Inside Tipsy: The $10 Million Cyber Host Club

"The Spring River warms, the duck is first to know"

Many people probably haven't heard of Tipsy yet, so let me give a quick introduction: an AI companion product that may have already pushed past the boundaries of "edgy," with revenue and profit growth rivaling Melvin Chen.

SensorTower shows Tipsy's August MRR at $17 million.

But on the other hand, Tipsy's DAU is still under 500,000.

Rough estimates put Tipsy's ARPPU (average revenue per paying user) between $90 and $100.

This is a product I've never seen before in the VC world. I've learned that their external valuation is already in the billions — practically the next Melvin Chen rising.

I reached out to my friend Ming, who previously led overseas growth at a star startup and was among the earliest practitioners in China to take AI emotional companion products global.

At the same time, Ming is a veteran otome game player with a dedicated VIP customer service rep at Paper Games, having spent enough on 2D men to fund one of their office bathrooms — making her perfectly suited to analyze Tipsy.

Here's what Ming wrote:

I first noticed Tipsy a few months ago during an app store winter, when former hard-tech investors suddenly started asking me about an overseas interactive content platform with under 100,000 DAU but $80–100 million ARR.

With Loopit and Rezona showing retention numbers too embarrassing to look at, and having witnessed CrushOn push MRR to a rumored $5 million at a scale where the founder could be arrested any day, my instincts told me this was probably another "cyber brothel."

But what interested me was what could possibly drive such high ARPPU, since products at similar revenue levels typically have DAUs several times Tipsy's.

So I dug into the product positioning and user profiles, and discovered this was a cyber host club.

Turns out it's not just Chinese women who are starved for love — women worldwide are starved for love — I mean, people everywhere need their White Horse Club.

Previously, Xianyu wrote about large models being a brutal manufacturing industry, and China's primary market will eventually return to the AI applications version.

And as the representative of this wave of AI applications rising from the dead, Tipsy has captured the entire market's attention, bearing the heavy responsibility of restoring the AI application dynasty.

The spring river warms, the duck is first to know — Tipsy's cyber ducks have brought spring to AI applications. If this really revives AI applications, everyone in the industry should kneel before Tipsy.

After all, where Melvin Chen goes, nothing grows — but Tipsy can actually heat up the entire赛道.

01

Before analyzing Tipsy, some context on AI companion products

The previous generation of emotional companion products (C.ai, CrushOn, etc.) primarily monetized through male-oriented sexual content.

This was the GPT-3.5 era when C.ai rose — AI couldn't even remember usernames or character relationships, tokens were prohibitively expensive, and domestic models could barely break even overseas.

AI emotional companion practitioners gradually discovered that rather than engineering complex prompts and memory systems, it was faster money to loosen NSFW restrictions on models and squeeze revenue through more characters for users to sample.

So these AI companion products became increasingly explicit, with character cover images showing increasingly less fabric.

Simply put, companion products of this era, despite wearing the label of AI social, were actually conversational content consumption products.

But thanks to longer context memory, the emergence of agents, and even improvements in multimodal capabilities, more complex emotional companionship gradually became possible.

This "complex companionship" specifically refers to gamified, scripted experiences.

Compared to simple conversation, complex content also raises the ceiling for relationship depth with virtual characters (here referring to roleplay relationships, not social relationships).

Thus products like Zeta overseas have seen revenue soar.

Rather than understanding them through the previous generation's "conversational content consumption," it's more accurate to say that as model capabilities have risen, they've absorbed some fanfiction and novel enthusiasts through roleplay and tabletop RPG formats into existing product forms.

This isn't to deny that NSFW content remains the primary revenue driver — just that compared to pure chat-based flirting, post-apocalyptic encounters and Chicago mafia shootouts leading to explicit scenes are obviously better experiences.

Screenshots can't be included here for content reasons; interested readers can check Tipsy's community.

02

Tipsy, inheriting the legacy of Catbox

With that context, let's get specific about how Tipsy differs.

Compared to AI companion 1.0 products, Tipsy still uses a chat interface, but the consumption experience increasingly resembles fanfiction and novels with multimodal features:

Characters have backstories, relationships have tension, and stories unfold based on user responses. Users are both reader and protagonist, able to rewrite the narrative direction at any time.

The other side competes for you through immersive storytelling — being favored, being fought over, experiencing ambiguity, conflict, and reversals, then deciding what happens next.

These attractions have long existed in web novels, fanfiction, and romance games, but AI lets stories unfold around each person's preferences.

Even using the same character and opening, different users can enter different plots.

This also provides more direct reasons to pay. With simple chatting, you can exit when satisfied. But when a story stops mid-plot, users become invested in what comes next.

How will they respond, will the relationship advance, how will that foreshadowing resolve. When this "need to continue" impulse combines with attachment to the character, sustained consumption gains momentum.

For example, while experiencing the product, I loved a scenario I created about being reborn as a green tea bitch:

I'm a "ordinary girl" beautiful without knowing it, who moves into a hot influencer's place as a roommate, accidentally passes behind him during his livestream, and my stunning beauty crashes his entire stream.

After "accidentally" overhearing his official girlfriend going crazy about this, I quietly ask, "Is your girlfriend okay?"

The hot guy is now completely shattered, thinking I'm cute, kind, beautiful, and moving, while finding his girlfriend too unreasonable.

One line撬出 so much plot — truly gave this sis the performance of a lifetime (being a green tea bitch hits different).

Here's a supplement from Qiu Mu:

Actually, the earliest product to pioneer story mode was ByteDance's Catbox, their domestic answer to STARFIELD — unfortunately Doubao's model was too terrible to achieve good results.

So it was Doubao that killed Catbox. Liang Chenqi really got screwed here 😭

But now he's broken free from ByteDance's constraints and struck gold — let's look forward to his triumphant return 💪

03

The Tipsy variable: female-oriented, deeper content consumption, higher ARPU

Beyond product form, I think Tipsy made the right call choosing female-oriented content.

Tipsy originally had gender-neutral content plus 2D IP characters — you could still see big-breasted female character assets a year ago:

But by the second half of 2025 it gradually pivoted to female-oriented companion products, spending higher acquisition costs for more expensive users, and catering to them extremely aggressively in product experience and operations.

The audience gradually shifted from men in tier-2/3 countries to adult women in Europe and America — truly impressive, turning from浪子回头 to going all-in on whale users 👍

SensorTower shows Tipsy's 30-day retention for paying users exceeds 50%. Market rumors suggest 90% of paying users are female.

Clearly, compared to Emochi with its much higher DAU, Tipsy leverages female users writing their own fanfiction with cyber boyfriends, with better content experience, to drive significantly higher ARPU than peers.

In emotional companion scenarios, male-oriented and female-oriented content differ significantly in modality. Compared to direct physical exposure, female-oriented content typically requires space for imagination.

Thus pure visual pornography is less effective than erotic literature, suggestively tense images, and alluring voices (for example, Maoer/Manbo radio drama audiences are also primarily female) for women.

A quick aside — so are there no AI products male users like?

Yes, brother, yes.

The famous CrushOn achieved millions in monthly revenue through sheer audacity, becoming AI applications' first small win, though its later growth was mediocre.

Also looking forward to the renowned Huang Guo (censored) carving out new territory through post-training H3.

Back to the point — female-oriented content requires more foreplay and world knowledge.

Before getting explicit, women don't just need to know if you love me, but what you love about me, how you'll love me, why you love me.

And it can't just be verbal responses — you need to describe your actions, what we've experienced together, what stories and environments we shared, what you sacrificed for me.

Answering such philosophical questions naturally requires more advanced, more intelligent models.

This also explains why, despite similar product forms, Tipsy targets whale users while Emochi relies more on IAA+IAP (ads + in-app purchases).

The former currently uses primarily Claude Sonnet + Gemini, while the latter uses cheaper self-trained models through post-training.

To some extent, Tipsy resembles a currently hot, upgraded-scale version of romance companion games — you know some women are willing to spend five figures on those.

Perhaps to配合融资 storytelling, starting from late June this year, Tipsy's growth team began aggressively spending on acquisition, showing a significant growth curve.

Per SensorTower data, Tipsy's DAU grew from under 100,000 in June to nearly 500,000.

However, estimates suggest aggressive monetization methods likely make Tipsy's overall retention rather感人, and it's uncertain whether the current scale can still be profitable.

From growth creatives, both paid acquisition and KOL/KOC campaigns target the wild, liberated side of European and American fangirls. They primarily cover mid-tier creators, concentrated in 2D, anime, and fanfiction communities.

Love and Deepspace remains the版本答案 from a creative perspective — first-person perspective still delivers strong immersion on social media.

But from a growth practitioner's view, there's nothing innovative anymore. Growth for this category has been templated since the STARFIELD/Talkie era, and some viral KOL content scripts already appeared in 2024 campaigns.

In short, while Tipsy is still small potatoes compared to otome games in whale user scale and spending depth, it has indeed cracked open a high-ARPU paid gaming market in the AI female-oriented赛道 that many straight male founders (naming EVE and other AI otome companies here) have wanted to capture.

So does Tipsy's success come from some special trick? Let's assume its data is fully real.

My assessment: no, it's just basic operations. Only the contextual memory bottleneck that previously constrained female-oriented content has been solved through model advancement.

This is what the AI industry should have been doing all along — continuously extracting gold from high-quality paying overseas users.

You know Tipsy's founder was previously in crypto.

What can I say, except that you AI folks mock crypto daily, yet when crypto people actually enter the space, they discover how incompetent you AI industry people really are.

04

But still, it's just a business

After all that praise, this doesn't mean I buy Tipsy's valuation. However high their revenue, the current product form is still just a cash flow business.

Before the multimodal era arrives, Tipsy's current form is essentially a personalized novel platform + Orange Light Games + AO3 business.

The content richness填充 challenges novel platforms face, and the user migration between platforms, remain unresolved;

Orange Light Games' generalization issues are well-known, and AO3 remains to this day a subcultural,公益属性 community.

Though its revenue structure more closely resembles whale-driven models, where a small portion of top spenders contribute the vast majority of revenue, the spending pool's depth and breadth currently can't compare to otome games.

Not to mention the compliance challenges Tipsy will face as a content platform scales up.

To repeat: this is just a business.

Qiu Mu also asked if Tipsy could become another Newborn Town or Soul? He meant the whale business like voice rooms.

Voice rooms sell relationship, presence, and status — all of which require a public/semi-public venue to function.

Users pay for priority position in a relationship and that relationship's hierarchy, and the spending must be visible to others to matter.

This leverages social products' network effects — the more people spending around the same relationship, the fiercer the competition, the more each person is willing to spend. Payment is driven upward by others' payments.

But Tipsy's payment is one-directional, a one-to-one transaction between user and product. User growth doesn't directly increase whale ARPU.

Without this multi-player competitive venue, you can't grow the continuously climbing spending curve of whale users.

Current user stickiness likely still depends on specific content on the platform — similar content appearing on other products, with equivalent experience and cheaper tokens, and they may migrate.

While writing this article, C.ai sold to The Walt Disney Company for $20 million — the初代版本神就这样歇菜了. Though Tipsy benefited, partially absorbing these users.

I also once wondered why products validated before the pre-Manus era could still hook people years later.

After looking into it, I was relieved — turns out it was investors who hadn't previously looked at AI companion, or even AI applications, now coming to look at AI.

Many investors from the previous AI companion wave are no longer in this赛道, or even this industry.

So, folks, when your pivot speed is slower than investors getting eliminated from the industry, you can survive to strike gold one day.

(This article's cover image was generated by ChatGPT; purely human-written)

Stay tuned — tonight at 8pm we'll be conducting an AI phone review, examining whether the 30-day-old Doubao phone and the newly released StepFun phone qualify as AI phones. Come watch.

⬇️

Subscribe to our Substack funeralai.substack.com