Last week, 720 new games launched on Steam. Of those, 530 still had fewer than 10 reviews as of today.
That figure comes from a census of 53,597 Steam games, covering the period from July 2023 to this July. The person who conducted it is Sulka Haro, who was the lead designer of Habbo Hotel in the early days.
The most striking finding in the census is this comparison. New games labeled as using AI made up 10.9% of all new releases the year before last, 19.9% last year, and 30.8% in the first half of this year. One-third.
But their share of revenue was just 10% of the platform’s total in the second quarter of this year, down from 17% in the first quarter. Supply is surging while revenue is falling, and the gap widened by another third in a single quarter.
AI has brought the cost of “making a game” close to zero. It hasn’t lowered the cost of “making people want to play a game” by a penny.
Steam has always been a battle royale. The top 1% of games by revenue take 94% of the platform’s estimated earnings. The other 99% split the remaining 6%. That pattern was set long before AI came along, and has nothing to do with what tools you use to make games. AI hasn’t changed it; it’s just made the denominator bigger, and growing faster.
The AI games that survive have something in common. Among the failures, 72% used AI for art—that is, the first thing players see. The survivors used AI in places players don’t keep their eyes on. AI voice acting was used by 24% of them, compared with just 8% of the failures. Localization: 18% versus 6%. The same tools: one group used them to replace craftsmanship, the other to amplify it.
I’ve seen this story before.
That’s how things played out with blockchain games a few years ago. Launch a token, and the barrier to making a game suddenly drops. The number of projects explodes. White papers pile up. Then everyone runs into the same wall. It’s not that players have nothing to play; they just don’t want to play your game. GALA and SAND fell from their highs back then—not because of the technology, but because of a lack of demand.
My view is that AI games will repeat this cycle over the next two years, but more slowly. Blockchain games lowered the barrier on the funding side: the money came first, and the product could come later. AI lowers the barrier on the production side, but you can’t fool people with what you make. Steam’s review section is the filter.
I’m not sure how reliable the AI labels in this census are. The labels are self-reported by publishers. Checking the box might hurt sales. Not checking it is hard to detect. Only the development team knows how much AI went into an art outsourcing job.
One thing is certain. In this wave of change, the valuable skill isn’t generating things; it’s filtering what’s generated and refining it until it’s playable. Those blockchain projects never developed that skill. People making AI games now are trying to build it.
$GALA $SAND $BTC #币安广场 #AI游戏 #Steam #链游 #CryptoAssets