In February, OSgrid reported losing more than 800,000 standard region equivalents. In that same monthly survey, the whole of OpenSim fell by 14,636, every public grid added together.
Both numbers come from the same report by the same person on the same day. They cannot both be describing something that went missing.
Working out which one is real turns out to be the actual story of the OSgrid cleanup, and it is stranger and considerably less alarming than the headline figure suggests.
The 800,000 regions belonged to one person
The explanation is sitting in the previous month’s report, and it is not a database failure. It is a hobby project.
“Due to an OSgrid user modeling all of North America, the total number of standard-sized regions in OpenSim is still an astronomical 988,206,” Hypergrid Business editor Maria Korolov wrote in her January 2026 survey. One OSgrid user had built a model of an entire continent, and in doing so had roughly sextupled the apparent size of the open metaverse.
Korolov’s handling of this is worth quoting because it is the key to the contradiction. She decided to leave those regions out of her chart and subtract them from the monthly ecosystem total, while leaving them in the individual grid listings. Her figure for how many to subtract was “an estimate of 825,435, based on subtracting OSgrid’s size before this grand experiment,” which she described plainly as “a rough estimate.”
So by February the ecosystem total had already been excluding the North America regions for months. OSgrid’s own per-grid line had not. When the continental project shut down, OSgrid’s individual number fell off a cliff while the combined total barely registered it. The two figures were measuring different things, and only the smaller one described anything a person could visit.
We are not innocent bystanders here. When OpenSim’s headline count was sitting near a million, this site ran a piece asking whether 900,000 regions proved virtual worlds mattered. That number was mostly one enthusiast’s terrain model, and a lot of coverage, ours included, treated it as a measure of a growing world.
Why OSgrid produces numbers that behave like this
The continental project was an outlier, but the volatility underneath it is structural, and the February report explains the mechanism directly.
OSgrid lets anyone attach a region to the grid, including people running regions on home computers that are not switched on around the clock. To stop somebody else grabbing your spot on the map every time you shut your machine down, OSgrid holds a map reservation for you. If the location goes unused for long enough it is treated as abandoned, the reservation reverts, and the region stops being counted in the grid’s statistics, until the owner reconnects and picks a new spot.
Read that again as a measurement problem. OSgrid’s reported land area is, in part, a count of how many hobbyists happened to have their computers on. Nothing is destroyed when the number falls. Somebody went on vacation.
This is why OSgrid’s figures lurch while the commercially hosted grids sit still. In the same February report, Wolf Territories Grid held 33,497 regions and Kitely 17,840, both paid platforms running on servers that stay up because somebody is billing for them, and Korolov notes in it that commercial OpenSim hosting starts at less than $5 a region. OSgrid charges nothing and has been running since 2007, and the price of that is a map that breathes.
Six months later, the land came back
The useful thing about a monthly survey that has run since 2009 is that you can simply wait and look again.
In February, after the cleanup, OSgrid reported 23,956 regions. In Korolov’s August 2026 report, OSgrid reports 25,665, which is 1,709 more regions than it had immediately after the event that was supposed to have gutted it, a gain of about seven percent, and it puts the grid second only to Wolf Territories on land area.
The wider picture moved the same direction. OpenSim’s public grids totaled 148,135 standard region equivalents in February and 153,618 in August, a gain of 5,483. The ecosystem has still not climbed back to January’s adjusted 162,771, so this is a partial recovery rather than a full one, but the direction is unambiguous.
OSgrid itself is unremarkable to visit today. Its front page reports the grid online, describes itself as the largest free OpenSimulator grid, and hands out a hypergrid address for Lbsa Plaza, the same public landing point people were teleporting to before any of this happened.
What actually did get worse
The recovery story is real, and stopping there would be its own kind of dishonesty, because the number that matters most did not recover.
Active users across OpenSim were 48,018 in February and 47,425 in August, a fall of 593. OSgrid’s own actives went from 4,122 to 3,907 across those same two reports, down 215. The August report is titled for the slump: the public metaverse shed 240 regions and lost active users that month, ten grids were marked suspended with Korolov noting that if they do not return within a month or two she will mark them closed, and registered users fell by 39,179, though she attributes almost all of that to the educational grid Eureka World, which had 39,000 registered users, not publishing its statistics.
So the honest six-month summary is that the land grew back and the people did not. Land is cheap and largely automatic; a region can be reattached by one person in an afternoon. Attention is the scarce thing, and no database cleanup caused that shortfall or could fix it.
The part I would argue for
Here is where I stop reporting and state a view: the panic in February was a category error, and it is one this whole beat keeps making.
A number like 988,206 regions got treated as evidence that the open metaverse was enormous, and then a drop of more than 800,000 got treated as evidence that it was collapsing. Neither was true, and both readings came from applying the instincts of enterprise infrastructure reporting to something that is not enterprise infrastructure. Volunteer-run, home-hosted, free-to-join grids do not produce clean quarterly figures, and a chart that wobbles because a hobbyist’s PC is off is not a chart of a failing platform.
What deserves the attention is the thing buried inside the scary statistic. One person, using hardware they own and software nobody charged them for, modeled North America and attached it to a public world that anyone could walk into. Then they were done, and they switched it off, and the commons carried on. That is not a fragility story. Compare it to Meta, which moved Horizon Worlds off the headset after more than 80 billion dollars in losses, and the question of which model is precarious gets harder to answer than the headlines allow.
The correct posture toward OSgrid’s statistics is not alarm and not celebration. It is patience. Read them annually, not monthly, and understand that the noise is the sound of an open system working.
Questions readers keep asking
Did anyone actually lose their builds in the OSgrid cleanup?
The 800,000 region figure did not represent user content being deleted. It was a single continental modeling project, one OSgrid user’s model of North America, being shut down, plus map reservations reverting on regions that had gone unused. A reverted reservation frees the map coordinates, and the owner picks a new location when they reconnect.
Is OSgrid still running?
Yes. Its own site reports the grid online, and in the August 2026 Hypergrid Business survey it ranked second among all OpenSim grids on land area with 25,665 regions and second on active users with 3,907.
Why does OSgrid’s land area move so much more than other grids?
Because it is the large grid that lets anyone connect a region from a home computer that is not always on. OSgrid holds a map reservation for that spot and lets it revert once the location goes unused, at which point the region stops being counted. Its reported land area therefore partly tracks how many volunteers currently have their machines running, where commercially hosted grids report from servers that stay up.
A note on reading the chart
The next time a virtual world posts a catastrophic number, the first question worth asking is not what went wrong. It is what the number was counting last month, and whether anybody changed the rule.
In this case somebody had, openly, in print, one report earlier. It just did not make as good a headline as a vanishing continent.
