About a fifth of Steam's 42 million are in a game
Valve counts signed-in clients, not players. Read in the same minute, the in-game share is 23.7%; divide a quiet morning by the daily peak and you get a figure that describes no moment at all.
What the counter counts
Steam's statistics page carries two figures side by side. At 09:46 UTC on 9 November 2026 it read 27,188,646 clients signed in, with a 48-hour peak of 42,306,574.
The peak came at 14:15 UTC the previous afternoon. Valve's own chart feed, whose series is labelled "Steam users logged in", puts that minute at 42,304,176. The 2,398 difference between the page and the feed is where the samples landed, not a revision. Both round to 42.3 million.
A client enters the counter when it authenticates and stays connected. It does not need a game open, and it does not need a person in front of it. Valve's label is precise about this: logged in. Headlines that read "42 million players" swap those two words for a different claim.
Valve does not publish a global in-game total on that page. Sites that quote one are summing the per-app counters Valve exposes through its public API. LiveCounts does that and prints both totals on one screen, which is what makes a same-minute division possible at all.
The division that holds
At 09:44 UTC on 9 November, LiveCounts showed 27,252,978 online and 6,472,228 of them in a game. A second read four minutes later showed 27,375,484 and 6,501,618.
Both pairs return 23.7%. Slightly under a quarter of signed-in clients had an app running at that hour.
| Metric | Value | Read at (UTC) | Counts |
|---|---|---|---|
| Clients signed in | 27,375,484 | 2026-11-09 09:48 | Authenticated sessions |
| Clients in a game | 6,501,618 | 2026-11-09 09:48 | Sessions with any app running |
| Share in a game | 23.7% | 2026-11-09 09:48 | Row one divided by row two |
| Top 100 titles, combined | 4,243,402 | 2026-11-09 09:48 | Sum of the hundred most-played apps |
| Peak sign-ins | 42,306,574 | 2026-11-08 14:15 | Busiest minute of the 48-hour window |
The last row is not a reading from the same minute as the rest. Everything above it comes from one Monday morning, when Europe is at work and the Americas are asleep. The peak comes from a Sunday afternoon.
The division that gets published
Put the in-game figure over the peak: 6,501,618 divided by 42,306,574 is 15.4%.
Every input is real and copied correctly. The output describes no moment that existed, because the numerator is a quiet weekday morning and the denominator is the busiest minute of the weekend. Sample an hour earlier or later and the answer shifts again. That is the tell: the ratio tracks when you opened the page, not what the platform was doing.
The 15.4% version reads as the better story. It sounds stricter and more damning than 23.7%, and it confirms what a tired reader already suspects about platform numbers. It is a third figure assembled from two.
There is a related number nobody can currently supply. We could not find a stored reading of in-game players for 14:15 UTC on 8 November, so we do not know the share at the peak. No source we could reach keeps that minute. Estimating it from the morning ratio would repeat the error in the other direction.
A tenth of the peak, two thirds of the in-game count
The hundred most-played apps accounted for 4,243,402 clients at 09:48. Against the 42.3 million peak, that is 10.0%. Against everyone in a game in that same minute, it is 65.3%.
One number, two meanings. The 10.0% is the one that travels, because it makes the top of the market sound like a rounding error beside the platform total. The 65.3% is the one that describes the platform: two out of three clients running something on Steam were running one of a hundred apps.
The remaining third is where the definition loosens. It spreads across thousands of titles, and also across things that are not titles in any useful sense.
Wallpaper Engine sits at rank 4 in Valve's most-played rollup for the week ending 8 November 2026, with a peak of 118,429 clients running it. Its Steam store page lists the genres as Animation & Modeling, Design & Illustration, Photo Editing and Utilities. The app animates a desktop wallpaper and then keeps running. Steam files it in the same column as Counter-Strike 2.
Source SDK Base 2007 appears further down the same table on Valve's stats page. It is a development runtime, and it counts as in a game.
So in a game means a session has an app attached to it. Nothing in the definition checks that the app is a game, or that anyone is at the keyboard.
What the counter is good for
None of this makes Valve's figure wrong. It is a load number: how many clients the login service and the content network have to serve. For capacity planning, patch delivery and storefront traffic it is the right measurement, and nobody outside Valve can reconstruct it.
It is also the only number Valve prints at platform scale, which is why it gets borrowed for questions it was never built to answer. A journalist wants to know how healthy the PC market is. The counter can only say how many clients are connected.
The peak behaves predictably. In the 48 hours we pulled, the high point landed at 14:41 UTC on 7 November and 14:15 UTC on 8 November, and the lowest reading of the window came at 06:36 UTC on the Monday, 24,728,471. Peak against peak is a fair comparison. This morning's concurrency against last night's peak is not.
How we read it
Two numbers can be divided only when both carry the same timestamp. Where a dashboard prints no reading time, we treat the figure as unusable for any ratio, including one we would like to publish. When the timestamps differ, we say so and stop.
The second check is the numerator. Sessions, clients and people are three different quantities, and a chart does not always tell you which one it is showing. Valve is consistent about it if you read the label. Aggregators are less so.
This is the third entry in a series about the same habit: why we stopped scoring games out of ten, then what the refund window data actually showed. Both were about checking what a published number counts before treating it as a verdict. This one is about a number that gets misread in the same direction every time it sets a record, because the record is a peak of connections and the headline is about players.
Valve's page updates every minute. The 48-hour peak will roll forward with it, and the next record will be a peak of logins, not a count of people playing.
