse.’

03

Giant wheel stalls

Before ‘Original God’ was launched, Cai Haoyu was already thinking about the next step.

At the beginning of 2020, Yuanshen opened mass production and started a round of closed tests. After reading the data, Cai Haoyu confirmed that the product would not cause problems, and his attention immediately turned to a new direction: the project code name Project SH (PJSH). The positioning is the open world of the city, the realistic two-dimensional, modern cities, shooting, the style of painting, the theme, and the way of playing are all directions that Miha You has never done.

‘Every flagship game of Mihayou must challenge the technical boundary at that time, not China, but the world.’ A person who understands the development of the company told us. PJSH is no exception. The internal positioning is an ‘up to the ultimate game’, and the technology, pipeline and market goals must be on the basis of ‘Original God’.

This is not the first time Mihayou has aimed at seemingly impossible targets to concentrate resources. The difference is that every ALL In, they all anchored the direction in the early days: when doing ‘Honkai 3’, Cai Haoyu has to restore the feel of the host level on the mobile phone; when doing ‘Original God’, the picture of the character standing on the top of the mountain and overlooking the earth can naturally appear in his mind.

But during the period of PJSH, this kind of picture can’t be spelled out.

A producer who has been in contact with the project told us that what Cai Haoyu wants to do is a ‘a city that makes everyone want to live’. His first reference was the GTA series, but quickly fell into a paradox: the reason why GTA was fun was based on the wanton destruction of the city. ‘Would anyone want to live in such a place?’

What’s more deadly is that the core parts of GTA – guns, cars, crimes – are almost useless in the two-dimensional cities they envisioned. You can’t shoot, you can’t grab a car, you can’t hit people, put aside these, GTA is no longer GTA. The direction was quickly overturned.

When the fire was not found, the team had already spread.

According to Mihayou’s own project establishment specifications, when the direction is unclear, you should start with a very small core team, and then expand it after verification. In the early stage of the project of ‘Original God’, there were only 20 or 30 people, and they were familiar with each other that they could make aligned judgment without an official meeting. A former Mihayou employee described that the company at that time was like a ‘black Internet cafe’.

PJSH skipped this stage. Cai Haoyu feels that the ‘black Internet cafe’-style team can’t support the ultimate game ambition, and what he wants is the strongest person in every link. The priority of the project is directly reflected in the allocation of resources. For a long time, many senior talents will communicate with the PJSH project first.

A large number of talents with bright resumes have been introduced from all walks of life, and the complete post structure and reporting level were quickly filled, and the team quickly expanded to hundreds of people, across domestic, Singapore and North America. Talents introduced in a large number from industry leading companies and overseas 3A studios bumped into each other with their own work habits. There was no common direction to twist the strength to one place, and Mihayou’s proud mesh structure began to fail.

The second edition takes the overseas 3A route, and introduces talents from Singapore and overseas shooting manufacturers to try large-scale cross-border cooperation. At the same time, try to transplant Mihayou’s best at things – multi-role switching and elemental responses – into the realistic urban framework. A former PJSH project backbone told us: ‘A person who has really done this category will basically start to dark when he hears this.’

The construction of commercialization and worldview has also taken a detour. Characters are divided according to occupation, but some characters are not useful in specific scenarios, and the payment logic is difficult to establish; the narrative introduced from outside the industry Talents have produced a batch of exquisite and complete short stories, which are exquisite and complete alone, but the framework for supporting the open world is limited. In the words of the above-mentioned backbone, ‘the things on the face are very good, but they are not alive.’

Cai Haoyu’s own position has also changed in the process. In the past, he was often the one who wrote the first line of code in his hand and set the first version of the setting. Now each module has its own person in charge, and the information is digested layer by layer before it comes to him. It is precisely the early stage of the unknown fire that needs the producer’s first-line body feeling the most, and his eyes are elevated.

Cai Haoyu admitted at an employee meeting: the status of the project is very wrong, the idea cannot be pushed, and the information is not transmitted in place. The team spans three time zones, and collaboration alone consumes a lot of energy. For a person who can align everything with a dozen people in the past, this is completely unfamiliar out of control.

He tries to remedy, compress the management level, and re-embed himself back to the front line. But what I see is far worse than the report in the report: the construction period is expanded to two or three times the estimate, the finalized plan is always out of shape when it falls to the implementation level, and the same problem rebounds repeatedly.

The subsequent iterations have fallen into a cycle: for each version, the technical reserves and pipelines accumulated before will become a burden, and the optional direction is getting narrower and narrower. A developer who has experienced four versions said: ‘The gameplay has not been settled from start to finish.’

Gradually, the sunk cost is getting higher and higher. There are veterans in the team who have done mature shooting games privately muttered: ‘Even if you honestly copy a benchmark in the market, and Mihayou’s art and technology, the result will not be bad.’ But Cai Haoyu could not accept mediocrity. If he can’t make a blockbuster, he’d rather not do it.

In the past, what made him win three games was always the same set of premises: the visible fire, the organization that runs efficiently after anchoring the direction, and finally brings a huge increase in experience. PJSH is the first time these premises disappear at the same time. A large ship sailed into the sea without a channel, and every time it turned around, it burned a huge amount of money and the trust of the team.

Until November 2022 – nearly three years in the project – a very short internal letter was sent out:

‘Everyone has always been successful in doing projects, and everyone feels it is right to walk. Since this road can’t go on, let’s bury it. Don’t waste your time writing a high-sounding email. Fail fast, learn fast.’

The genius producer who started from the university dormitory and believed that he could win every time he bet, finally hit his own boundary in this expensive trial and error.

04

Crossroads

After the dissolution of PJSH, Cai Haoyu experienced the longest direction blank since his establishment.

It was probably the most frustrating moment in recent years. It’s not because of losing money, but for the first time I can’t find the next direction worth investing in. Before every bet, he could see the picture of the end in his mind. PJSH failed to spell this kind of picture for three years, and the blank feeling did not dissipate after the project was terminated.

It’s not a new game that fills this gap, but another line that was buried early.

There is a road map in the cultural manual written by Cai Haoyu himself, and the end point is called ‘Honnons of God’: a virtual world that makes people willing to live in it. Realizing it requires an ability that he calls the ‘fucking vision’ – the creator has a vague thought in his mind, and the system helps him sort out the concept and generate objects that are comparable to reality.

This is almost the picture depicted later by the generation AI, but when these words are written, there are still more than three years before ChatGPT came out.

Along this roadmap, Mihayou began to try with traditional AI technology very early. In 2018, the team launched two research and development lines at the same time around the virtual role. The logic is simple: to treat the two-dimensional character as a living person, one depends on the skin, and the other depends on the soul.

The code name of the skin is ‘Lu Ming’, and the animation characters are made to the extreme with graphic image AI. The project was later disclosed as a technical demo, and it gained tens of millions of broadcasts, and some of the technologies were precipitated into game products. A perceived result is that the several games that Mihayou later released are almost invisible to wear molds, and behind them are supported by the technology accumulated by Lu Ming. This road is relatively smooth.

The road to be the soul entered the dead end. The team wants the characters to not only look good, but also respond to players naturally and deal with various scenarios. The behavior tree is originally used: pre-define each situation for the character, and then manually write each response. Liu Wei later compared this road to ‘carved’, and the engraved no matter how detailed it was, after all, the characters could not deal with all the scenes. In 2021, I can’t go on.

At the end of the same year, Yan Junjie, the former vice president of Shangtang Technology Engineering, left the old owner’s house to start a business and found an old friend Liu Wei. He brought two concepts that were almost untold in China at the time: AGI, versatile artificial intelligence; GPT, a large-model route based on the Transformer architecture. He talks about Scaling Law, the ability to emerge, and tries to persuade Liu Wei that general intelligence is not science fiction, but a road that can be approached by engineering methods.

According to the recollection of relevant sources, the route made by Yan Junjie is completely different from that of Mi Hayou. Many concepts are heard by Liu Wei from his mouth for the first time. But he couldn’t deny the man in front of him. It was not a technical argument that made Liu Wei finally make an investment decision, but a simple intuition. This person is too reliable, and what he is talking about is not like the editor.

The company founded by Yan Junjie was later named Minimax, which is one of the earliest large-model startup companies in China. After Mihayou followed every round, it was not only financial investment, but also a channel for continuous learning. Those things that were completely incomprehensible at the end of 2021 have gradually become perceptible and verifiable in the next two years.

When the ChatGPT was released at the end of 2022, the public was shocked, and Mihayou’s internal reaction was closer to confirmation: what Yan Junjie said a year ago was fulfilled in front of the world.

Almost at the same time, PJSH has just been cut off. The blank period of Cai Haoyu’s direction and the opening of the model era collided. The ‘Honnon-Earth Region’ written in that cultural manual was once just a distant and ultimate vision, but now there is a possible path in technology.

Mihayou immediately formed a team and began to try interactive products based on large models. In 2023 and 2024, they have been focusing on large models as role dialogues and interactive experiences, trying to graft AI capabilities into the content that Mihayou is good at. But after two years, they have found that they are always stuck at the application layer that calls other people’s models, and what they can do is subject to the boundaries of the underlying ability.

At the end of 2024, an internal discussion brought the long-established anxiety to the table. After that discussion, a judgment was settled down: ‘Intelligence is the only important thing, and everything else is a condiment in the face of intelligence.’

This sentence, in turn, is also a question: if you really believe, why are you still making condiments?

There was hesitation at the time. The cost of training the basic model is a hard constraint, not just shouting can be crossed. Two things pushed them. First, the cost curve of the industry is faster than expected, and training a large model is no longer something that only a few companies can afford it. The second is that DeepSeek let them see a living sample: a Chinese team with more than 140 people, with a small scale, extremely high talent density, free discussion, no KPI, completely driven by interest, has made world-class low-level innovation.

This sample convinces them that the direction is feasible, and at the same time, it also reflects their own problems. Mihayou is already a global company of 6,000 to 7,000 people. Once the organization becomes larger, it will produce inertia, which is almost as unavoidable as gravity, and Mihayou is no exception. Cai Haoyu and Liu Wei discussed this issue a long time ago, and the two finally reached an unanimous judgment: within a large company, there are too many ties, and exploring the frontier AI requires a small and focused environment.

Cai Haoyu wants to do this himself. It’s not to hang in one direction and leave it to the team to run, but to end it yourself. This means that he needs to completely withdraw from the daily management of six or seven thousand people. The result of their negotiation is very straightforward: then leave the big company and start a new business.

After quitting the daily management of Mihayou, Cai Haoyu took a small team to start from scratch, read the paper every day, and chat with the people at the forefront of the industry. At the beginning of 2025, the new company started with pre-training and built its own large model from the bottom layer. This road has just begun, and talents are still recruiting.

After Cai Haoyu led the team to fade out, the daily management of Miha You 60,000 people fell on Liu Wei. Describe the current situation, Liu Wei has no room to stay:

‘This is a must-have car. If this train is missed, we will not play in the next five years and ten years. We have not gotten in the car yet, the car is already running in the front, and we are just chasing after it. I hope I can at least get close.’

05

After the sea map is taken away

After Cai Haoyu’s focus was shifted, what left to Liu Wei was not only a company of 6,000,000 people, but also a more urgent question: what will the next game do?

Turning earlier than the outside world thought. A person close to Liu Wei told us that since at least 2021, it has decided that the pure two-dimensional project will no longer be established. It was only in those two years that the core energy was pressed on PJSH, and the backlog of transformation did not really enter the implementation period until the project was terminated.

The first step in execution is to seek change.

Liu Wei calls the set of things that needs to be changed as a ‘three-piece set’. A producer dismantled it like this: the style of cellulin’s painting style is responsible for pulling new ones, the card drawing system is responsible for realizing, and the development cycle with holy relics and abyss as the core is responsible for retention – three parts are combined into a commercial engine that has run for ten years. If it is not changed, it will only be another traditional tour.

Therefore, there is one more unwritten rule for the new project: at least two of the three-piece suit are changed.

‘The original intention is to control the dilution of the brand and encourage the team to do new things,’ said a person who understands the change.

This is not an easy decision. Cai Haoyu wrote a judgment in the cultural manual very early: the formation and maturity of the business model is a process of gradual evolution of a society, the fastest three to five years, more than ten years and twenty years.

‘All Mihayou projects, regardless of success or failure, are attached to the commercialization of three-piece suits and roles. Now that we can’t do this, we can change them to a new one? This is a real confused place.’ An old employee said.

After taking away the old chart, the new chart has not yet been drawn. The regulations fall to the implementation level, and it is inevitable to go out of shape. A producer who has been in contact with multiple project teams has observed that when ‘do not do a three-piece set’ becomes a prerequisite for project establishment, some teams invest as much energy in innovative packaging as innovation itself. And once the project is scaled, the window to adjust the direction will become narrower and narrower.

The first batch of people who got the opportunity to start the project was the veteran who proved himself in the old project. The direction of the multi-person fighting is led by the executive producer of the ‘Honkai 3’ period.

Veterans bring new projects because their capabilities have been verified. The industrial capacity and content quality control accumulated by Mihayou in ten years are scarce resources in the industry. But the paradox is also here: these abilities grow in the soil of the two-dimensional content, and to be transplanted to unfamiliar categories, there is no ready-made road to take.

The ten-year ‘Iron Law’ internalization has become the default value of producers’ thinking. And content-based games still have a major difficulty: it is impossible to use the white box prototype to verify, and it is necessary to pile up a certain amount of content to test. This means that, dozens or hundreds of people have been invested in several months or even several years, once it is found that there is a problem with the framework, it is already difficult to turn around.

Among several research projects, the multi-person co-fighting direction of the code-named KL is relatively smooth. The 3D action battle is the comfort zone of Miha You. A person who has been in contact with the project said that in other directions, it is necessary to hurt the muscles and bones, and ‘KL at least knows what to solve.’

But even on the recent road, there are still many questions that have not been resolved. According to the source, after the game’s single-game experience is finalized, how to build the peripheral system has also experienced a round of struggle.

From a distance, the bigger problem is commercialization: if you don’t sell characters, the skin has become one of the few alternatives, but if the appearance is commercialized, at least two of the three conditions of big DAU, strong social networking, and display space are required.

The other two projects have also experienced different degrees of challenges.

‘Star Cuckoo Land’ was originally aimed at the life simulation track of ‘Dong Sen LIKE’. In the past two years, many manufacturers have been laying out this direction, but the category threshold is higher than that of the outside world. A relevant person said, ‘The domestic plan with life simulation experience is difficult to recruit itself.’

The project was first small-scale test in November 2025, and the feedback is not bad. But a person who understands the inside story has a reservation: the first test is mainly to verify the single-player experience, and the life simulation is to be a long-term, and the social and UGC really decide to keep the modules that are difficult to verify in the short-term test.

How to build a sustainable social system is still the core long-term proposition faced by such products.

‘Honkai: Causal Elf’ is advancing the fastest. The project was launched in 2024, and the following year will be publicly tested for players, in order to improve and optimize the game through real player feedback.

The team has taken a lot of steps, trying to maintain the tonality of the content-based company, and combines the gameplay of ‘Catching Pet + Automatic War Chess’. This combination has indeed attracted new attention. Many players who have never played the Miha game have begun to pay attention to this product, but the evaluation of the new gameplay is also mixed.

After the first test, the project is undergoing reconstruction, and the gameplay experience needs to be further verified.

The three projects have different directions, but the same problem recurs: How to do social interaction?

In the past ten years, Mihayou has been the best single-player experience mobile game in the world, and this accumulation has shaped the ability structure of the entire company. Do everything socially needs – understand the relationship between players, design interactions, tolerate the emergence of uncontrollable – not in this structure.

The current difficulties are not necessarily the end of the game. A person close to Liu Wei said that this batch of projects has gone through the transition period of the transition period, with the initial idea, the compromise after facing the reality, and the stepping and accumulation after trial and error. There will be a batch of projects in the future, and the transformation will be more thorough.

It’s just that in the transition stage, it is necessary to traverse the area where the old chart has expired and the new chart has not yet formed.

06

weight of the hull

When the ‘Original God’ project team was 1,600, there were only more than ten PMs (project managers).

The PM of ‘Original God’ is not a project manager in the traditional sense: those who manage tasks should be able to use the editor to make a running task prototype, and those who manage the level should be able to independently build a white box level. It is not to let PM replace planning, but to ask them to know how life is done. A former employee who worked in ‘Original God’ said: ‘If you plan to deal with you, you must be able to see it.’

Two or three hundred people of similar projects in the industry may be equipped with five to seven PMs, and ‘Original God’ only uses a dozen people to drive 1,600 people, maintaining a major version of the six-week update rhythm, and almost never delays.

But there may be less than 300 of the 1,600 people who really affect the direction of the product.

Another old employee of Mihayou pointed out the hidden boundary of the system: ‘The original god can be understood as adding capacity to more than 1,000 people on a framework of 300 people.

The core team defines the skeleton, and thousands of people who have joined since then filled the skeleton. They don’t need to understand how the skeleton is built, they just need to be delivered to the specifications at their own station.

Quantitative change, not qualitative change.

An early core employee of Mihayou used Ubisoft as a comparison: the same function, Mihayou’s labor input may be equivalent to two to three times Ubisoft, but the product profit margin is 20 to 30 times that of Ubisoft. He calls this ‘Multi-Strength Miracle’ – the ultra-high profit margin has digested the loss of management efficiency. This is an effective strategy at a stage where the profit margin is sufficient to support. But it is almost impossible for new projects to replicate the same profit margin.

It is precisely because people outside the skeleton only need to fill the content, ‘adding people’ has become the easiest way to expand. After the cabin was full of people, the flat structure that Mihayou was proud of in the early years began to deform from another direction.

From the first line to the boss, there are only three or four floors, and the appointment of each team leader needs to be extremely cautious. This means that if a new group is not created, the capable person will not have a day out. An example of a former head project employee gave an example: Initially, there was only one level group. Later, the B and D groups were divided, and the technical stack and focus were different, and there were reasonable business logics.

‘But it’s not right in the back. The task team leader went out to open group C, and the direction completely overlaps with group A. Why? Because I can do group C, I can have one more team leader. Later, I started group E, and the functions are still similar.’

‘You have the first department of Shenfu and the second department of the gods and surgery, because two doctors who are qualified to be directors cannot be stuffed into the same department.’

The more groups, the lower the efficiency. But if you don’t dismantle it, people can’t keep it.

The power of expansion has gradually shifted from business needs to digesting internal promotion pressure. But even if the organization is constantly fissioned, the core positions that can really make people grow are still limited. The most important thing in the game industry is to do things in key positions and open up the horizons in actual combat.

Industrial pipelines rarely offer such opportunities. The above-mentioned people close to the top of Mihayou said: ‘A thousand people’s projects, the pipeline is so thin, how much can you see?’

This organization that was born to focus on a big tree is naturally not suitable for planting many trees at the same time. It has cultivated a group of people who can deliver accurately in the existing system, but rarely provides the overall experience required to become a producer, and it is not easy to grow the ability outside the comfort zone.

It is difficult to grow from this structure, and it needs to be introduced from the outside. Mihayou is also actively doing this.

But running in is not an easy task.

A person close to the new project mentioned that people with a social design background found that after joining a certain project team, they found that the entire team lacks the accumulation of social and player interactions, and all design ideas are based on stand-alone experience. After all, they have not been done in the past.

The talent structure is a major challenge in the transition period. The incentive method of Mihayou is another important thing.

Mihayou’s salary is at a high level in the industry, but the incentive structure is not bound to long-term equity and options. An early core person’s feelings about this are: the price given at the time of entry is not low, and the overall distribution logic of the company is closer to ‘results in return’, and does not share the upper limit of excess income with the team.

This model is effective in the execution stage of high determinism: the goals are clear, and they are delivered according to the diagram. However, when the company needs to explore new directions, some people need to persevere in uncertainty and iterate through trial and error. The incentive logic and business needs are dislocated: the return is relatively determined, the upper limit is visible, and taking risks has become a matter of higher cost than the benefit.

The three forces are pulling each other within the organization: the corporate culture wants people to take the initiative, and the problem should be solved when they see the problem, and the merits can only be achieved by expanding the boundaries; industry The pipeline requires people to passively, and deliver them according to specifications on extremely narrow workstations, without crossing the line, not deviating; the incentive mechanism does not provide long-term binding. The three forces are intertwined, and the tension acts on everyone.

Almost all content-based companies organized from workshops will hit the same wall at a certain stage. Miha You run faster than most companies, so face it earlier.

The problem is not difficult to see. The hard thing is that what this company needs to adjust is precisely to get it to what it is today.

07

Do it for ten years, do three generations

Last year, Liu Wei began to frequently meet with bosses, producers and strategic leaders in the industry. He not only talked about projects, but also talked about project establishment logic, review mechanisms and talent training. A person he talked to said: ‘He is taking the initiative to make up the class.’

There are a lot of courses that need to be supplemented. Mihayou has a very early project manual, which has a phased review process of green light, yellow light and red light, and is mainly reviewed by three founders.

After Cai Haoyu faded out of daily management, Liu Wei took over the decision-making of the project.

He began to introduce player tests, user interviews and data analysis at the key nodes of the project, and even early projects try to do more verification in the white box stage. Products that have been launched are also being adjusted. A support department employee recalled that in the past, the project team believed in the judgment of the main creator, and now he has begun to take the initiative to test more at the key nodes, ‘I still need to hear more about the players.’

If it is said that Cai Haoyu’s product judgment is a talent, Liu Wei’s more come from acquired learning, he is very clear about his boundaries, so he took the initiative to introduce systematic verification to make up for it. According to the observation of this employee, Liu Wei has obviously grown in recent years. ‘He’s a lot of aesthetics to improve.’

In addition to the introduction of research, changes also take place outside the walls of Mihayou.

In the past few years, Mihayou’s foreign investment hardly touched games. With their self-developed strength and profit margin, it is better to invest in others than to do it yourself.

But since the beginning of this year, two investments have broken the convention. The former ‘Outside the World’ producer Xuanzi started a business as a female direction, and the former producer of ‘The Shore of the Land’ Li Kaiming started a stand-alone strategy, and successively received capital injections from Mihayou. There is almost no intersection between the two categories and Mihayou’s ability circles.

‘What he is looking at is not a single project, but the long-term vision and ability of this person,’ said a person who has known Liu Wei for many years.

The same logic has appeared in earlier investments. In 2021, when he invested in minimax, the AGI route that Yan Junjie talked about was almost inconclusive to Liu Wei, but he judged that the person was worth betting.

Not necessarily fully agree with and understand the direction, but fully trust those who do things. The above-mentioned person said that Liu Wei is a person who is interested in the big and let the small, and he will not struggle too much after a person is recognized.

After all, investment is separated by a layer. What Liu Wei wanted to do more was to invite people in.

One of the senior producers, who has been in business for nearly 20 years, has been deeply involved in the categories that Mihayou has never set foot in before, and has accumulated a deep accumulation in core experience polishing and long-term operations. According to a relevant person, he was in contact with Mihayou a few years ago, and he was not optimistic about Mihayou in this direction at that time. After a few years, Liu Wei still invited him.

This is a abilities model that is completely different from Cai Haoyu. Before Cai Haoyu made every flagship game, the end picture was already formed in his mind. He is a rocket, the direction is defined by him, and the team revolves around him. Liu Wei can’t be the rocket head, but what Mihayou needs may not be another rocket head.

Now no one can see the determined end point, and the direction should gradually emerge in the process.

From ‘a genius defines the direction’ to ‘a system to discover the direction’. The statement of ‘from one tree to a forest to a hundred flowers blooming’ at the annual meeting is not only the change of the slogan, but also the switching of organizational operation logic. Whether this conversion can run through, it may take five to ten years to answer.

A person who understands the project idea uses Nintendo as an analogy: each studio is responsible for a vertical category, and the producer has been deeply involved in the same category for decades. ‘The Zelda team has done RPGs for decades, and Mario Kart is another independent studio.’

Mihayou no longer expects to be a flagship as soon as it comes up, but let people who are good at a certain category continue to accumulate in specific fields. Not every tree will grow into ‘Original God’, but each tree must stand firm in its own category.

‘To be a new type, it is a delusional idea to expect a generation of products to be able to achieve the results of ‘Original God’. At the summary meeting at the beginning of the year, Liu Wei said it very bluntly. Every new direction is like Mihayou in 2011, and the accumulation is almost zero, and it is necessary to start over. ‘You must have the patience and determination to do ten years.’

He said, looking back at the past, the core is one thing: continuous cultivation in a vertical category. From 2011 to 2016, it took Mihayou five years to find the smallest prototype of the two-dimensional category – ‘Honkai 3’, which established the role and internal The three-in-one framework of content and payment. After that, ‘Original God’, ‘Broken Iron’ and ‘Zero Jue’ are all from 1 to 10 and 100 in this framework. ‘In 2022, the whole industry feels that ‘Original God’ is successful, but it has only allowed us to achieve results in this category in ten years.’

In 2023, he began to think about the next step in the company. The final conclusion is to follow your own way. For a new drape, it may take four to six years to go from 0 to 1, and it will take the same time to mature. This is the logic behind ‘doing ten years and doing three generations’.

‘The model in the past was when the goal was very clear. Now the goal is not clear, the direction is uncertain, and it needs to be explored.’ said a person close to Liu Wei.

Take the time to come and go again. This strategy sounds reasonable, but it has a prerequisite: have enough patience.

Miha has almost never really failed. From ‘Honkai Gakuen 2’ to ‘Original God’, to ‘Broken Iron’ and ‘Absolute Zone Zero’, every bet is won, which makes the company lack a personal understanding of failure. “When a company tells you that it has never failed, either life is not long enough, or there are not enough projects,” said the people.

Many respondents expressed a similar view: Confusion and pain are the stages that must go through in transformation. It is not terrible. At least Mihayou still has the capital of trial and error, there is no listing, and the cash flow is abundant, and time is on its side.

The real test is not to think clearly enough at the moment, but to dare to move forward when you can’t think clearly. Liu Wei retweeted a sentence internally: ‘The desperate mentality may not be successful, but the hesitation of looking forward and afterward will definitely fail.’

We asked a person who has been observing Miha for a long time, see if it has changed.

‘A lot of good changes. The process is inevitably anxious, confused, and overkilled, but I can clearly feel that they are changing.’

End

In a sharing, the main creator of ‘Original God’ once mentioned a law that Cai Haoyu left to the team: happiness and conservation.

‘If you are very happy to do one thing, you will have a high probability that you will not satisfy others. But if you are anxious, painful, confused, and leave unhappiness to yourself, the things you make may make others happy.’

Not every time. ‘But we will always stick to this law.’

In the most difficult period of the development of ‘Original God’, once Cai Haoyu lost control of his emotions at the meeting, and after scolding everyone, he scolded himself: ‘This project can’t go on.’

Then of course it went on.

In the past ten years, the company has grown into a towering tree by turning the pain and anxiety of a few geniuses into the ultimate industrial output. Now they say they want to plant a forest.

The forest is not planned, it is tried. The seeds have been planted, some are sprouting, and some are still in the soil. The answer is time to answer.