MiniMax: a new flavour of consumer internet company with AI branding

Two Chinese AI companies went public within 24 hours. One popped 13% on Day 1. The other rocketed by 109%. The difference? One is built for enterprise contracts. The other built for attention.

I’ve been watching the Chinese AI IPO wave closely, and MiniMax’s debut last month still surprised me. Shares opened at HK$78.80 and closed at HK$168; doubling in a single session. This happened literally one day after Zhipu AI’s own IPO (which we wrote about here), which managed a respectable but comparatively modest 13% gain. They're both AI companies out of China but they had completely different outcomes.

Digging into MiniMax’s prospectus, I realized these companies are different not just in size or strategy. They’re also operating in parallel universes. MiniMax’s universe is one where AI monetization looks more like TikTok while Zhipu AI is more like Oracle, selling to boring enterprises.

The revenue engine isn’t “frontier AI,” it’s attention

MiniMax generated $53.4 million in the first nine months of 2025 (per their filing). It's not crazy revenue but the breakdown is interesting:

• Talkie, their AI roleplay app: ~35% of revenue

• Hailuo AI, their video generator: ~33% of revenue

• API business: ~29% of revenue

• Ads and other: The remainder.

  historically over 50% in 2024

What's interesting in the last year of operations (2024), Ads made up 50% of their revenue and Hailou did not even exist. Their revenue mix changed very quickly! Now compare that to Zhipu AI, which gets 85% of its revenue from enterprise on-premise deployments to banks and government agencies. Zhipu’s model is heavy B2B, while MiniMax is predominantly B2C - making just 30% from the API business (which is also not really enterprise).

And here’s the kicker: MiniMax’s API business actually makes money. We’re talking 64-69% gross margins. Their consumer side? Still bleeding cash as they buy users. But the API profits are real, which is more than Zhipu can say; their API business is reportedly underwater.

Talkie is big, but Hailuo is the plot twist

Talkie, MiniMax’s AI companion app, is doing serious numbers: 27.6 million monthly active users as of September, with 1.7 million actually paying for the privilege. The app briefly ranked #5 among free entertainment apps in the US last summer, ahead of Character.AI in downloads.

But the real story is Hailuo AI. This thing basically didn’t exist in 2024. By September 2025, it was generating $17.5 million; fully a third of MiniMax’s revenue. Users have created over 590 million videos on the platform. It’s got 5.65 million monthly actives and 310,000 paying users. This also drastically changed the revenue distribution in 2025, reducing dependence on ad revenue, which contributed about half of the revenue in 2024. (Total Revenue in 2024: 30M; in 2025: 53M in 9 months of 2025, projected 71M for 2025)

I keep thinking about CEO Yan Junjie’s quote:

“In the era of large models, the real product is actually the model itself.”

The apps are just delivery mechanisms. If the model is good enough, users find you.

Some Weird Financial Engineering

There are some interesting circular relationships in how MiniMax makes money. They’re paying Singapore-based ad-tech platforms millions to acquire users $11.4 million to “Supplier K” in the filing. But the same entities (called out as “Customer F” and “Customer G”) are paying MiniMax $9.4 million and $8.7 million respectively to run ads on MiniMax’s apps.

But it seems like standard practice in consumer apps - they’re buying traffic from these platforms and simultaneously selling ad inventory back to them.

The 385-Person Unicorn

MiniMax runs on 385 employees. That’s it. Zhipu has 700+. For a company with 200 million global users spanning chatbots, video generation, and enterprise APIs, that’s quite lean.

The average age of their R&D team is under 30. While they are spending heavily on R&D, 80% of this goes straight to renting cloud compute. They rent all their training compute with no data center capex or GPU hoarding. In fact, there is not a single mention of GPUs anywhere in their prospectus. This is really funny given that they are an AI company!

The Lawsuit That Didn't Stop the Party

In September, Disney, Universal, and Warner Bros. Discovery sued MiniMax for $75 million. They allege Hailuo AI generates “an endless supply of infringing images and videos” featuring Darth Vader, Minions, Superman i.e. the whole catalog. The studios claim MiniMax markets itself as a “Hollywood studio in your pocket” while refusing to implement copyright filters they already use for violence and nudity.

Statutory damages could hit $150,000 per infringed work. Do the math on hundreds or thousands of works, and you’re looking at exposure that could dwarf MiniMax’s $53 million annual revenue.

They IPO’d anyway. The stock doubled. Either investors are betting this settles for pennies on the dollar, or they think the growth trajectory is so steep that legal risk is just background noise. Honestly? It might be the best marketing Hailuo ever got.

What This Means for AI Monetization

I’ve been thinking about the key takeaway here - there seem to be multiple ways to be an AI company. MiniMax proves you can build a defensible business by:

1. Going consumer first - Viral apps give you distribution speed no enterprise sales cycle can match

2. Letting the model be the product - Technical excellence creates organic growth, not just feature checkboxes

So while Zhipu AI’s enterprise-heavy approach is not wrong, it's definitely slower. In a market where model capabilities are leapfrogging every six months, speed might be the only moat that matters.

My Take

MiniMax is betting that attention compounds faster than contracts; that viral consumer adoption creates data flywheels no enterprise sales team can replicate. Going global-first (73% revenue overseas) also insulates them from China's domestic pricing wars, where DeepSeek undercuts everyone at $0.14 per million tokens.

The 109% IPO pop versus Zhipu's 13% tells you what the market thinks. Whether they're right depends on two things: can Hailuo keep its viral momentum without becoming a legal cautionary tale, and can Talkie retain users once the novelty fades?

What MiniMax proved is that in this AI gold rush, capturing attention might be worth more than capturing contracts. That's going to rattle some boardrooms.

What do you think? Is MiniMax's consumer-first model sustainable, or is Zhipu's enterprise focus the safer long-term bet? Drop a comment or share this with someone tracking the AI IPO wave.

Sources: Reuters, CNBC, SCMP, company prospectus filings, and various tech publications covering the IPO and lawsuit.