Twenty-two years of Steam releases

Most writing about Steam quotes one year and calls it a trend. The interesting shape only appears when you put the whole series in one column, next to the policy decisions that caused it.

The series

One source, fetched directly from SteamSpy on 4 August 2026, for every row. See the warning about counting at the end.

From 2014 to 2025 that is a factor of 11. From 2004 it is a factor of 1,294. Not a trend, a different medium.

The gate, and what happened each time it moved

The curve is not weather. Every bend in it has a decision behind it.

Before 2012, curation. Valve chose who got in, and the ceiling shows: 364 games in 2009, then 329 in 2010 and 293 in 2011. Two consecutive declining years. A hand-picked storefront does not scale, and for three years it visibly did not.

2012, Greenlight. Valve stopped choosing and let the community vote. The effect is the loudest number in the table: 2014, plus 203 percent.

2017, Greenlight abolished. Replaced by Steam Direct: pay a fee per title, meet the legal requirements, and you are in. No vote, no committee. The run from 4,415 to 8,081 across 2016 to 2018 is the sound of a gate being removed.

2019, and this is the part almost nobody mentions. After five years of 28 to 66 percent growth, with the gate now fully open, the number fell 4 percent. From 8,081 to 7,738. Kotaku's headline at the time said the numbers used to climb every year and then held steady, which was the polite reading.

That single down year is the most informative point in the series, because it is a natural experiment. When the gate came off, the obvious prediction was that supply would rise forever. Instead it hit a ceiling within one year and briefly went backwards. Roughly 8,000 games a year was as much as the existing population of developers could produce with the tools of 2019. Permission was never the binding constraint. Capacity was.

Then look at what broke that ceiling. Growth resumed slowly, +21 %, +8 %, +5 %, and then 2024 jumped 32 percent, the largest rise since Greenlight landed a decade earlier.

Nothing about the gate moved in 2024. No policy changed. What changed was the cost of making a game.

The other side of the counter

A storefront has two numbers, and almost everyone only quotes the first.

Steam is not shrinking. Peak concurrency grew 6.4× between 2013 and 2026, which for a platform this old is remarkable.

It is also barely half the growth on the supply side.

The number nobody publishes

Put the two series in the same division, using completed years only.

Peak concurrent players per game released that year:

• 2013/14: 6.6 m ÷ 1,637 ≈ 4,032
• 2025: 40.27 m ÷ 18,119 ≈ 2,223

A drop of 45 percent. Every new game in 2025 arrived to a crowd that was, per title, a little over half the size of the one waiting in 2014, even though the crowd itself had grown six-fold.

This is not a claim that games earn 45 percent less. It is a claim about attention density: the ratio of people present to things asking for their time. That ratio has been falling for eleven years and there is nothing in the series suggesting it will stop.

(Our calculation. Concurrency is the peak of the year, not an average, and the 2013 peak is paired with 2014 releases because that is the earliest concurrency figure we could source. Both choices flatter the older year, so the real decline is likely steeper, not shallower.)

A warning about the counting, and a correction

Different counters produce very different numbers for the same year, and this piece is itself an example of getting that wrong.

When we first published this on 4 August 2026, the table used figures assembled from several summaries. They were up to 14 percent off the numbers SteamSpy itself reports, and one claim was simply wrong: we described 2019 as growing 2 percent when SteamSpy shows it falling 4 percent. The table above is now a single series, fetched directly from SteamSpy on the same day, and the 2019 paragraph has been rewritten.

The disagreement between counters is not sloppiness, it is definition. Does software count? Do delisted titles stay counted? Are re-releases new? Does a demo count? Each tracker answers differently and stays internally consistent. For 2025 alone the published figures range from 18,119 (SteamSpy) to 21,273 (Indie Launch Lab), with SteamDB and others in between.

The rule we broke and now follow: one source for one series, never mixed inside a line. A number assembled from several counters describes no storefront that exists.

We have deliberately left 2026 out of the table. SteamDB reported 11,979 releases for the first half of the year; SteamSpy shows 9,020 for a longer stretch. Those cannot both describe the same thing, and we would rather show you an incomplete table than a projection built on a contradiction.

What this series says

Three things, in order of how sure we are.

Certain: supply grew eleven-fold in eleven years, and the sharpest recent acceleration, 2024, happened without any change to who is allowed in. The gate has been fully open since 2018.

Very likely: the 2019 decline shows the ceiling was never the gate. It was how many people could make a game with the tools of the day. That ceiling has been lifted once by removing the gate and once by collapsing the build cost, and the second lift is still in progress.

Worth watching: attention per title has nearly halved. Every discovery mechanism on the platform, the queue, the charts, the wishlists, was designed when that number was four thousand.

Sources

• SteamSpy year index, fetched directly on 4 August 2026, for the full 2004 to 2025 release series
• Kotaku, "Annual Steam Release Numbers Used To Climb Every Year, But In 2019, They Held Steady"
• SteamDB, Steam Game Release Summary by Year, for the divergent 2026 half-year figure
• Indie Launch Lab, Steam Game Release Reports 2010 to 2025, for the divergent 2025 count
• Concurrency records as reported for 2013, 2024, 2025 and the 11 January 2026 all-time peak

Retrieved 4 August 2026. SteamDB refuses automated access, so figures attributed to it come from published summaries rather than our own request. Where we calculated something ourselves it says so in brackets. This article was corrected on the day of publication, and the correction is described above rather than made quietly.