Become a member of GB MAX to gain exclusive access to the industry and to the most influential global B2B leadership community in the business of gaming, entertainment, and tech.
Join nowand also get a VIP ticket to GamesBeat Next (Nov 2-3, SF).If you don’t check in with Fangda Wan and Shao Yun at Magic Find Ventures, you might not be keeping up with everything you need to know about AI and games, particularly in China.
I occasionally suffer from such FOMO, and so it was interesting to get a different, non-Western perspective on AI from Wan, cofounding general partner at Magic Find Ventures in China and formerly a vice president on the investment team at NetEase Games. I also spoke with her partner, Shao Yun, founding general partner at Magic Find Ventures also a former SVP at NetEase Games. They’re getting a lot of pitches for AI and games, especially in China, and are raising a fund to invest in game AI startups.
They’ve raised $30 million so far for a $50 million target for Magic Find Ventures I, a venture capital fund that will invest $500,000 to $2 million in pre-seed and seed startups that combine gaming and AI in novel ways.
Wan said the fund is planning to invest in 25 to 30 portfolio companies, and they have been investing since the first close in January 2026. So far, the fund has made five investments in the following categories: Super creators — individuals or small teams building AI-native games and interactive experiences; next-generation interactive experience platforms; and creator infrastructure serving those teams.
The fund operates in Singapore, Shanghai and Hong Kong. The investment team is four people, including Wan and Yun as general partners. We keep the team deliberately small to ensure swift decision-making and give founders direct access to the GPs.
Wan shared a couple of examples of the combination of AI and games at Chinese companies. They illustrate different futures of AI-native design.
The first is History Simulator: Chongzhen (《历史模拟器:崇祯》) by Qinggan Studio, released on Steam in May. You play as the last Ming emperor, writing imperial edicts in free text — any policy you can imagine — and an LLM simulates how the empire, the court factions, and history itself respond.
Nothing in the game is pre-scripted. It’s an interesting answer to the token-cost question that I raised during our interview: they price it as a base purchase plus AI-simulation credits, essentially passing compute cost through as a game resource.
Player reviews are genuinely mixed — the simulation still has rough edges — but that’s exactly what the “sprouting stage” Wan described looks like: the first commercial proof that an LLM-driven systems game can ship at all.
The second game Wan pointed out is Last Serenade (《恋恋终序》) by LumiTopia — a portfolio company of Magic Find Ventures. It’s a 3D romance game built on Unreal Engine 5 where AI does two jobs. The first is companionship: the characters hold persistent, open-ended conversations — the team fine-tuned their models on a large corpus of world-setting narration and hand-written dialogue for vivid character expression, backed by an agentic memory system that tracks each player’s history.
The second is what makes it “infinite-flow”: inside the same scenario — the same train, the same manor — the AI lets you take on different roles and drives the story down a different path each time. One stage, endless plays; closer to immersive theater than a branching visual novel. The team started lean and had a playable 3D demo in six months; their closed beta hit next-day retention above 70%, and it has kept climbing across test rounds. To Wan and Yun, it’s a shining example of their thesis in action: AI as the foundation, not a feature.
What they’re seeking: On the company’s website, the vision is threefold: they seek players who reshape reality, obsessions that demand creation, and the courageous who storm into the new era. They are genre- and platform-agnostic; what they really focus on is the core essence of creator motivations, their design philosophies, and whether they are willing to fully embrace AI to achieve their ambitions.
Here’s an edited transcript of our interview. (Wan translated for Yun).
Fangda Wan: I think it’s probably quite a consensus now that AI will be the most disruptive force for the gaming industry in the decades to come, and we’ve already passed an inflection point where 2D image generation, coding, and LLM narrative-telling are getting heavily embedded in the game production pipeline.
What really triggered us to launch this fund are two things — the democratization of creative tools that would enable small studios to compete with big incumbents on triple-quality, highly sophisticated content, and the emergence of true AI-native gameplay that will signal the “Angry Birds” moment. It’s also quite a blessing that our base is in Asia, China, where its share in the global gaming market is accelerating continuously, and here the utility of AI in games is met with real enthusiasm and support.
Shao Yun, my partner and founding GP, ex-SVP of NetEase Games, brings a rare combination of production experience, talent mentorship, and creative genre-incubation. He led the teams behind hit titles such as TianXia, Knives Out, and Eggy Party, and coached teams behind Identity V, Where Winds Meet, and Diablo Immortal. His foremost passion is to help young talent sharpen their design craftsmanship and unfold new forms of experience for players.
My own path is the other half. I spent years investing in how the younger generation wants to be entertained — TapTap, Kizuna AI, IMVU, early bets on Replika and other AI companionship tech. I love digging into the latest cultural and consumer trend shifts.
We suspect there might be a big gap in how AI is perceived in the gaming world of West vs. Asia, and early-stage funding into gaming right now is quite scarce. We’d be happy to share with you our perspectives, our observations in China, and how we might be of help to Western developers as well.
GamesBeat: That’s fine. Let’s see. Do you want to maybe introduce me to what you guys are doing? (Wan translates for Shaoyun)
Wan: Yeah, sure. Maybe we can introduce ourselves a bit first. [To Shaoyun: I will speak for you directly.] I’ll start with Shaoyun. He is the founding partner of our fund. He graduated early from college and started his career at NetEase at 21, back in 2006. Fortunately, he was able to witness two transitional periods of the business model within NetEase. First, NetEase grew from a single-title MMO-based studio to a multi-title studio around 2010, establishing itself as a top gaming company in China. The second transitional point was from PC to mobile, around 2015 to 2018. During that time, it grew both in revenue split and from a single MMO-focused company to a multi-genre innovation content gaming company.
If you look at NetEase’s portfolio construction right now, besides MMOs, we also have multiple titles that are strong leaders in their respective genres. For example, Identity V (an asymmetrical battle game), Knives Out (a battle royale that has been a long-time leader in the Japanese market), and Eggy Party (the first successful UGC platform in China). Luckily, all three titles I just mentioned were incubated by studio heads under Shaoyun’s management. Ever since he entered NetEase, besides being a game producer himself, he was responsible for campus recruiting as well as a game producer mentorship program.
Over his 20-year history, not only did he grow into a big producer overseeing titles like
Knives Out, Tianxia, and Eggy Party, but the producers from his mentorship program also
produced titles like Diablo Immortal and Where Winds Meet, which is growing in popularity in the Western market. He ended his career at NetEase at the end of 2024 because of internal structural shifts, started looking at AI games last year, and that’s when we reconnected.
Wan: For me personally, I have a very hybrid background. I started my career as a corporate lawyer doing IPOs for internet and gaming companies in China, and I was also once an entrepreneur doing lifestyle businesses in China and the US. I joined NetEase’s corporate strategy team back in 2017 under Simon.
We grew the team from about six people to about 200 people at our fullest capacity. I was responsible for game ecosystem investment, including frontier gaming tech, young generation lifestyle, and new entertainment platforms. For AI-specific investments, we invested in vertical AI SaaS tools that enable different functions of a game pipeline, like facial animation generation for triple-A games. We also invested in AI virtual companionship, like Replika back in 2021.
GamesBeat: I remember Replika. Very interesting company. Wan: Yeah. We got reconnected last year. I left NetEase at the end of 2023 and did a small fund doing crypto investments for the past two years. Last year, I also thought it would be a great time to start re-evaluating how AI will be incorporated into the game and
entertainment experience. That was also the time Shaoyun was leaving NetEase. We
reconnected and shared a passion to look at new genres, new player behaviors, and help young entrepreneurs and producers refine their craftsmanship to grow into mature
creators. So that’s why we launched this fund together.
GamesBeat: Interesting. When you decided to make AI into one of the main things you would invest in, did you feel like you were early, late, or just in time?
Wan: That’s a great question. Shaoyun can understand English listening, but for speaking, I will help translate. Whenever you ask a question, we will discuss it and come up with an answer together. [To Shaoyun: For us, is it too early, too late, or just right?]
Shaoyun: For me, I think it’s just right. Previously, we started doing this internally at
NetEase around 2021. At that time, I felt we had hit a bottleneck. However, when DeepSeek came out, I felt the bottleneck was suddenly broken. That moment didn’t feel late; it felt like a previous limitation was overcome. When that happened, I felt I had to do this. I think it’s just in time. The technology has developed exactly to this point, giving me a huge sense of mission and passion to bring AI into the pipeline and let more producers use it. I feel I am the most suitable person to do this.
Wan: From his perspective, back in 2021, after Stable Diffusion and Midjourney came out, the art pipeline at NetEase was already aggressively pushing internal artists to use AI. At that time, we felt we had already pushed the limits of AI and there was nothing else we could do to further improve productivity. But for him personally—his college friends were actually the co-founders of DeepSeek—DeepSeek was a real awakening moment in the Chinese market. People felt, “This is something China can also do, and we can do it faster, cheaper, and more efficiently.” For him, it was an awakening moment to push for it, not just in investment, but to inspire existing producers and designers to leverage AI to empower themselves. It’s not just about cost-saving; it’s about amplifying personal taste and creative ambition.
For me personally, I would say we are definitely still early. Every premium in early-stage investment comes from non-consensus, meaning we have to be earlier than everybody else to get the right price, invest at the right stage, and get the right exit premium. We are early in terms of how AI can improve the player-facing experience. On the cost side, in order to be commercially validated, new AI-native games only make sense if the incremental player value they drive is significantly higher than the unit cost of that token. We are really looking forward to seeing that gap widen out.
GamesBeat: I understand there is a difference between the reception for AI in China versus the United States. We see a lot more resistance to it here. Gamers seem distrustful of new technologies and think there is a lot of “AI slop” out there. Game developers are suspicious of management that likes AI because they think it will eliminate their jobs, and they don’t see anything really useful yet that’s better than human game development. What do you think of that, and how do you think you can change perceptions in the West?
Shaoyun: I will divide this into two aspects. First, for Chinese players, the acceptance of AI is indeed much higher because they are highly result-oriented. As long as the game is good and the quality is high, they don’t care if it comes from AI production or not. That is the only thing Chinese players care about. So for Chinese companies, using AI achieves massive cost reduction and efficiency gains. Players won’t complain as long as the content is good.
For the West, we have some bold observations. First, why are bosses of large companies saying AI is bad? Fundamentally, AI is a tool that represents innovation and the disruption of small teams. Large studios tell everyone not to use AI because they are actually afraid of this innovation and disruption. Players might be led by this rhythm. I think players being led by the rhythm of these large company CEOs is actually not very constructive for the players themselves. Yes, there is a lot of slop right now, but true innovation or original gameplay often comes from this slop. There is AI slop, but there is also “friend slop” (human slop). For players, good is good, bad is bad. But if you are led by Western CEOs, it actually slows down the development of the whole industry. Wan: First, on the ideology level, it is true that Chinese players don’t have that ideology
rejection towards AI. It’s more about the result. If you give me a good game with a good
experience at a good cost, I don’t care whether you use AI or handcrafting. It’s much less of a concern.
He also thinks some CEOs or executives from bigger studios in the West are trying to push this divide further. For example, [one CEO] famously promised they will not use AI.
From an outside perspective, we have no way to validate if that’s true. But it’s a very good ideological weapon against AI, so players will tend to think only the products produced by these incumbents are worth buying. We think there are marketing incentives behind this ideological war to protect their competitive moat. To your second question on how to influence the Western market: honestly, on the developer adoption level, if they want to launch a game and start a business, everyone will use AI to save costs. It would be foolish to start small and not use AI. Ultimately, on the player side, the final question is what kind of quality product you can deliver. If we can deliver a super high-grade product with an interesting, new interactive experience, players will accept it and admire the craftsmanship.
GamesBeat: Are there some signs of progress that you see that show the fast, aggressive approach is the right one when it comes to AI tools or AI game development?
Shaoyun: I think it’s mainly in three steps. The first is what landed last year: AI emotional
companionship and chatting. The second, this year, is that coding capabilities have risen
significantly. However, returning to game design and 3D pipelines, it still hasn’t helped
much. The next step is what we are seeing now: small teams working on 3D pipelines,
multiplayer networking, etc. If AI capabilities can touch those boundaries, then those are
the things we can see accelerating right now.
Wan: We can speak from both the tech maturity standpoint and the consumer experience
perspective. From the tech perspective, back in 2021 and 2022, 2D image generation got
more mature—right now probably 90 to 95% mature. The second is code generation. The significant milestone was Opus 3.5 earlier this year (and now 3.7), which people think is approximating the capacity of a CTO in a mid-sized gaming company. Most studios we see are heavily using coding tools to accelerate. For 3D, startups like Meshy and Tripo are accelerating market penetration, but mostly they can only satisfy casual or hybrid-casual games; they are not there yet for AAA 3D assets. On the video generation side, tools like Kling and Vidu are heavily used in short dramas, and we are seeing studios use them to produce cutscene animations and trailers.
A fourth point is whether AI coding tools can help solve multiplayer games. Right now, most AI games are PvE or single-player. There are multiple layers of LiveOps problems that AI doesn’t have the capacity to solve yet. We hope the next generation of coding tools can help solve this bottleneck.
On the player experience side, the first commercially validated genre is for sure virtual
companionship, like Replika and Talkie. Since last year, we are seeing a cluster of new
platforms validating that AI can deliver player-specific, personalized experiences. We look forward to seeing multiplayer-based games and AAA-like quality games coming out once the 3D art asset tools are more mature.
GamesBeat:** I wonder how companies can figure out how to deal with token pricing. To get a game character to say something—like a blacksmith you can talk to all day—could run up a huge token bill if the player does nothing but talk to the AI. How do you solve it?**
Shaoyun: We roughly consider this in three layers. The first layer is the game structure and gameplay design. For example, using a caching engineering approach. It’s built on top of large models, but we use an engineering method so it doesn’t push inference every time. We use caches, structural generation, and only generate an experience for a certain period. The second layer is training our own open-source specialized models, like small behavior tree models running on the player’s own device. The third layer is genuinely using large inference models. We are very optimistic that it will become cheaper and cheaper. DeepSeek in China is extremely cheap right now. Future costs will just be like a utility bill, so token cost won’t be an issue.
Wan: Three layers manage that problem: First is on the engineering harness side and
how developers embed it into gameplay. You don’t want the player to just talk with the NPC; you embed the conversation into the general gameplay. You need a harness engineering system to ensure not every single instance goes to the LLM token. You use cached content and pre-written content. One studio, Aizivilization, we interviewed cut their cost by 80% this way, managing it to $2 per player per month. The second layer is training specialized models based on open source. You can train your own smaller model tailored to game settings and run it locally. The third layer is betting on scaling laws for LLMs, where costs will continue to decrease to a micro-utility level.
GamesBeat: Is there any more obvious revenue model that games can switch to so they don’t have this dependence on token cost?
Shaoyun: The business model is still traditional. I look at it from a core value perspective: we provide games and entertainment. The business models are still selling these things. However, AI provides a much better, stronger experience. A key shift in the future is moving towards customized services. Everyone will spend money on bespoke services tailored to them. So the business model will be more equalized and customized.
Wan: When we see the business model, especially for GenAI games, we look at human nature. You pay to be stronger, prettier, or more outstanding. Secondly, regarding the income structure of players, except for the top 1%, the gap between middle and working class might even out due to AI subsidies. The pay tiers will be less hierarchical. Everyone will have a chance for personalized content. You pay for a personalized service tailored to your own taste, rather than paying to show off social status.
GamesBeat: I agree that the end state of AGI is that we don’t have to work anymore and we can just play games. I wonder what those games would be like. You have an infinitely smart opponent.
Shaoyun: I still have faith that games are the final fortress. Games are for human
consumption, and AI has no consumption desire. Ultimately, even with a simulated world,
the interactive granularity will reach a level where you are experiencing a fully
personalized, human-centered story, interacting with AI NPCs in a highly advanced virtual reality.
Wan: Our founder believes games are the last fortress of human creativity. AI doesn’t have the human preference or taste that we do. We don’t foresee AI replacing the human design and experience completely.