Has miHoYo's golden run come to an end, and does AI make it harder for small game developers to break through?

Four tech enthusiasts set out from a dormitory at Minhang Campus of Shanghai Jiao Tong University, and eventually built one of the largest game giants with a valuation of 175 billion yuan. After *Genshin Impact* was launched in 2020, its global revenue exceeded 1 billion US dollars in 171 days. miHoYo, once a small game startup, quickly rose to the top tier of the global game industry.

Over the past decade, inspiring startup stories have been far from rare in China's game industry. But this growth path is now almost impossible to replicate.

In 2026, AI has penetrated into the deep-water zone of the game industry. While developing a single game has become easier, it is far more difficult for a game company to grow from scratch to a leading position. The recently released semi-annual reports of domestic game companies are sending out this exact signal.

The "era of universal growth" is coming to an end 

If we only look at the overall market size, the 2026 game market is not performing badly. Statistics show that the revenue of China's domestic game market in the first half of the year exceeded 188.4 billion yuan, with a growth rate of over 12%, which is almost the most outstanding report card in recent years.

However, if we shift our focus from the entire industry to individual listed game companies, the situation is far less optimistic.

Recently, 64 listed game companies have released their semi-annual reports. Statistics show that in the first half of this year, the total game revenue of these companies increased by only about 9% year on year, a significant slowdown compared to previous periods. Among them, leading enterprises including Tencent, NetEase, and Century Huatong contributed about 79% of the total revenue.

A large number of mid-tier and small game companies are still under severe performance pressure. The number of loss-making companies reached 22, doubling compared to the same period last year, of which 16 companies turned from profit to loss. In addition, among more than 20 A-share listed game companies, only 8 implemented interim cash dividends, while the remaining 13 chose not to pay dividends.

A more notable change lies in personnel costs. The total salary cost of these 64 companies reached about 80.3 billion yuan, 900 million yuan less than the same period last year. The main reason is the reduction in headcount, and the drop in average monthly salary is also a contributing factor. Among the companies that disclosed the number of employees, the average per capita monthly salary cost is 34,000 yuan, a year-on-year decrease of 1,500 yuan.

This trend is driven by the development of AI. At present, AI has been applied in multiple business links of game enterprises, including art, programming, testing, game design, distribution and operation. The survey data released by China Audio-Video and Digital Publishing Association this year shows that the penetration rate of AI technology in game enterprises has reached around 86%, and the application proportion of AI in art design, distribution and operation links is quite high among representative leading enterprises.

Not long ago, Values Value released the report *Talent Signals Reshaping the Games Industry*, which surveyed more than 1,800 game practitioners from 90 countries. 78% of the respondents said that AI has changed their way of working this year, and 35% of founders said they have decided not to recruit for certain positions because AI tools can complete the corresponding work.

In the game industry, AI is no longer a new tool tried by a small number of companies, but is becoming a new round of infrastructure in the game industry following game engines, cloud computing and mobile internet.

Big players reap huge gains while small firms are squeezed out of the market? 

AI undoubtedly brings a round of efficiency dividends. In the past, production capacity was the scarce resource in the game industry. Whether a team could produce high-quality art, complex gameplay and a huge worldview was itself a core competitiveness. Now, these capabilities that used to rely heavily on manpower and experience are being partially "commoditized".

Leading companies have made extremely aggressive layouts and investments. Tencent not only develops AI large models in-house, but also integrates AI directly into game R&D and player experience.

For example, in *Delta Action*, Tencent uses Agent to improve operational efficiency, and promotes the end-to-end production pipeline of scenarios through Hy3D to boost efficiency. *Peace Elite* is exploring AI NPCs, and the total number of users who have experienced related gameplay has reached 167 million, with the highest daily active users reaching 17.7 million. Tencent also added the in-game native Coach Agent "Scarlett" to *Roco Kingdom: World*, allowing AI to provide specific gameplay suggestions based on players' backpacks, lineups and match environments.

NetEase did not choose to spend heavily on chasing a general large model, but emphasized that it aims to be "the AI expert that knows games best". In its 2025 financial report released this year, NetEase disclosed that its AI native pipeline has achieved large-scale deployment in art, game design, programming, animation and other links, with the efficiency of some links increased by 300%. In the technical R&D link, AI code generation tools improve development efficiency by 50%, and the quality of AI-generated logic can reach the expert level.

It is understood that many popular games under NetEase are now full of AI applications. *Eggy Party* has launched the "AI model generation" function, allowing players to quickly create and share UGC content. At the same time, a series of AI native gameplay has been created and implemented on a large scale. *Where Winds Meet* has deployed tens of thousands of vivid and realistic intelligent NPCs. The *Naraka: Bladepoint* mobile game provides players with AI teammates that are almost indistinguishable from real people, with full multi-modal capabilities and real-time voice communication support. In the "Film Crew Mode" of *A River of Sorrows* mobile game, players only need to input text, voice or video into the game to generate characters, animations and short videos with one click.

"In the next three years, we will invest a maximum of 100 billion yuan to deeply explore the AI field. Even if we fail to achieve the final success in the end, we will accept the result, just treat it as setting off a grand firework", Liu Wei, co-founder of miHoYo, once said at an event. In addition, miHoYo has also invested in AI large model companies. Before MiniMax went public, miHoYo became its angel round investor with a post-money valuation of 200 million US dollars, and continued to increase its investment in subsequent rounds of financing.

Recently, Century Huatong and Perfect World also joined hands with the new investment fund Monolith Capital respectively, contributing 150 million yuan and 50 million yuan respectively, both targeting the artificial intelligence sector.

However, there may be a significant "temperature difference" between leading companies and small and medium-sized game companies. Looking at the AI layouts of Tencent, NetEase and miHoYo today, you will find that their AI applications are completely at a different level from those of small and medium-sized game companies.

Zheng Yinhe, a technical leader at miHoYo, publicly shared at a technical summit that in order to test multi-agent collaboration, the team set up dozens of AI Agents to run at the same time, and left work without setting an upper limit for Token consumption. As a result, the Agents ran continuously for 13 hours, burning up 2 million yuan worth of Tokens.

"If it were an independent game development team, it might have gone bankrupt overnight."

When it comes to AI applications, the "experimental cost" of leading companies may be the "survival cost" of small firms. This is exactly the industry temperature difference that AI is creating.

Is the entrepreneurial path for game companies getting narrower? 

This investment gap will inevitably be transmitted to the entire growth path of game companies.

From the production side, AI is indeed a rare opportunity for game entrepreneurs. It lowers the production threshold of art, programming, game design, 3D and other links, allowing more people to enter this industry, and enabling small teams to develop products that previously required a large-scale team to complete.

But at the same time, leading companies are also reaping this round of AI dividends. Moreover, for large companies, AI can not only help reduce costs, but also allow them to continue to increase investment on the original R&D system, and use the same or even more resources to carry out more attempts. They have more funds to afford trial and error, can test multiple AI tools and technical routes at the same time, have mature IP to carry new technologies, a huge user base to verify products, and enough time to wait for AI to generate real returns.

Small teams may not have such advantages. For them, the Token expenditure spent on AI tools may mean a trial and error cost that is difficult to allocate. Especially when the team starts to try multi-agent collaboration, complex 3D generation or AI native gameplay, the consumption of Tokens and computing power may increase rapidly.

More critically, there is a complete industrial chain between a game demo and a commercial product. Distribution, operation, continuous content update, solving server and technical costs, building communities, handling user feedback, and dealing with various unexpected situations that may arise after the product is launched, these links have become more important after AI lowers the production threshold. And these are exactly the traditional advantages of large companies.

In addition, the competition brought by AI will not only stay at the level of "who can make a game faster". For leading companies with more sufficient capital and resources, they are also relying on AI to extend from game R&D to IP and content, and even new business models.

Tencent has launched "Fire Dragon Manhua Drama" to explore the collaborative incubation of "game + manhua drama". NetEase Lingyang develops products based on the game IP ecosystem. The proportion of AI-generated video materials in the internal R&D end of 37 Interactive Entertainment has reached 80%. Friend Times has launched AI manhua dramas for its works *Go Lala Go!*, *Heartbeat Fall* and original series such as *Senior Brother, Stop Kneeling, I Practice the Ruthless Path* on multiple platforms.

The same set of AI capabilities, in the hands of large companies, has been amplified into the productivity of the entire game ecosystem.

In this era, startups can still make good products, and even make hit games faster than before. But making a hit game is only the first step. How to control R&D costs, afford customer acquisition and operational investment, and turn a one-time product success into sustainable commercial capabilities, will increasingly test the team's capital reserve and operational capabilities.

When AI gradually changes from a "tool" to a fixed cost and innovation infrastructure of game companies, leading companies can continuously amplify AI dividends with capital, IP, users and ecosystem, while mid-tier and small companies may increasingly lack the ammunition for trial and error.

Therefore, this may be the most noteworthy change in the game industry in the current AI era: the upper limit of entrepreneurship is getting lower. Demos will become cheaper, hit games will still be rare, but the path from a hit game to a leading company is getting harder and harder to cross.

This article is from the WeChat official account "Tech Planet" (ID: tech618), written by Hua Wei, authorized for release by 36Kr.