Module 10 · Business

Distribution and marketing: the fight for visibility

Distribution stopped being the wall: Steam killed the retail gatekeepers — anyone can publish. But a new wall grew in its place — visibility. The real "tax" isn't the platform's 30% cut, it's obscurity: you keep 70% of nothing if nobody found you.

~18 min💰 business + GTM

The gist in 30 seconds

The wall used to be distribution: shelf space in retail, a publisher gatekeeper, 10–20% for the developer. Steam (2003→open to everyone by 2013) flipped the power: platform > developer > player, a 30% cut, anyone can publish. But that produced a new wall — visibility: 10,000+ games on Steam, median sales in the hundreds to low thousands of copies. The real "commission" isn't 30%, it's obscurity. Platform economics: Steam 30% / 25% (>$10M) / 20% (>$50M), Epic 12% (+0% on the first $1M), mobile historically 30% — but 2024–2026 antitrust (Epic vs Apple/Google) is breaking it (Google ≤20% + external payments; Apple forced to allow external payment links with no cut in the US). Subscriptions (Game Pass) pay for play, not for a sale — a shift of incentives toward retention. Marketing in 2026 = wishlists (the best predictor of launch), community (the Discord flywheel), virality (Lethal Company: $0 marketing → ~10M sales on "clippability"). The winner isn't whoever made the best game — it's whoever got found.

The mechanism: power moved, the wall stayed

The shift of platform power

In retail the chain was publisher → retailer → developer → player: shelf space was finite, gatekeepers (the publisher, the chain's buyer) decided who shipped at all; the developer got 10–20%, and the barrier to entry cut off everyone but the big players. Steam (2003 as DRM for Half-Life 2, open to everyone by 2013) removed manufacturing, the shelf and the gatekeeper: platform → developer → player, 70% to the developer. But an infinite shelf + open entry = a new wall: visibility.

The real commission is obscurity

On mobile the median is under $10K over a game's entire life; on Steam there are tens of thousands of releases a year. "You can keep 70%, but if nobody found you it's 70% of nothing." Discoverability, not the platform cut, is where revenue actually gets cut. The break-even point is sobering: with cost $C$ , price $p$ and commission $r$

u^{*} = \frac{C}{p (1 - r)}

A $50K game at a $20 price with a 30% cut (you take $14/copy) breaks even at $50000 / 14 \approx 3571$ copies. Doesn't sound like much — but most indies sell 500–5000 over the whole lifetime; 3571 is already above the median. The commission moves the threshold by percent, visibility moves it by multiples.

Platform economics (and the 2026 antitrust break)

Channel	Platform cut	2026 wrinkle
Steam (PC)	30% · 25% (>$10M) · 20% (>$50M)	infrastructure/discovery/refunds; ~74% of the PC market
Epic Games Store	12% (+0% on the first $1M/year)	cheaper, but the traffic is small (~3% of the market)
Apple / Google (mobile)	hist. 30% · 15% (<$1M / subscriptions)	the break: Google ≤20% + external payments (US/UK/EU from mid-2026); Apple obliged to allow external links with no cut in the US (case at the Supreme Court)
Consoles (PS/Xbox/Switch)	20–30% (via the publisher)	usually covered by the publisher, not the developer

The "30% debate" dragged on for years (Valve: that's servers, payments, discovery, refunds, cloud saves; critics: it's rent). What moved it wasn't competition, it was the courts: Epic vs Apple and Google (2020→) forced the mobile stores in 2024–2026 to crack open external payments and trim the cut — a rare case of platform power being broken by a regulator rather than by the market.

Subscriptions: paid for play, not for a sale

Game Pass (2017) and PS Plus are "Netflix for games": studios get roughly $0.50–2 per player per month (estimates), that is, money for being played, not for being bought. Incentives shift: time-in-game > a single playthrough, retention > virality. Microsoft is estimated to be down $10–20B cumulatively — a bet on the ecosystem and lock-in (the Ultimate price jumped $19.99→$29.99 in Oct 2025→$22.99 in Apr 2026: the model lives in price instability). For an indie a subscription means access to millions with no marketing budget, but also dilution of one-off sales and DLC.

Marketing = a visibility machine

If the wall is visibility, marketing is the assault on it. Three levers (for the Steam specifics — the page and the algorithm — see Steam, wishlists and launch):

Wishlists — the best predictor of launch (≈100 followers → 1–5 copies on release day; 5–10K wishlists → $50–200K at launch). They're built over months: announce 6–12 months out, devlogs, Steam Next Fest, a mailing list.
Community — the Discord flywheel: announcement → early subscribers → regular devlogs (on a schedule) → fan content/playtests → back into social → 1000+ evangelists at launch. A direct channel with no algorithmic filter.
Virality and influencers — trust transfers from the streamer to the audience; 10 keys → a couple of streams → hundreds of wishlists. TikTok/Shorts is the new "trailer distribution" via user clips. Lethal Company (2023, solo, $0 marketing) took off on "clippability": every 5-minute session produces 10 potential clips.

🕹 What to play — and what to notice

Lethal Company virality out of nothing

Solo developer Zeekerss, ~$0 marketing → ~10M copies, $100M+ revenue, ~$33M net to one person. The fuel is "clippability": funny deaths, co-op chaos, a mechanic you understand in 10 seconds ("grab the scrap, don't die").

🎮 Notice: play a couple of sessions (or watch clips) and count how many "clippable" moments come out of 5 minutes. Compare with a slow cinematic game — it has almost none of them, and its viral odds are near zero. You can't plan virality, but clippability raises the odds.

Steam "New Releases" the median of obscurity

Steam's new-release feed is exactly that wall: thousands of games, most of them with single-digit review counts. A visible power law: the top 1% takes almost all the attention.

🎮 Notice: open Steam → New & Trending → scroll through into "just new". Count how many games in a row have <10 reviews. These aren't bad games — they're invisible ones. That is what all of marketing works against: not "make it better" but "get found".

Balatro word of mouth + streamers

Solo, a card roguelike, low art cost. 5M+ sales (by Jan 2025) with almost no budget — on word of mouth, streamers and "one more run" clippability.

🎮 Notice: look at how people find out about Balatro — almost always "a friend/streamer showed me". That's trust transfer in its pure form: not an ad, a recommendation. Work out what your Novgorod's word-of-mouth hook is — what would make someone show the game to a friend.

Deep end · discoverability as a power law and a funnelskippable

Sales follow a power law

Outcomes on an open platform are a power law: the median is near zero and the top 1% takes the overwhelming share (Stardew 41M+, Balatro 5M+ against thousands of games at 500 copies). That's not a bug, it's a property of a market with infinite supply and finite attention: attention is the resource, and it concentrates. So the "average outcome" is meaningless for planning — compute the survival threshold (break-even) and the distribution, not "expected" sales.

Marketing is a funnel of signals

Wishlist → launch e-mail → purchase: every step has a conversion rate. Wishlists predict launch better than reviews because they're an intent signal ("I want this"), not a rating. Multiply: $launch \approx W \cdot c$ , where $W$ is wishlists and $c$ is conversion (1–5%). Which makes all of pre-launch an exercise in accumulating $W$ : Next Fest, devlogs, community, streamers. The platform cut is a multiplier on a sale that already happened; visibility decides whether it happens at all.

Deep end · economics: commissions, antitrust, subscriptionsskippable

Why 30% held and what moved it

30% is the "default price point" inherited from the console era; it held because Steam owns discovery and the audience (network effects: players are where the games are; games are where the players are). Epic undercut to 12% + 0% on the first $1M — and barely moved its share (≈3%), because a cheap cut doesn't buy traffic. What moved it wasn't competitors but antitrust suits: Epic v. Google (Google lost, injunction) and Epic v. Apple (Apple obliged to allow external payment links with no cut in the US; the litigation is heading to the Supreme Court). The lesson: platform power = power over discovery and payment, and it gets broken by a regulator, not by price.

A subscription changes the objective function

Pay-per-play (~$0.50–2/player/month) redefines what you optimize: not "sell a copy" but "hold attention". That's good for indies (access with no marketing) and dangerous for design (chasing an engagement metric → Goodhart). Plus a subscription cannibalizes one-off sales and DLC — which is why big publishers keep entering and leaving the catalog.

Analogy

Distribution used to be like getting a book onto a store shelf: a handful of gatekeepers (publishers, chain buyers) decided what existed at all, and if they didn't pick you, you didn't exist. Steam tore the store down and built an infinite library where anyone can put a book on the shelf. But now 10,000 new books arrive every month and the shelf is infinite — and the problem isn't "getting on the shelf", it's being found on it. The 30% the library takes is small change next to the real tax: obscurity. Marketing is how you get yourself a spotlight in an infinite library.

Why it matters

For an engineer who will one day ship their own thing (your Novgorod), this redefines the priorities: making it good is necessary but not sufficient; with infinite supply the found one wins. Understanding commission economics, the shift of platform power and the mechanics of visibility (wishlists, community, clippability) is what keeps you from building an excellent game nobody sees. And the frame "when supply is infinite, the bottleneck isn't production but discovery and ranking" is one of the central ones in modern systems.

🔁 Beyond games — where this transfers

The lesson is about the attention economy: when supply is unbounded, the bottleneck moves from production to discovery and ranking.

ML / AI (your domain): the visibility crisis is exactly the recommendation/ranking problem and the attention economy: when models/apps/content are infinite, the bottleneck isn't training, it's being found, and the power sits with whoever holds the ranking algorithm (Steam discovery queue, app store, the feed) — platform power = ranking power. Wishlist-as-signal ⇄ implicit feedback / an intent signal for ranking. Clippability/virality ⇄ designing for share-ability and the growth k-factor. The power law of outcomes ⇄ winner-take-most platform dynamics and why cold-start / discovery is a hard problem. "70% of nothing" ⇄ a great model nobody finds is worthless: deployment and distribution are first-class problems, not a postscript (in MLOps the model is 10% of the product). Pay-per-play subscriptions ⇄ usage-based pricing and optimizing for engagement with its Goodhart risks. Platform antitrust ⇄ the debate over concentration of compute/platform power in AI.

Product/startup: "build it and they will come" is a myth; go-to-market and distribution often decide more than the feature set.

Platform economics: network effects, two-sided markets, commission vs traffic, the regulatory break-up of a discovery monopoly.

Principle: with infinite supply, optimize for being found, not only for quality; compute the survival threshold and the distribution, not the "average" outcome.

🔧 Take it apart and do the math

🧮 Launch economics ~20 min

For a hypothetical release of your Novgorod, compute the break-even point

u^{*} = C / (p (1 - r))

for Steam (30%) vs Epic (12%) vs mobile-2026. Then estimate the wishlist funnel: how much

W

you need at 2% conversion to break even at launch.

🎮 Clippability audit ~15 min

Take a viral game (Lethal Company, Content Warning) and your own idea. Write down 5 "clippable" moments for each. If yours has few — what would you add so that every session produces content? Separately, sketch the Discord flywheel: a devlog schedule, the first community hook.

Checklist: computed break-even on 3 channels; built a wishlist funnel; ran a clippability audit; sketched a community plan; connected visibility to ranking/attention economy.

Connections

foundation

Steam, wishlists and launch — the page and the visibility algorithm on Steam (tactics); here it's strategy and economics.

Studio economics — the commission and sales feed break-even, burn and portfolio risk.

The monetization spectrum — how to make money once they have found you.

Cloud and subscriptions — subscription/cloud as a distribution channel and a shift of incentives.

Questions worth asking

Is 30% fair? Why didn't competition push it down?

Valve's argument: 30% covers servers, payments, refunds, discovery tools, cloud saves, multiplayer infrastructure, Steam Community — and, above all, the audience. The critics' argument: it's rent, the costs stopped being 30% long ago. The key fact is why competition didn't cut the price: Epic undercut to 12% (+0% on the first $1M) and still holds ≈3% of the market against Steam's ≈74%. Because a cheap cut doesn't buy traffic: players are where the games and the friends are (network effect), and the developer cares more about discovery than about an extra 18%. The price was moved not by competitors but by antitrust courts — though so far mostly on mobile (Google ≤20% + external payments; Apple obliged to allow external payment links with no cut in the US, with the case heading to the Supreme Court). On PC, Steam's 30% model holds — not because it's fair, but because a discovery monopoly is hard to challenge with the market.

Why is visibility a bigger problem than the commission?

Because the commission is a multiplier on a sale that already happened (30% vs 12% changes revenue by percent to tens of percent), while visibility decides whether the sale happens at all (0 vs N — that's multiples and orders of magnitude). The indie median is 500–5000 copies over a lifetime; with an infinite shelf and finite attention, sales follow a power law — the top 1% takes almost everything. "You keep 70%, but 70% of nothing is nothing." So the rational priority is: first get found (wishlists, community, clippability, streamers), and only then optimize the commission by picking a storefront. New developers systematically overrate the commission ("I'll go to Epic for 88%") and underrate discovery ("there are fewer players there → 88% of less = less money").

Can you plan virality?

Virality itself, no — it's stochastic (Stardew, Balatro and Lethal Company all leaned on luck to some degree, and most solo projects don't take off). But you can raise the odds by designing for clippability and share-ability: every session produces "moments" (a funny death, an epic win, co-op chaos), the mechanic is legible from a 10-second clip, there's group play (streamer + friends = content). Lethal Company isn't "randomly lucky": it structurally generates ~10 clips per 5 minutes, while a slow cinematic game generates roughly zero. The growth analogue is the k-factor: how many new players one player brings in through content/invites; you can't guarantee virality, but you raise k by design. Plus distribution of odds: build a game that scales if it takes off (co-op, replayability), so that luck, if it arrives, converts into retention.

How does this connect to recommender systems and ML?

Directly: the visibility crisis is the attention economy, and its engine is ranking/recommendation. When supply is infinite (10K games, endless content, thousands of models on HuggingFace, millions of apps), the bottleneck moves from production to discovery, and the power sits with whoever holds the ranking algorithm (Steam discovery queue, app store, the recommendation feed). That's the same reality as ML systems: training a model/feature is cheap, and distribution and ranking are the moat. A wishlist = an implicit intent signal for ranking; clippability = designing for share-ability and the k-factor; the power law of outcomes = winner-take-most and why cold-start (a new item with no signals) is a hard recommendation problem. And "70% of nothing" is a direct MLOps lesson: a model nobody finds/deploys is worth zero; deployment and distribution are a first-class part of the product, not a postscript. If you can think about a game's visibility, you're already thinking about ranking and cold-start.