The Steam2 Teraleak: what the 12TB Valve archive actually contains

What happened

The community is investigating a reported archive of roughly 12 TB from Steam’s old “Steam2” content-delivery infrastructure. The material covers approximately 2003 to 2013 and is being discussed under the name Steam2 Teraleak.

The first widely shared signal came from Valve researcher Gabe Follower on August 29, 2026. Follow-up reporting describes thousands of historical depots: versioned packages that held game files, updates and related content on the old Steam platform. That makes this a historical content archive, not one isolated build and not a modern Steam database.

The most important correction is about how the material became available. Based on the provenance checks reported by Gabe Follower and the technical reporting so far, the files appear to have been obtained through an old publicly accessible endpoint. There is no verified evidence at this stage of someone breaking into Valve’s current network. That still leaves important questions unanswered: who assembled the archive, when it was collected, and why a retired service was reachable for so long.

The archive is also being described as around 12–13 TB depending on how the size is counted. The number is therefore best treated as an estimate, while the exact inventory is still being catalogued.

Why does it stop in 2013?

The date range matches the end of Steam’s older distribution model. Before SteamPipe, Steam used formats such as GCF and NCF to package content. In 2013 Valve began moving games to SteamPipe, a newer system based on standard HTTP file trees and differential updates.

That migration explains why the collection appears to have a sharp cutoff around 2013. It does not prove every file came from one single server or one single moment, but it makes the Steam2 explanation technically plausible. The archive is best understood as a view into a retired generation of Steam infrastructure, before the current content pipeline became standard.

What has been found so far?

Portal 2 and the F-Stop connection

Portal 2 is currently the clearest example of how different Valve’s development path could look before release. Researchers and specialist reporting have identified a 2009 beta build with a different setting and structure from the final game. The version is associated with a more ruined Aperture Science, an earlier presentation of GLaDOS and different uses for the personality cores.

The same material includes assets related to F-Stop, the cancelled project that preceded Portal 2. F-Stop was built around a camera and photography mechanic rather than the portal gun. What has been described so far looks like development material and assets, not a complete, polished F-Stop release. That distinction matters: having a project’s files is not the same as having a fully playable build.

Left 4 Dead

An early Left 4 Dead build from around June 2008 has also been reported. Compared with the released game, it is said to contain a different HUD, alternate voice lines, changed map sections and unfinished assets. These differences show the game while its cooperative horror formula was still being assembled.

Counter-Strike: Global Offensive

The archive reaches into the early development of Counter-Strike: Global Offensive. One of the earliest builds is described as being much closer to a modified version of Counter-Strike: Source than to the final CS:GO identity. Researchers have pointed to an earlier interface style, prototype maps and unfinished or cut weapon and equipment material.

Those discoveries are useful precisely because they document a transition rather than a finished product: CS:GO did not appear fully formed, and the archive exposes part of the process that connected Source to the game that eventually launched.

The “Weaponizer” and the Episode 3 question

The most sensational claim concerns Half-Life 2: Episode Three. A model or asset known as the “Weaponizer” has been linked by researchers to early Episode 3 material, and a related directory name has been discussed in connection with an older leak.

This is not proof that a complete Episode 3 build has surfaced. At the time of writing, the evidence supports a narrower conclusion: the archive contains an asset or group of assets that may be associated with Episode 3-era work. It does not establish that the project was close to release, that a playable build is present, or that the files reveal anything about Half-Life 3.

What kinds of games and files are included?

The name “Valve leak” is misleading if it suggests that the archive contains only Valve games. The old Steam infrastructure hosted content from many publishers, and the material being catalogued appears to reflect that.

In practical terms, the archive is closer to a historical snapshot of Steam’s distribution layer than to a neatly organised library. It can contain several versions of the same product, duplicates, placeholders, development tools and media files next to the material people are most interested in. The community still has to separate those categories and verify what each package actually contains.

What has not been proven

Several claims are circulating faster than the evidence supports. The current research does not establish that:

• a complete or playable Half-Life 2: Episode Three build is included;
• Half-Life 3 files have been found;
• F-Stop exists in a complete, finished or easily playable form;
• every one of the roughly 12 TB is unique game content;
• the archive is a clean dump from one event rather than a collection assembled over time;
• the old endpoint was accessed through a conventional hack or intrusion.

It is also too early to call this automatically a “source-code leak”. The material described by the strongest reporting is primarily depot data: builds, assets, packages and associated files from a legacy content system. Some packages may include development tools or other internal material, but that is different from confirming that Valve’s source repositories were exposed.

Why this matters

For players, the appeal is obvious: these files can show how familiar games changed before release. For preservation researchers, the value is broader. Old betas often preserve cut mechanics, temporary art, forgotten interfaces and the technical assumptions of a particular moment in development.

The archive also illustrates a less glamorous side of digital preservation. A platform can move on while its old distribution layer remains poorly understood, duplicated or publicly reachable. Once a large historical collection escapes into the open, it becomes difficult to establish which version is authentic, which files are duplicates and which material was already known.

That is why this story should be followed as an archival investigation, not treated as a promise of secret Valve games suddenly becoming available. The responsible editorial approach is to document verified findings without reproducing acquisition paths or directing readers to the files themselves.

What happens next?

The next phase is likely to be slow cataloguing: matching depots to games, comparing timestamps and builds, identifying duplicates, checking whether an executable actually runs and separating Valve material from third-party content. New screenshots and names will probably appear before the full inventory is understood.

Valve had not issued a public statement on the archive at the time of this research, August 31, 2026. Until the company responds or the community verifies more of the collection, the safest summary is also the most accurate one: a large amount of old Steam content appears to have become accessible through a legacy public endpoint, and it includes valuable development history from Valve and many other publishers. It is not, by itself, proof of a playable Episode Three or Half-Life 3.