When Netflix started streaming movies and YouTube made it trivial to upload video from a laptop or phone, the skepticism was familiar: existing solutions worked fine, bandwidth was limited and people were not going to shift all their viewing, from movies and short clips, onto the open internet. Scale proved the skeptics wrong. By pairing large content catalogs and user-generated video with CDNs (content delivery networks), more accessible and lower-cost editing tools, adaptive bitrate streaming, and always-connected mobile devices, internet video took the video store shelf and made it infinitely long. A Blockbuster store could stock a few thousand DVDs. Streaming platforms and UGC (user generated content) services together could surface millions of titles and clips on-demand.
3D is entering the same cycle, but faster. For decades, creating compelling 3D required highly-trained teams working in digital content creation tools. That is changing. Generative AI now produces usable 3D assets from text prompts or a handful of images, while capture pipelines are turning photos and video into dense radiance fields and Gaussian splats that can reach production quality. The barrier to creation is dropping, and the volume of content is rising.
Distribution has not kept pace. Today, most realtime 3D on the web runs through polygonal formats such as glTF. These formats do not inherently require simplification, but in practice, developers aggressively decimate meshes, compress textures and standardize lighting to fit within device and bandwidth budgets. The result is a look many brands recognize: physically-based materials that still read as plasticky, generic reflections, and hero products that do not match their high-end renders. The alternative, pixel streaming from cloud GPUs, works well for demos and small events, but breaks down economically at internet scale. At 100,000 active users, you are paying for a stadium of GPUs (assuming you can secure them in the first place) and an enormous amount of video bandwidth, which is not a sustainable way to power everyday shopping or entertainment.
The path forward is to stream compact field representations directly to devices and render locally. Instead of pushing a fixed sequence of pixels, you distribute a compressed scene that can be cached at the cloud edge while phones, headsets and laptops sample it from arbitrary viewpoints in realtime. At bitrates that are competitive with high-quality video, you trade a single camera path for fully interactive viewpoints without renting a cloud GPU for every viewer.
Driving down the barrier to creating 3D assets, from expert-only DCC workflows to prompts, captures and small teams, only matters if distribution scales just as efficiently. When capture-based and AI-generated scenes can be streamed as compact volumetric representations at video-like costs, 3D stops being a special-case asset and starts to behave like a native medium across the web, mobile and XR devices. The same pipelines that feed product detail pages can power immersive try-on, spatial storytelling and ambient experiences that live in the world rather than on a flat screen.
2026 will not finish this stack, but it will reset expectations. As radiance fields, splats and AI-assisted asset generation become normal, teams will stop asking whether today's web 3D and pixel streaming are "good enough" and start treating volumetric streaming as a first-class medium. The networks stay the same. What we will send over them will change.
Will McDonald is the Chief Product Officer at Miris (www.miris.com), the developer of a SDK that streams high-quality 3D to any device instantly.