Module 10 · Business

Studio economics: the math of making games

Games are a hit-driven business with high fixed costs: you pay a full team for years before the first revenue, and the outcome is distributed by a power law. Everything follows from that: why AAA budgets exploded, why studios avoid risk, and why the shape of the organization is a strategic bet on managing that variance.

~18 min💰 economics + org structure

The gist in 30 seconds

Making games is a hit-driven business with brutal economics. AAA budgets grew ~10× in 20 years: GTA VI (out Nov 19, 2026) is estimated at $1B+ — the most expensive game in history — against ~$265M for GTA V; a $200M game has to sell 3–5M copies to break even, so a failure is catastrophic and investors flee into sequels. The "missing middle" (AA) collapsed in 2015–20 and came back in the 2020s (BG3, Hades, Balatro) thanks to streaming-as-visibility + Game Pass + talent freed up by layoffs. Indie math is simple and harsh: break-even = cost / (price × (1−commission)); most sell 500–5000 copies, the top 1% (Stardew 41M+, Balatro 5M+) lives for years off a single hit. Survival is governed by burn and runway. And org structure is strategy: Valve (flat), Supercell (small cells, kill fast), Nintendo (auteur + hardware), Tencent (invest in everything), solo (constraints as innovation) — each produces different games. The through-line: high fixed costs + hit-driven outcomes = portfolio risk, and the shape of the company is a bet on how to manage it.

The mechanism: a hit-driven business and its math

High fixed costs + a power-law outcome

The key structure: you pay the whole team salaries years in advance and only learn whether it's a hit at release. Most titles lose money or break even, a few hits carry the whole portfolio — like venture. Which gives you the two main survival metrics. Burn — how much you spend per month; runway — how long the money lasts:

runway = \frac{M}{n \cdot s}

where $M$ is money in the bank, $n$ is team size and $s$ is cost per person per month. 30 people × ~$10K/month = a burn of $300K/month; $9M in the bank → $9 M / 300 K = 30$ months of runway. Fail to ship or raise a round before the runway ends and the studio is dead, however good the game was.

The AAA escalation: $200M+ is the norm

Why it costs so much: talent (100 people × 4 years × $150–300K = $50M+ in salaries alone), technology (engines, mocap, voice acting = $20M+), marketing ($50M+), scope creep ("next-gen" demands new systems) and live service after release. A $200M game has to sell 3–5M copies at $60–70 to break even:

u^{*} = \frac{B}{p (1 - r)} = \frac{200 M}{60 \cdot 0.7} \approx 4.8 M

A failure is catastrophic (Cyberpunk 2077 >$300M + day-one bugs; Concord shut down after ~2 weeks in 2024). Investors demand less risk → sequels, franchises, safe bets. That's a direct consequence of hit-driven math, not cowardice.

The "missing middle" of AA — and its return

The AA studio (50–100 people, $10–50M) nearly died out in 2015–20: it couldn't compete with AAA on marketing/graphics and was riskier than indie ($30M needs 1M+ sales). It came back in the 2020s (BG3, Elden Ring, Hades, Helldivers 2, Balatro) because: streaming made quality visible (people watch BG3 on Twitch for free), Game Pass gives you an audience with no marketing, and after the layoff wave talent preferred stable mid-size studios. Current AA economics: $10–50M, 30–80 people, 3–4 years, a target of 500K–2M copies, 30–50% margin (higher than AAA — less marketing).

Indie: simple but harsh arithmetic

Break-even $u^{*} = C / (p (1 - r))$ : a $50K game at $20 with a 30% cut breaks even at ~3571 copies. Reality: most indies sell 500–5000 over a lifetime; 10% do 5000+; 1% do 50K+; the top 1% do 500K+ (Stardew 41M+, Balatro 5M+, Hollow Knight). Sustainability thresholds: solo with minimal costs — 500–1000 copies per game; a team of 5 — 10K+. The path: the first game builds an audience (it may run at a loss), the second reaches that audience plus new ones, the reach compounds; or one hit feeds you for years.

Org structure is strategy (shape determines product)

Model	Structure	What it produces · the price
Valve	flat, self-organizing	masterpieces (HL2, Portal) · slow, paralysis, top talent only
Supercell	small cells, "kill fast"	high-ROI mobile · brutal churn, smaller games
Nintendo	auteur + hardware	icons, high attach rate · expensive, slow to online
Tencent	holding company, stakes in everything	risk diversification, distribution in China · little control, regulatory risk
Solo	one person	Stardew/Balatro, high margin · luck, doesn't scale

Each shape is an answer to one question: how do you survive hit-driven variance. Valve bets on rare geniuses and tolerates long cycles; Supercell bets on many cheap bets with fast culling; solo bets on minimal cost and high variance. Structure isn't neutral: it determines which games you're capable of making at all.

🕹 What to take apart — and what to notice

Larian / BG3 AA→blockbuster over a decade

Larian built expertise through Divinity for years before BG3 became a phenomenon. A classic AA renaissance: quality + streaming visibility + time, not a one-off $500M swing.

🎮 Notice: look at Larian's history (Divinity: Original Sin 1→2→BG3) — that's compounding of audience and skill across games, not instant success. Tie it to your own path: the first game builds the base for the second.

An AAA failure the price of one bet

Concord (2024) was shut down after ~2 weeks; Cyberpunk 2077 (>$300M) survived only on CDPR's reputation and years of patches. One miss on a mega-bet can kill a studio.

🎮 Notice: find a AAA release that flopped and estimate its break-even (budget / ($60×0.7)). How many copies did it need and how many did it sell? Get a feel for why investors demand sequels after something like that.

Stardew / Balatro a solo hit feeds you for years

One person, minimal costs, a margin close to 100%. Stardew (41M+, ~$150M+) and Balatro (5M+) are proof that the constraints of solo development can become an advantage.

🎮 Notice: compare the margin of a solo hit (no salaries, just tools) with AAA (hundreds of salaries × years). At equal revenue, who comes out further ahead? That's why the solo model is durable despite depending on luck.

Deep end · the portfolio math of a hit-driven businessskippable

The expectation lives in the tail

With a power-law distribution of outcomes the expected value is dominated by the tail: a few hits produce almost all of a portfolio's revenue, and the "average" title loses money. So planning around "expected sales" is meaningless — you have to think in terms of the distribution and the portfolio: either make many cheap bets and kill the losers fast (Supercell), or hold enough capital to survive one big bet failing. A single $200M bet with no portfolio = a binary life/death for the studio; hence the rational pull toward franchises (a known IP narrows the variance).

Burn, runway and the "valley of death"

While $runway = M / (n s)$ ticks down there's no revenue — that's the "valley of death" between the start and release. Growing the team $n$ speeds up development but shortens the runway (you burn faster) — a fundamental speed↔survival trade-off. Classic deaths: scope creep pushed the schedule past the runway; hired too many too early; the hit didn't cover the accumulated burn. The discipline: a minimal team until validation, a playable slice early (see scope and prototype), money for 6–12 months past the planned release (it will slip).

Deep end · org archetypes as risk strategies (Conway's law)skippable

The shape of the organization shows up in the product

Conway's law in game dev: the system mirrors the team's structure. Flat Valve produces a handful of deeply polished things and can't ship quickly; Supercell's cells produce many small, fast-to-test games; auteur Nintendo produces coherent visionary titles but adapts slowly to online. You don't pick the "best" structure — you pick which class of product you want and how much variance you can stand.

Kill fast as variance management

Supercell cancels games that miss their KPIs by around month 3 — that's an explicit strategy against hit-driven risk: spawn many bets cheaply, cut the losers early on a metric, double down on the winners. The price is brutal churn and "monetization ahead of design". The solo model is the opposite pole: one bet, but at near-zero cost, so a failure isn't fatal and a hit is life-changing. Both are rational answers to the same variance from different sides of capital.

Analogy

A game studio is a venture fund investing in one or two startups at a time, except each "startup" is a multi-year all-or-nothing bet where you pay the whole cost up front and only find out whether it's a hit at release. AAA is a single mega-bet (one $200M swing; a miss kills the fund). Supercell is a seed fund (many small bets, kill the losers fast, double down on the winners). Solo is a bootstrapped founder (tiny cost, high variance, one hit changes your life). Org structure is the fund's thesis about how to manage variance.

Why it matters

For you, someone who will one day lead their own project (Novgorod) and works in the industry, this is a survival map: how much money you need (runway), how much you have to sell (break-even), why big bets pull toward sequels and which org shape fits your risk. And the frame itself — high fixed costs + a heavy-tailed outcome = portfolio thinking; team structure shows up in the product — carries over to any R&D organization, including research bets in AI.

🔁 Beyond games — where this transfers

The lesson is about the economics of hit-driven R&D: high fixed costs, heavy-tailed payoffs, portfolio and team structure.

ML / AI (your domain): hit-driven + high fixed costs + a power-law outcome ≈ the economics of frontier training and research bets: a huge fixed cost (a training run, a research program) paid up front against an uncertain heavy-tailed payoff; most experiments fail, a few carry the portfolio (research project selection). Burn/runway ⇄ the compute budget and the discipline of "ship before it runs out". AAA scope creep → the pull toward sequels ⇄ the pull toward safe model increments vs risky bets, and why big labs "franchise". Org archetypes shaping the product ⇄ Conway's law in AI: flat research vs directed, small autonomous teams vs a monolith determine which AI you're capable of building. A portfolio of parallel bets with fast culling (Supercell) ⇄ an eval-gated portfolio of experiments: run many, kill the ones that miss the metric. And "constraints as innovation" (solo) ⇄ how compute/data constraints force architectural inventiveness (efficient models out of resource-constrained labs).

Venture/startups: power-law returns, portfolio construction, burn/runway, the "valley of death" — literally the same laws.

Organizations: Conway's law — a system's architecture mirrors the team's communication structure; the shape of the company is a strategic choice, not a detail.

Principle: with heavy-tailed outcomes think in portfolios and survival thresholds, not "averages"; grow the team deliberately (speed cuts runway); pick the structure to fit the product class and your tolerance for variance.

🔧 Do the math and classify

🧮 Runway and break-even ~20 min

Model your own mini-studio: team

n

, cost

s

/month, money

M

→ compute the runway. Then, given a price and a commission, work out how many copies you need to cover the burn accumulated by release. Play with it: +2 people speed up development by X months but cut the runway — where's the point past which hiring kills the studio?

🗂 Studio classification ~15 min

Take 3 studios you admire and assign each to an archetype (Valve/Supercell/Nintendo/Tencent/solo). What does their structure allow and what does it forbid? Which shape fits your Novgorod in phase 1 (Canada) and phase 2 (Russia)?

Checklist: computed runway and break-even; found the point where hiring cuts survival; classified 3 studios into archetypes; picked an org shape for your phases; connected it to research portfolios/Conway's law.

Connections

foundation

Distribution and marketing — the commission and sales volume feed break-even and runway.

Industry economics — the macro cycle of capital and layoffs; here it's the view from inside a single studio.

foundation

Indie production — scope/planning at the project level: how not to stretch the schedule past the runway.

The monetization spectrum — the revenue model you pick sets the whole economics of the studio.

Questions worth asking

Why are AAA budgets growing, and is it sustainable?

They grow because of a chain of factors: top-talent salaries ($150–300K × hundreds of people × years), "next-gen" expectations (new physics/graphics/AI systems every generation), marketing ($50M+), scope creep and ongoing live-service costs. GTA V cost ~$265M, GTA VI is estimated at $1B+ — the most expensive game in history. Is it sustainable? Probably not in its current form: break-even is already 3–5M+ copies, and a failure (Concord, early Cyberpunk) is catastrophic, so the industry rationally contracts toward franchises and sequels (they narrow the variance) — which strangles novelty. This is part of the barbell: expensive safe franchises at the top, cheap indies at the bottom, with the middle (AA) squeezed out and then partly restored via streaming/Game Pass. Long term the pressure runs two ways: cheaper production (including GenAI tooling) and a shift to service revenue that spreads payback across years instead of a single launch.

What killed the "middle" AA and why did it come back?

What killed it in 2015–20 was a squeeze from both sides: from above, AAA raised the bar on marketing and graphics (AA couldn't compete for visibility); from below, indies got free Unreal/Unity and asset stores (cheap competition). AA got stuck at the worst point of risk: a $30M game needs 1M+ sales, one failure burns the studio down — riskier than a cheap indie, without a AAA marketing budget. It came back in the 2020s for three reasons: (1) streaming made quality visible with no marketing budget — people watch BG3/Hades on Twitch, word of mouth replaces ads; (2) Game Pass pays for catalog inclusion, providing an audience and a revenue cushion; (3) talent, after waves of layoffs and burnout, preferred stable mid-size studios to mega-corporations. The result is a renaissance (BG3, Elden Ring, Hades, Helldivers 2, Balatro) with 30–50% margins (higher than AAA thanks to less marketing). The lesson: the economics of a genre/segment get flipped not only by costs but by visibility channels.

How much money do you actually need to make a game?

You compute it through runway, not by feel. runway = money / (team × cost per person per month). A solo developer on savings with cheap tools can go for years at almost zero burn — which is why the solo model is durable. A team of 5 at ~$8–12K/person/month burns ~$40–60K/month, so a year of development is $0.5–0.7M in people alone, and you need a reserve for the release to slip (it always slips) — budget for 6–12 months after the planned ship date. The critical mistake is growing the team early: more people speed up development but cut the runway (you burn faster), and if the hit doesn't cover the accumulated burn, that's the end. The rule: a minimal team until you've validated with a playable slice, hire only once the product risk is gone, keep money in reserve for the overrun. And remember the power law of outcomes: plan to survive a failure, not to "break even on average".

Why does org structure affect the product so much?

Because of Conway's law: a system tends to mirror the communication structure of the team that built it. Valve's flat self-organization produces a handful of deeply polished things but can't ship quickly or predictably (no authority to kill or push a project → long cycles, risk of paralysis) — and it only works with top talent. Supercell's small cells with "kill fast" produce many small, fast-to-test mobile games with monetization ahead of design. Auteur Nintendo produces coherent visionary titles integrated with the hardware but adapts slowly (years to get decent online). Tencent's holding structure diversifies risk with stakes in everything but doesn't let studios act independently. You don't pick the "objectively best" structure — you pick which class of games you want to make and how much variance you tolerate: a structure simultaneously allows one thing and forbids another. The same law in AI: flat research vs directed, small autonomous teams vs a monolith determine which systems you're capable of building at all.