Skip to content

Dead End: VRML & The First Metaverse

Abstract

In 1994, developers had a vision of a three-dimensional internet, a walkable web of virtual spaces that translated William Gibson’s cyberspace and Neal Stephenson’s Metaverse into code. They had the idea, the language, and the standard. What they did not have: enough bandwidth, enough computing power, and a clear answer to the question of what people were actually supposed to do in this three-dimensional internet.

Snow Crash and the Blueprint

In 1992, Neal Stephenson’s science fiction novel Snow Crash was published. In it, people move through the Metaverse, a persistent virtual world, a three-dimensional internet in which users interact as avatars, trade, and form social hierarchies.

Snow Crash was literature, not a manual. But it described something so precisely and vividly that developers read it as a specification. Everyone working on a 3D internet in 1994 knew Stephenson’s novel.

Tim Berners-Lee had introduced the World Wide Web as a hypertext system in 1991: two-dimensional, pages and links. A group of developers wanted to anticipate the next paradigm: Mark Pesce, Tony Parisi, and others laid the groundwork for a 3D description language for the web at the first WWW Conference (Geneva, 1994).

The result: VRML, the Virtual Reality Modeling Language. Version 1.0 appeared in 1994, designed by Pesce and Parisi with contributions from Paul Strauss (Silicon Graphics). The syntax was derived from Silicon Graphics’ Inventor file format: a declarative description of three-dimensional scenes, with geometry, materials, light sources, and transformations.

The Technology: What VRML Could Do

A VRML file described a 3D scene as text. A browser plugin (or a dedicated VRML browser) rendered the scene and let users navigate it.

#VRML V1.0 ascii
Separator {
  Sphere { radius 1.0 }
  Material { diffuseColor 1 0 0 }
}

VRML 2.0 (1996, later standardized by ISO as VRML97) extended the model with interactivity: sensors, script nodes, animated objects, audio, interpolation. Worlds could react to user interaction. Objects moved. Doors opened as you approached.

Silicon Graphics (SGI), then the maker of the most powerful workstations in the world, was a natural champion. SGI’s VRML browser Cosmo Player was one of the most widely used. The company saw VRML as a way to justify its expensive 3D hardware: broad VRML adoption would make powerful graphics workstations interesting for consumers.

Microsoft briefly integrated VRML support into Internet Explorer. Netscape had a VRML plugin. It looked like a rising standard.

The Ecosystem of Early 3D Worlds

The first virtual communities emerged on a VRML foundation:

WorldsChat (1995), a chat program in a VRML-like 3D environment, operated by Worlds Inc. Users moved through a space station as avatars. Up to a thousand simultaneous users in the early years.

OnLive! Traveler (1996): 3D chat with voice transmission and lip-synced avatars. Remarkable for 1996.

Activeworlds (1995), persistent, user-buildable 3D worlds. An early model of a virtual reality shaped by its users.

These systems had real users. People spent time there. Relationships formed. Virtual land was traded.

Info

The bandwidth math:

A simple VRML scene with a few hundred polygons and textures quickly reached several megabytes. In 1996, most users had 28.8k or 56k modems (28,800 to 56,000 bits per second). At 28.8k, a one-megabyte VRML file took about 5 minutes to download before anything appeared at all. Complex worlds were practically impossible.

High-performance graphics cards were expensive. The 3dfx Voodoo (1996) brought consumer 3D acceleration for 300 dollars, not affordable for everyone. VRML on a Pentium 90 without a graphics card was an experience of 5 frames per second in a 320x240 window.

The internet was not ready for the three-dimensional internet. And the three-dimensional internet was not ready for the hardware that actually stood in people’s homes.

The Failure

VRML97 became an ISO standard. That was the high point, and the beginning of the decline.

The problem was not standardization. The problem was the lack of applications. VRML worlds were impressive as demonstrations. They were impractical for everything else.

What do you do in a 3D chatroom that you cannot also do in a 2D chatroom? VRML offered no convincing answers. Navigation was frustrating, there was no spatial audio, no sense of embodiment, no meaningful reason to be in a 3D space instead of an IRC channel.

Commercial disinterest: Companies tried 3D product catalogs in VRML. Conversion rates were worse than 2D catalogs. Customers wanted to buy products, not navigate 3D spaces.

Technical fragmentation: Different VRML browsers implemented VRML97 inconsistently. A world that looked correct in Cosmo Player was broken in other browsers. The web had HTML compatibility problems; VRML had them squared.

X3D (2001) was VRML’s successor, built on XML. It standardized much of what VRML97 had left open. But the industry had stopped caring.

Silicon Graphics collapsed in 2009 (after two bankruptcies). The hardware justification for elaborate 3D worlds disappeared with it.

What Meta Ignored

In 2021, Mark Zuckerberg announced the renaming of Facebook Inc. to Meta and a multi-billion-dollar bet on the metaverse. Horizon Worlds, Horizon Venues, VR meetings in Oculus headsets.

The fundamental problems of 1996 are partly solved in 2024, partly structurally the same:

Bandwidth: Solved. Fiber and 5G make loading 3D worlds trivial.

Computing power: Largely solved. Modern GPUs render complex 3D worlds in real time.

Input devices and comfort: Unsolved. VR headsets are heavier, bulkier, and more expensive than smartphones. Hours of use remain uncomfortable. The Meta Quest 3 costs 499 dollars, still not a mass-market device.

Killer application: Still unsolved. Horizon Worlds had under 200,000 monthly active users in mid-2022, against billions of invested dollars. The question “what do you actually do there?” remained unanswered.

Metaverse critics did not cite VRML; they barely knew it. But the failure followed the same pattern: impressive technology demonstrations, no everyday relevance, users leaving the medium after a short exploration.

📚 Sources