Astrocade hits 5M MAU and 75K user-created games — building the TikTok for AI-generated playable content

Astrocade

Amir Sadeghian's pitch for Astrocade is simple: what Twitter did for text, Instagram did for photos, and TikTok did for video, Astrocade wants to do for games. Type a prompt, publish to a feed, reach millions of players. No coding required.

The platform launched publicly around seven to eight months ago and has reached 5 million MAU with 75,000 games created by hundreds of thousands of creators. The games are HTML-based, entirely web-delivered, and currently free to play and free to create. Sadeghian describes the content as "ultra-casual" — sessions measured in minutes, not hours, with mechanics closer to a mobile time-killer than a console title.

What's actually being created

The surprise is that the most viral content isn't reskins of existing games. One breakout genre has users washing things — Mona Lisa's painting, a dirty car — a loop Sadeghian compares to the power-washing and rug-cleaning videos that dominate Instagram. Simple, satisfying, and nobody predicted it. Sadeghian argues this is the nature of mass creation tools: the power law generates a few massive hits, but the long tail surfaces game mechanics no professional studio would have greenlighted.

About 12% of creators get outsized play counts; the rest still get thousands of plays per game. Sadeghian says a game his AI can generate for $10–20 would have cost a studio a million dollars and a month of development time.

“We launched the platform publicly around seven or eight months ago. Our MAU is like 5,000,000 right now. Hundreds of thousands of creators are now creating games. We actually processed trillions of tokens to make this happen — it's completely free. The biggest advantage of Astrocade is the core of the experience: you play things rather than consume them. Engagement is just so much better.”

The hard technical problems

The creation tool is the obvious challenge, but Sadeghian says the recommendation system has been harder. Video platforms can measure completion rates against a fixed runtime. Games are open-ended — finishing a game in ten seconds might signal satisfaction, not abandonment — and extracting meaningful engagement signals from that requires purpose-built infrastructure. No off-the-shelf recommendation engine handles this content type at scale, according to Sadeghian.

At 75,000 games and $10–20 per game in inference costs, the rough token bill implied by the conversation is around $1.5 million in compute to date. Sadeghian says Astrocade routes simpler tasks to cheaper models and more complex generation to frontier models, with the expectation that inference costs will keep falling.

Monetization and the Apple question

The platform has not started monetizing. The plan is an ad layer for brands and creator-facing microtransaction tools, modeled loosely on YouTube's creator monetization stack rather than a subscription. Sadeghian isn't worried about the revenue model — gaming is structurally profitable and larger than Hollywood by revenue — but the timing is open.

The Apple question is more uncertain. Astrocade is fully web-based today and hasn't confronted the App Store rules that would apply if it moves onto iOS as a native app-creation platform. Sadeghian's read is that the distribution Astrocade brings should be attractive to Apple, though the tension is real: Apple currently takes a cut of in-app purchases on games, and Astrocade's ad-supported model routes around that entirely.

The structural bet

Sadeghian frames the generational opportunity around a consistent pattern: every time a creation tool became accessible to non-experts, a platform-scale company emerged. The counterargument — that passive consumption still dwarfs active gaming by total hours — is one he largely accepts, which is why he avoids the word "games" internally and calls the content "playable content." The audience is not replacing Netflix; it's occupying the gaps around it.