We gathered hundreds of real game budgets. Some come from public sources (press coverage, developer post-mortems, conference talks, web archives and court filings), others were shared with us privately, under NDA. We converted every budget to US dollars and adjusted it for inflation, so a 2010 game can be compared with a 2024 one.

We then trained a machine-learning model to predict those budgets from what can be observed about a game. That includes the size, roles and experience of the team listed in its credits, the studio's and publisher's track record, the game's install size, launch price, platforms, languages and early sales. It also uses design markers like multiplayer or 2D. We checked the result on games the model had never seen.

We then ran that model across ~100k Steam games to build a much larger catalog of estimated budgets. This tool doesn't query the ML model directly, it queries a simpler, faster stand-in model fit on that catalog. For each input, it returns nine estimates: the typical budget (median) plus eight other levels, from q10 (the cheaper end of comparable games) up to q90 (the more expensive end). The gap between them tells you how varied budgets are among games that look like yours.