Game companies are making huge profits in the large model track.

In 2026, among the most profitable investors in China's large - model industry, there was an unexpected group: game companies.

MiHoYo participated in the angel round investment of MiniMax in December 2021. The pre - investment valuation was only $170 million, and four institutions jointly injected $31 million.

In January 2026, MiniMax went public on the Hong Kong Stock Exchange with an issue price of HK$165. On the first day of listing, its stock price soared by 109%, and its market value exceeded HK$100 billion.

After that, the stock price soared all the way to a record high of HK$1,330. MiHoYo held about 6.4% of the shares before the IPO, and its conservative book floating profit reached HK$10 billion.

Return multiple: more than 100 times.

In the same month, Zhipu AI rang the bell on the stock exchange under the halo of "the world's first large - model stock". It went public on the Hong Kong Stock Exchange on January 8, 2026, with an issue price of HK$116.2. In the following three months, the stock price continued to rise, reaching a maximum of HK$1,010 in April.

And in its list of early shareholders, there was another game company —— 37 Interactive Entertainment.

The first - quarter financial report of 37 Interactive Entertainment just released in 2026 showed that its investment income soared by 981% year - on - year, an additional profit of 326 million yuan. The core source of this money was the change in fair value after Zhipu's listing.

One invested in MiniMax, and the other invested in Zhipu. These two game companies almost grabbed the two most valuable tickets in China's large - model industry.

What's more unexpected is that they didn't just invest in these two companies.

1. 37 Interactive Entertainment: Invested in Three of the "Four Little Dragons"

In China's large - model industry, there is a well - recognized term "Four Little Dragons of AI": Zhipu, DarkSide, Baichuan Intelligence, and MiniMax.

37 Interactive Entertainment invested in three of them at once.

First, it was Zhipu. In 2023, 37 Interactive Entertainment increased its capital in the Jiaxing Beisheng Hengwo Fund by 25 million yuan. This fund directly or indirectly invested in the equity of Zhipu AI.

In January 2026, Zhipu went public on the Hong Kong Stock Exchange with an issue price of HK$116.2. After that, the stock price once soared to HK$1,010, and the market value reached a maximum of nearly HK$380 billion. 37 Interactive Entertainment's recognized investment income in the first quarter alone increased by 326 million yuan.

Then it was DarkSide. 37 Interactive Entertainment invested in this company early by participating in the Shixiang Global series of funds. In March 2026, market rumors said that it was promoting a new round of financing, and the industry expected its pre - investment valuation to be about $8 - 10 billion.

This is not the end. Baichuan Intelligence, a company based on a general large model and deeply involved in the vertical field of medical health, had a post - A+ round valuation of 20 billion yuan in April 2026, and 37 Interactive Entertainment was also on the shareholder list.

It invested in three of the Four Little Dragons, only missing MiniMax. Because that ticket had been in MiHoYo's hands since the angel round.

But 37 Interactive Entertainment's investment map is far more than just large models.

According to public information, 37 Interactive Entertainment has invested in more than 30 hard - tech companies, covering computing power chips (Xingkong Technology, Fangqing Technology), brain - computer interfaces (Qiangnao Technology), embodied intelligence (Guanglun Intelligence), semiconductor materials (Jingzheng Electronics), and XR hardware (LeiNiao Glasses).

It's no exaggeration to say that 37 Interactive Entertainment has almost scanned the entire AI industry chain from top to bottom, and its investment scope covers half of the core of the AI circle.

A listed A - share company that started from game distribution with a revenue of 15.966 billion yuan in 2025 has a higher investment frequency and wider coverage in the AI field than most market - oriented VCs.

2. MiHoYo "All - in" on MiniMax and Made the Most Profits

Different from 37 Interactive Entertainment's "scatter - gun" approach, MiHoYo's strategy is completely the opposite. It focuses extremely.

In the large - model industry, MiHoYo mainly bet on MiniMax. But this single bet became one of the most classic deals in China's AI investment history.

Back in late 2021, MiniMax had just been established. Its founder, Yan Junjie, left SenseTime to start a business in the field of general large models. At that time, ChatGPT hadn't become popular yet, and the market's understanding of large models was still relatively vague.

MiHoYo entered in the angel round and, together with Hillhouse Capital, IDG, and Yunqi Capital, injected $31 million. The pre - investment valuation was only $170 million. Yan Junjie has a close personal relationship with Liu Wei, the chairman of MiHoYo, and Liu Wei still serves as a non - executive director of MiniMax.

More than three years later, MiniMax became one of the fastest - growing AI companies in the world. Its AI companion app, Talkie (Xingye), exploded in the North American market, with over 212 million registered users globally.

In January 2026, it went public on the Hong Kong Stock Exchange with an issue price of HK$165. On the first day, the stock price soared by 109%, and the market value exceeded HK$100 billion. After that, it continued to rise, reaching a maximum of HK$1,097, and the total market value once exceeded HK$313 billion.

MiHoYo held about 6.4% of the shares before the IPO. Based on MiniMax's historical high - point market value, the book value of its corresponding holdings once exceeded HK$18 billion. Even conservatively calculated, it far exceeded HK$10 billion.

Since entering in the angel round at the end of 2021, the return multiple of this investment has steadily exceeded 100 times.

More interestingly, MiHoYo is not only a shareholder of MiniMax but also its client.

MiniMax's large - model technology has been applied to "Honkai: Star Rail" since 2023, with core scenarios including intelligent dialogue of NPCs and dynamic generation of storylines. In MiniMax's prospectus, MiHoYo is listed as a "leading senior independent investor", both investing and using, and deeply bound.

MiHoYo's layout in AI is not limited to this.

As early as 2018, the company established the "Entropy Research Department" to develop its own large AI model, Glossa. The digital human "Luming" has about 1.7 million fans on Bilibili, and the cumulative views of 32 videos have exceeded 100 million.

Its self - developed subsidiary, Wudinggu Technology, was established in July 2025 with a registered capital of 500 million yuan. After Cai Haoyu, the founder, stepped down from important positions in the company in 2023, he founded Anuttacon overseas to specifically explore AI - native games.

In addition, MiHoYo also participated in the establishment of multiple AI investment funds, indirectly deploying in fields such as AI applications, semiconductors, and XR through funds like Shanghai Yuanchuang Future, Xiamen Wanwu Phase I, and Shanghai Tongge Phase II.

One is like a sniper, and the other is like a shopper. With two different styles in the same industry, both have made a fortune.

3. Not Just Them: The Game Industry's Collective Bet

37 Interactive Entertainment and MiHoYo are not alone. Looking at the whole picture, the entire game industry is pouring money into the large - model industry, and each company has its own strategy.

Century Huatong aims at computing power. As early as 2022, it invested in the domestic GPU company Moore Thread through an industrial fund.

In December 2025, Moore Thread went public on the STAR Market. On the first day of listing, its stock price soared by 468%. Century Huatong contributed about 640 million yuan in net profit from this investment in the fourth quarter of 2025, accounting for 53% of its annual net profit attributable to shareholders in 2024.

Century Huatong also jointly built the Yangtze River Delta Intelligent Computing Center with Tencent, planning to have 40,000 cabinets, and 10,000 have been delivered. A game company is doing the work of computing - power infrastructure.

Kunlun Tech is more radical and simply gets involved itself.

Since September 2023, it has cumulatively invested 680 million yuan to increase its capital and take a controlling stake in the AI computing - power chip enterprise Aijieke Core, directly connecting the entire chain of "computing power, large models, and applications". Its Skywork series of large models cover multiple fields such as multi - modal reasoning, search, and music. It is one of the few game companies in China that are truly all - in on AGI.

Youzu Networks bets on domestic computing power. In 2025, it successively invested in two domestic GPU companies, Biren Technology and Xiwang. Biren Technology has gone public on the Hong Kong Stock Exchange, with a pre - investment valuation of HK$20.9 billion. Youzu also invested 1 billion yuan to participate in the establishment of Wuxi Yunxing Intelligent Computing, entering the AI computing - power industry.

Giants Network bets on the product side.

It invested in the AI video - generation company Aishi Technology and the AI image - generation platform LiblibAI. Its game "Supernatural Action Group" is the first domestic product in a large - DAU game to deeply integrate large AI models. The NPCs are driven in real - time by large models, and the number of AI - involved game rounds exceeded 25 million in the first week of its launch.

As the absolute leader in the game industry, Tencent is also a super player in the AI investment circle. It participated in MiniMax's Series B (holding about 5.7% of the shares after the IPO), invested in Zhipu AI, and spent 18 billion yuan on self - developed large models such as Hunyuan and new AI products like Yuanbao in 2025, planning to at least double the investment in 2026.

But Tencent is more like both a player and a referee, which is a story on another level.

Here is an interesting comparison. According to IT Juzi statistics, from January to October 2025, there were 139 financings of over 100 million yuan in China's AI field, with a total amount exceeding 60 billion yuan.

However, traditional VCs are shrinking. During the same period, the investment amount in the AI model layer decreased by about 52.9% year - on - year, and the number of investment events decreased by about 35% year - on - year. The funds are accelerating towards the top.

While VCs are hesitating, game companies have already made their investments.

4. Why Game Companies?

Why are game companies leading the way in this field instead of other industries?

The most direct reason: They are extremely wealthy.

The game industry is one of the few cash - cow industries in China's Internet. According to public media reports, MiHoYo's game revenue exceeded 78 billion yuan in 2025, and its net profit is expected to exceed 40 billion yuan. It has more cash on hand than the funds managed by most VCs.

Since 2020, 37 Interactive Entertainment has cumulatively distributed more than 10.26 billion yuan in cash dividends and made external investments of over 20 billion yuan. This money is self - earned, and there is no need to report the exit schedule to LPs or meet DPI requirements.

When VCs invest in large models, they need to go through the investment committee, calculate the fund's duration, and consider the IRR (Internal Rate of Return). When game companies invest in large models, they are spending their own profits and can invest whenever they want.

But having money alone is not enough. What really enables game companies to outperform VCs in the large - model industry is that they are the biggest buyers of large models themselves.

The financial report of 37 Interactive Entertainment already shows the results.

The proportion of advertising materials generated with deep AI participation exceeds 70%, and the proportion of AI - assisted advertising placement reaches 50%. The Kimi large model of DarkSide is deeply integrated into 37 Interactive Entertainment's customer - service system and the AI assistant "Ask Xiaoqi". The GLM large model of Zhipu is directly used for game - storyline expansion.

MiHoYo is even more direct. MiniMax's model runs directly in its own game products, from NPC dialogue to storyline generation. It's not just a concept verification but a launched product.

When game companies invest in AI, they not only get equity returns but also real - cost savings. By investing a sum of money, they first earn from increased efficiency and then from the exit.

VCs can't achieve this. Sequoia invested in MiniMax, but Sequoia itself doesn't consume tokens and has no scenarios to use large models. Its only way to get returns is: Wait for the company to go public and sell the stocks.

There is also a deeper driving force pushing game companies forward: anxiety.

The game industry's good days are not as easy as they seem. With tightened game licenses, soaring user - acquisition costs, and a ceiling on user growth, the traditional model of spending money to buy traffic is no longer viable.

In the first quarter of 2026, 37 Interactive Entertainment's revenue decreased by 12.3% year - on - year, but its net profit attributable to shareholders increased by 59%. The reason behind the increase in profit despite the decrease in revenue is that AI is reducing costs, and investments are contributing to the growth.

When the growth logic in the game industry changes from who spends more money to who is more efficient, AI is no longer an optional course but a mandatory one.

With money, application scenarios, and anxiety, these three driving forces combined, game companies are entering the large - model industry much faster than anyone expected.

5. A Bigger Story: Industrial Capital vs. VC

Looking at a broader perspective, the fact that game companies are investing in large models actually reflects a bigger trend. In the game of AI investment in China, industrial capital is pushing VCs to the sidelines.

After MiniMax went public, among the investors with returns exceeding $1 billion, Alibaba ranked first (holding about 13.66% of the shares before the IPO), MiHoYo ranked second (holding about 6.4% of the shares before the IPO), and Sequoia China ranked third (holding about 3.81% of the shares before the IPO).

After Zhipu went public, investors with returns exceeding $1 billion included Legend Capital, Meituan, and Ant Group. Both industrial capital and VCs have top - tier institutions among the leaders.

From 2025 to 2026, in the largest exits in China's large - model industry, such as Zhipu, MiniMax, and Moore Thread, the biggest winners are almost all industrial capital. It's not that VCs are not good, but industrial capital has an identity that VCs will never have. It is also a buyer of AI.

VCs invest in AI as a bet on the future, while industrial capital invests in AI to solve current problems. This difference is not very obvious at the exit stage, but at the moment of investment decision - making, the gap is huge.

When MiHoYo bet on MiniMax at the end of 2021, few people in the market believed that large models could be a profitable business.

But MiHoYo believed it, not because it had some advanced understanding of AI, but because it clearly knew what it needed. It needed an underlying technology that could drive NPCs, generate storylines, and make the virtual world more realistic. And MiniMax was exactly doing that.

37 Interactive Entertainment invested in Zhipu and Dark