Behind Steam Next Fest’s Record 8,202 Demos: AI Supply Glut, Player Backlash, and Demo Self-Preservation Under Clone Panic#

By Wubian (NTGame Editorial · AI + Gaming Watch Column)

Hello everyone, I’m Wubian.

On October 19, this edition of Steam Next Fest opened. The official news page puts the count at 8,202 playable demos, more than double the same period last year — a genuine all-time record. By all rights, this should be the liveliest week of the year for indie developers: the shelves have never been this full, and players have never had this much to choose from.

But the headlines of this Next Fest are likely to be stolen by two pieces of news that run opposite to “lively.” The first comes from players: a post petitioning the platform for a “No AI” filter struck a broad chord in the community, on the grounds that works containing AI content have flooded in and browsing has become genuinely exhausting. The second comes from developers, and it cuts deeper: Digital Sun, the studio behind Moonlighter, postponed its public playtest outright out of fear that its new game would be reverse-cloned by AI before launch; another studio simply cancelled a demo it had already scheduled for Next Fest.

Put these two stories side by side and you’ll see they are two faces of the same thing: after AI pushed the cost of generating game content down to nearly zero, the supply side produced two diametrically opposite reactions — “flooding the shelves” and “self-preservation.” On one side, developers use AI to churn out works quickly and stuff them onto shelves until players start petitioning to “filter these things out”; on the other, developers, afraid of being cloned by AI, no longer dare to hand over playable demos. This article takes those two faces apart.

First, the sourcing. I have previously written a workflow-transformation look at Epic x Disney delaying a game and going all-in on “AI-first” development, an analysis of Google Playground teaming up with Unity on an AI game-creation platform, and an ecosystem look at Microsoft building a decompilation team and AI “reviving” shut-down games — in those pieces, AI was “a production tool” or “a preservation means.” This piece does not repeat those threads; it focuses on AI content oversupply, backlash, and community governance: how the AI pile-up crushes the shelves, how players push back, how developers contract in reverse, and how “the demo,” the indie game’s lifeblood for customer acquisition, is being rewritten. All specific facts in this piece (numbers, dates, quotes, company names) come from public reporting by multiple overseas and domestic outlets retrieved on October 20, 2026; single-source or secondhand claims are flagged at each point; anything officially unpublished or unverified is stated as such, without speculation.

1. First, the Facts: The Biggest Next Fest Ever, and Two Opposite Bills in the Same Week#

Let’s lay out the hard information about this Next Fest. The event window, the demo count, and the caveats around how that count is measured are all in the table below:

What does 8,202 mean? Last year’s figure couldn’t catch up even if you cut it in half. The demo count itself fluctuates dynamically — one observer watched the live search results page and saw the number drift across a range of several thousand — so arguing over the exact digit matters little; what’s worth recording is the order of magnitude: before counting a single released game, the demo shelf alone has already grown to more than twice last year’s size.

And next to that record number hangs another bill: how many of these demos used AI-generated content? There are currently two sets of figures for that question; here they are as reported:

The two metrics cannot be mixed. The 17% is a “disclosure” figure — the platform requires developers to proactively declare player-facing AI-generated content (art, audio, story, localization), and only declared items count; the industry widely believes the actual share is higher than this. If the 40% figure is accepted, it implies a rather glaring inversion: works containing AI content take up no small share of the shelf, yet account for a clearly lower share of sales. But it must be stressed again: the latter is a Chinese paraphrase, the original unseen, and I treat it as an unverified citation without drawing conclusions. Whichever metric you adopt, though, both point to the same thing: AI content’s share of supply is rising significantly.

2. Shelf Overload: The Players’ “No AI” Filter Petition, and the Technical Wall It Runs Into#

2.1 Three Voices From the Petition#

The mood on the player side is captured by three community posts. These quotes were relayed by multiple overseas and domestic outlets on the day the Fest opened:

• One user wrote: “I wish I could filter out all games containing AI-generated content.” This is the petition’s core demand — blunt, specific, actionable.
• Another user’s comment reads more like a sigh: “It’s all cheap AI stuff everywhere now; I don’t even want to browse anymore.” Note the second half — not “don’t want to buy,” but “don’t want to browse.” When discovery cost is pushed so high by junk content that people give up browsing, that is a more substantive injury to a platform whose selling point is a “massive shelf” than bad reviews.
• A third user’s thinking carries a hint of fighting fire with fire: he hopes the flood of vibe-coded games will drag down engagement metrics and force the platform to impose stricter submission rules — let the data speak, and let the shelf rot visibly enough that the platform has to act.

All three posts point to the same demand: put a “No AI” filter on the shelf. The petition drew cross-outlet coverage on the very day the Fest opened; the sentiment is not fringe.

2.2 Technically Easy — What’s Missing Is the Toolchain’s Last Mile#

Interestingly, technically speaking, this filter is almost a “free” feature. Breaking it down:

In other words, nearly every building block the filter needs already exists: the disclosure system, the tag system, the developer-reported data. The only thing missing is the final step — making “AI” an official tag and giving players a checkbox. And that May change cutting user-defined tags happened to narrow the community’s path to building its own filters as well. The governance tools exist; they just aren’t connected.

Pulling the governance timeline out separately makes the irony of “not connected” more concrete:

From May to October is precisely the window in which AI-containing supply piled up fast: more and more AI works on the shelves on one side, a narrowing path for community-built screening tools on the other. With pressure squeezing from both supply and demand sides, it is no surprise the petition erupted on opening day.

As of writing, the platform has made no public response to the “No AI” filter request; overseas outlets reported that several similar posts exist in the community, but no complete list was given. So whether this petition eventually becomes a feature, there is currently no official position to cite — this article records only the request itself and its technical feasibility, and makes no predictions.

3. The Backlash: Developers Begin Reclassifying Demos From Acquisition Channel to Security Exposure#

If the players’ petition was merely “unpleasant to use,” what is happening on the developer side is far more serious — they have begun re-evaluating the demo from an acquisition channel into a security exposure.

3.1 Digital Sun: Postponing the ReVamp Public Playtest — “Too Big a Risk”#

Start with the heaviest hammer. Digital Sun, the studio behind Moonlighter and Cataclismo, postponed the public Steam demo of its new game ReVamp. Their words on the Steam blog, as quoted by multiple outlets: “We are worried by recent news of games being reverse-engineered, cloned and released before the original even launches… With months to go before release, releasing a large public build is too big a risk.”

What kind of project is ReVamp? It is pitched as a tower-defense x metroidvania hybrid (“Tower Defensevania”): players play Dracula, running a modular castle that can be rearranged to fend off invaders. That “the castle layout is itself the strategic core” design is precisely the most visible source of clone risk — rules, numbers, and module-combination logic are all sitting in the playable build; read them and you can copy them. In the delay statement, the team also made a public appeal to the platform: we hope protection for original works keeps strengthening.

The new schedule looks like this:

Note what that ordering implies: first move to a small invite-only Discord wave test, pushing the public demo to two months before launch. Waves, invitations, small samples — all essentially caps on “how much content gets exposed early”: enough to gather feedback, not enough to feed a cloner’s full reverse-engineering. One point must be nailed down here: as of writing, there is no public evidence that ReVamp has actually been cloned; all reporting points to a preventive judgment about risk, not a realized loss. Digital Sun is buying insurance against a threat that has not yet occurred.

Looking more closely at ReVamp’s gameplay skeleton makes the worry easier to understand: its strategic core is playing Dracula, running a rearrangeable modular castle to repel invaders. Value in such designs is highly concentrated in “system rules and module-combination logic,” not in any single art asset — the rules and numbers are all written into the playable build, and the deeper you play, the more complete the reverse-engineering. In other words, the more generous the demo, the more feasible it becomes to lift the core gameplay whole. For works whose main selling point is “system design,” releasing a public demo early amounts to laying out the most valuable part first.

3.2 Alliance Arts: The Demo Cancelled Outright#

Another studio’s move was even more decisive. Developer Alliance Arts directly cancelled the Next Fest demo for Drunken Goddess Reflux, an anime-styled Russian-roulette horror game. Their statement voiced concern that demo content could be “quickly imitated or copied using AI or other tools and released before the game’s launch.”

Compare the two studios’ situations: Digital Sun only postponed, leaving a January 2027 target for the public playtest; Alliance Arts pulled a demo already scheduled into Next Fest — I hardly need to explain how valuable the Next Fest exposure window is for an unknown new title, and they still judged it not worth the clone risk. The two studios chose differently, but the judgment is the same: right now, the downside risk of “releasing a playable build early” has gone up.

Setting the two moves side by side makes the differences clearer:

One postponed, one cancelled, with different balances between what was kept and what was given up — but the same key phrase appears in both statements: imitated or copied, and the window before launch. That is no coincidence: it shows that “AI cloning” is no longer an abstract worry among developers but is being assessed as a concrete threat with a working method.

3.3 The Clone Wave Is Not a Hypothetical Enemy: Needle in a Haystack and the Cloner’s Cloners#

Is there a real-world sample of this risk? Yes — and the absurdity is very specific. The indie game Needle in a Haystack is being buried under a flood of fast-made knockoffs. The knockoff wave even evolved into a matryoshka of dark comedy: cloners accusing others of cloning their clones — copycats turning around to claim that someone stole “their original work.”

The technical variable driving this anxiety, according to reports from two game outlets (Kotaku and GameDeveloper), centers on concerns about the decompilation and code-recombination capabilities of a new flagship model (Claude Opus 5.5). The sourcing must be nailed down here: this is a concern and trend judgment from media and the developer community about capability boundaries, not a confirmed conclusion that “a specific game has been cloned by that model.” But the concern itself is the signal — once developers tacitly accept that “one demo is enough for AI to spit out a knockoff” is approaching as a technical premise, the decision model for demo releases changes.

3.4 How Wide Does This Retrenchment Go?#

An honest answer: unknown. Overseas outlets mentioned several similar reactions but gave no complete list; the confirmed public cases are Digital Sun’s postponement and Alliance Arts’ cancellation. That “list incomplete” state is itself worth recording — if more studios follow suit and pull demos, this section’s judgment needs upgrading; if these are the only two cases, then they are just two isolated examples under a headline topic. This article reserves judgment and keeps watching.

4. Recalculating the Demo: Hand Players a Controller, or Hand Cloners a Map#

Now let’s assemble the previous two sections and talk about what I consider the sharpest contradiction in this wave: the cost-benefit structure of releasing demos is being rewritten by AI cloning.

4.1 Why Demos Used to Be an Indie Game’s Lifeblood#

For indie games, demos were never a side dish of marketing; they were the main course: the Next Fest exposure window directly determines wishlist scale, wishlists determine the starting point of launch sales, and launch word-of-mouth determines the long tail afterward. The whole chain is staked on “letting as many people as possible play early.” That is why this Next Fest could climb to about 8,200 demos — under the old model, handing over a playable build was almost the only solution for indie customer acquisition.

4.2 Which Parameter of the Model AI Cloning Changed#

The old model rested on a premise: once a demo was out, an imitator’s cost of reproducing a “sellable” version remained high — art, code, tuning, content volume, each requiring manpower. AI has dismantled that premise. Assembling the factual chain above, the cloner’s workflow looks roughly like this:

Each step alone is nothing new; the problem is the chain: the more complete a demo is, the closer to the final game, the more detailed the “map” handed to cloners. The old measure of a demo’s sincerity — generous content, complete systems — has been inverted entirely into a liability in the face of clone risk.

4.3 The Retrenchment Dilemma, and the Middle State It Forces#

Developers are now squeezed from both ends: without a demo, they forfeit the Next Fest exposure dividend, which can be fatal for small teams; with one, they feed the chain above. Digital Sun chose the middle state — Discord wave testing preserves the feedback channel, the public playtest is pushed to just before launch, trading “exposure for safety”; Alliance Arts simply forfeited. This is not cowardice; it is rational actors repricing under a new cost structure: when “being cloned” shifts from a low-probability event to a risk that must be seriously provisioned, the expected return of a demo has to be recalculated.

One more observation worth leaving on the table: if demo retrenchment goes from isolated cases to a trend, the first hurt will be exactly those small teams that earnestly make original games and need exposure most — big publishers have marketing budgets as a cushion; small indie developers have only the Next Fest window. If clone risk keeps rising, the participant structure of demo events will be reverse-screened — probably an outcome nobody wanted.

5. The Backdrop and Governance Signals: Order for AI Supply Is Still on the Way#

5.1 The Industry Backdrop in the Same Week#

Zooming out, this week’s cluster of news sits inside a larger backdrop — AI’s penetration of game production is still accelerating:

The production side charges ahead while the protection side visibly lags — this scissors gap is the root of those “self-preservation” moves in Section 3.

5.2 A Governance Signal: Certification Arrives, the Filter Doesn’t#

There has also been movement on governance: the UK trade body (UKIE), together with a law firm (Sheridans), launched a four-step AI compliance and certification process, culminating in an AI Certification Mark usable in fundraising roadshows. A caveat: this comes via trade-press paraphrase, not directly reviewed this round — a single-source claim. Its significance lies in direction: when “whether AI was used, and how” starts to require certification, the industry is trying to build verifiable credentials of trust for AI content, aimed at investors and the distribution chain.

But set against the players’ demands, the misalignment is obvious: what players want is a consumer-side filter to “screen out AI content while browsing the shelf,” and the tool is one official tag short; what the industry delivered is “compliance certification for fundraising,” aimed at the B side. The platform stays silent, industry bodies move first, and both players and small developers are waiting for tools. How long this governance gap lasts will decide how far that Section 3 “self-preservation retrenchment” spreads.

Laying the existing governance tools into a table by “who they serve, and how far they’ve gotten,” the gap is plain:

Reading down the left column, the consumer-side cell is nearly empty: the disclosure data closest to players has been sitting in the system all along, yet players don’t even have a checkbox; the earliest landed action happened in the fundraising layer, furthest from players. The order in which tools are supplied is exactly inverted relative to how urgent the needs are.

5.3 Division of Labor Within the Site’s AI Coverage#

Finally, a restatement of where this piece sits in the site’s AI coverage line, for readers’ convenience:

• Epic x Disney pivoting fully to “AI-first” development — vendor transformation with AI as a production tool, and developer growing pains;
• Google Playground and Unity’s AI game-creation platform — the ecosystem shock as AI drops the creation barrier to a single sentence;
• AI “reviving” shut-down games and Microsoft building a decompilation team — AI as a preservation means and official co-optation;
• The engineering methodology of AI agents rebuilding the Modern Warfare 2 source code — the capability ceiling of the decompilation technology side.

And this piece covers the other face of the same coin: what happens on the supply side when AI’s productive spillover drives the cost of “replicating someone else’s work” to near zero — oversupply, backlash, and community governance starting to strain. Only with all five pieces together is the full picture of the “AI + gaming” line complete.

6. Conclusion: When Production Cost Approaches Zero, What Becomes Scarce Is Trust and the Launch Window#

Having worked through six sections, back to the two bills at the top. The record of 8,202 demos and the two studios’ demo retrenchment look like two unrelated news items in the same week, but they are two symptoms of one cause: after AI pushed the cost of producing content to near zero, every signal built on “content volume” has depreciated — demo counts, shelf richness, even “playability” itself is no longer scarce.

What is truly becoming scarce are two things. The first is trust: players’ willingness to spend discovery cost recognizing a genuine work has become expensive after AI-generated filler piled the shelves, which is why players petition for a filter — what they want is not a technical feature but making trust identifiable again. The second is the launch window: for original creators, the head start of “being the first to turn this idea into a finished product on the table” is the one thing AI cloning cannot copy — Digital Sun pushing its public playtest to two months before launch and Alliance Arts cancelling its demo outright are both locking that window.

When we discussed this topic in the newsroom, I said the most ironic part of this wave is that the same batch of AI tools is used by some to stuff the shelf until it bursts, while others dare not even step onto the shelf because of them. Whether it’s the players petitioning for a filter or the studios postponing demos, what everyone wants is the same thing — a market where “things made earnestly are treated earnestly.” Governance tools will come sooner or later — tags, filters, certification, all of it; until then, learning to protect your launch window is probably the new required course for every indie developer in this transition.

— Wubian, NTGame Editorial

Further Reading#

• Epic x Disney Delay: Going All-In on “AI-First” Development — “No More Hand-Written Code” and the Developer Growing Pains Behind Big-Publisher AI Workflow Transformation — the other side of production-side AI transformation
• Google Playground and Unity: One-Sentence Playable Games and the Ecosystem Shock of AI Game-Creation Platforms — an ecosystem projection after the AI creation barrier collapsed
• AI “Reviving” Dead Games Goes Mainstream: Microsoft Builds a Decompilation Team and the Legal Limits of Preserving Old Games — a full breakdown of decompilation capability and official co-optation
• AI Agents Rebuild the Modern Warfare 2 Source Code in 3 Months: Multi-Agent Engineering Methodology and a Paradigm for Restoring Old Games — real-world tests of the reverse-engineering technology side
• AI Game Rebuilds Hit by Bethesda and Activision Takedowns: Copyright Holders Begin “Closing the Net” in the AI + Gaming Red-Line Moment — the copyright holders’ first counterattack