Two reports landed ten days apart. They measured the same technology. They reached opposite answers.
One says AI has taken over. One says almost nobody is paying for it.
Both are right. The gap between them is where a seller makes a living.
What CGTrader counted
CGTrader's 3D Model Market Trends 2026 came out on 23 July. The data runs June 2025 to May 2026.
AI-made models are roughly 1 in 6 uploads on that marketplace. They account for about 1 in 90 dollars of revenue.
The rest of the report explains the pressure behind that. Supply grew 18.5% in the year. Demand grew 5.1%, so average prices fell 9.7%.
More shelf, the same number of shoppers. That is the whole story of the price drop.
What the Steam census counted
Sulka Haro published three years of AI on Steam on 13 July. He checked every Games release from mid-2023 to mid-2026.
That is 53,597 titles. Not a sample, a census. For each one he recorded whether the store page carries Valve's AI disclosure.
AI-flagged launches went from about 13 a month to about 530. Non-AI launches barely moved, from around 1,030 to around 1,320.
Depending on the window you pick, 60 to 90 per cent of Steam's release growth is AI-flagged games. That is the takeover headline, and it is true.
The gap is the whole point
Uploading is cheap now. Selling is not. That is the finding in both reports.
Haro's cleanest number is this one. An AI-flagged game reaches 100 reviews at about 55% of the non-AI rate. He notes that ratio has not improved in two years.
So the rising sales share is volume, not quality. More shots, same miss rate.
Print the caveats or drop the numbers
Haro estimates revenue with the Boxleiter method. He says plainly that it is only order-of-magnitude accurate for any single game.
Keep 1 in 6 as uploads. It is not the share of live listings, and it is not what a buyer sees on a search page.
CGTrader is one marketplace with its own survey attached. It is not the whole market, and it has an interest in the answer.
Where I half-agree with the CEO
CGTrader's chief executive told 404 Media that AI "can accelerate the more mechanical parts of the process".
The first half of that is right. Generation is genuinely fast at getting a shape on screen.
The second half is where I disagree. For somebody selling assets, the mechanical part is the product.
Buyers pay for topology, UV layout, clean bakes and sane object names. Creative direction is the cheap half of the job. The mechanical half is the invoice.
My conflict, in the open
I sell the AI 3D modeling and game-ready workflow course. I also sell 358 products, and not one of them was generated.
That is a conflict, so I am stating it before I ask you to trust the rest. I teach the workflow because the cleanup is the job.
What AI retopology does to your bake is the same argument, made with wireframes instead of revenue.
What the money actually rewards
Two numbers from the CGTrader report point the same way.
- Top-rated models held 65.4% of all CG revenue, through the entire price correction.
- PBR models are about 40% more likely to ever sell than non-PBR ones.
Blender is the most-bought format on the site, and it grew at twice the market rate.
Haro's tail is worse than most people expect. The top 1% of Steam titles take about 94% of estimated revenue. The median game with any reviews at all makes roughly 300 dollars.
The disclosure text gives it away
Haro read the disclosure wording on 138 successful AI games and 400 flops.
Among the flops, 72% mention AI art, images or textures. Among the successes it is 57%.
The successes were three times more likely to disclose AI voice work, and used it for text and localisation too. The median flop is an art-asset play and nothing else.
The boring alternative
My 692 base mesh bundle is 26 volumes and 692 models for 98 dollars.
Nobody generated it. It sells because it is a reference library you keep, not because it is new this week.
The base mesh category has 40 packs in it now. That took years, and the arithmetic above is why I am not in a hurry to change the method.
What would change my mind
One number would end this argument. If AI-flagged Steam games reached 100 reviews at 90% of the non-AI rate, I would rewrite the piece.
Same on the marketplace side. Move 1 in 90 dollars to 1 in 10 and the position is dead.
Until then, upload share is a headline. Revenue share is the market. Do not confuse the two when you decide what to spend your week on.
