Game Companies Investing in AI: Look Beyond Stock Fluctuations

I spent the weekend tinkering with an "AI Hard Tech Investment Observation Table" and stepped on quite a few pitfalls. According to the original Huxiu article, from the beginning of the year to September, gaming companies completed nearly twenty investments in the hard tech track. Perfect World subscribed to a frontier tech fund, investing in semiconductor equipment and new materials; Yoozoo Network also partnered to land projects in Wuxi. Ordinary readers tend to only remember "doubled," but judgment requires looking at gains/losses, lock-up expirations, revenue, investors, and tracks.

This tutorial is for beginners; no data terminal is needed, Excel is enough. I've used Excel for about three weeks, and the experience is that the table doesn't need to be complex. First, ensure you can understand why it rose, why it fell, and why others dare to invest.

Build a Table First

Open Excel, click File, then New, and select Blank Workbook. In the first row, enter from A to H: Company, Event, Date, Stock Impact, Lock-up Pressure, Revenue Source, Investor, Hard Tech Direction. After entering the first row, it will display as plain text; don't worry. From the second row onwards, fill in one company per row.

In A2 write MiniMax, B2 write First post-IPO lock-up expiration, C2 write July 9. Lock-up expiration means stocks previously locked and unable to be sold freely have matured; the market worries someone will sell. Public reports say this involved 153 million shares, close to half the total share capital; another report says the unlock ratio accounts for 63% of total share capital. In D2 write Intraday drop breached 20%, closed down nearly 18%; E2 write 153 million shares, conflicting ratio metrics. Two numbers fighting usually means the market conflated "floatable quantity" with "percentage of total share capital." The solution is to note the source metric, not just write a single percentage.

In F2 write Mainly C-end apps, Token billing switch caused controversy. Token can be understood as paying by usage volume, whereas subscription is monthly/yearly packages. According to TMTPost, the new generation model M3 performance didn't meet market expectations, and the billing mode switched from subscription to Token consumption, effectively raising prices. The key for this track is whether users are willing to pay for each call.

In G2 write Alibaba ~13%, miHoYo ~5.24%. Public reports indicate Alibaba is the largest single external shareholder, a cornerstone investor, and additionally subscribed to new shares; miHoYo's lock-up period lasts until January 2027, and they publicly stated they wouldn't reduce holdings. Don't just write "has Alibaba." Add a column Identity to separate financial, strategic, and industrial roles.

In H2 write Model company, not semiconductor equipment. Gaming companies investing in AI doesn't mean they're investing in chips. Semiconductor equipment, new materials, data centers, and model applications are different tracks. Mixing them up distorts the table.

Then fill in two gaming companies. In A3 write Perfect World, B3 write Subscribed to frontier tech fund, H3 write Semiconductor equipment and new materials, F3 write Fund contribution, not directly consolidated revenue. In A4 write Yoozoo Network, B4 write Wuxi partnership project landing, H4 write Hard tech track. If specific numbers aren't seen, write To be supplemented. The most common beginner mistake is leaving cells blank. Blanks are okay, but mark them To be supplemented to remind yourself not to treat "don't know" as "doesn't exist."

Pros/Cons and Judgment

After trying it out, this table breaks down the hype into fields. Seeing "MiniMax went up a hundredfold," you won't just follow emotions; you'll check lock-ups, revenue, and investors. It's also suitable for beginners to practice. Recently, I used LSEG to pull capex metrics; professional data is useful but has a high barrier. Using Excel for this observation table establishes a framework in three rows. It also exposes contradictions. For example, lock-up ratios: one side says close to half, the other says 63%. You don't need to immediately judge who is right, just know the market is worried about selling pressure.

There are plenty of pitfalls. Dates get messy easily; beginners like writing just "July," which messes up YoY calculations later. YoY means comparing to the same time last year; unify format as YYYY-MM-DD. Valuation and market cap are also easily mixed up. Stock price fluctuation is secondary market; financing valuation is primary market. Best to split into two columns. Some also misunderstand gaming companies investing in AI as "gaming companies transforming into AI." From the data, this looks more like switching capital entry points: gaming companies bet profits and financing channels on hard tech, while AI companies reciprocally gain traffic, content, and tools.

This table doesn't predict rises/falls; it only looks at who is selling, who is buying, and where money is invested. The MiniMax controversy this round, superficially about lock-ups and price hikes, fundamentally stems from thin revenue structures for model companies, with a gap between C-end popularity and monetization.

Zhipu is compared because public reports show its 2025 total revenue was 724 million RMB, with B-end contributing 99%. Enterprise general large models accounted for 366 million RMB, open platform & API for 190 million RMB, enterprise agents for 166 million RMB, and MaaS API ARR reached 1.7 billion RMB. B-end contracts are harder to sustain than C-end popularity but look more like cash flow.

You can use the same table to fill in OpenAI, Anthropic, Zhipu, and MiniMax, focusing only on stock price/valuation, lock-ups or financing, revenue structure, major investors, and technical routes. After filling in four companies, you'll see what each relies on to make money.