Video Infrastructure

8/1288 → 15/802; +1.25 percentage points of formation share.

The most telling detail here is the split between generation and orchestration: companies like VideoGen, Argil, and Focal are still racing to produce raw video from a prompt, while a second wave, Latted, Palmier, and Overlap, has stopped competing on generation entirely and instead builds the editing and agent layer that sits on top of whatever model already exists, treating the underlying video model as a commodity to be plugged in rather than a moat to be built. That divide shows up again in Cloudglue and Dragoneye, which skip generation altogether and focus on making existing video legible to AI, either as searchable context or as detection models trained in minutes, a sign that the harder unsolved problem is comprehension, not creation. It is worth noting that this formation is accelerating at 3.01x while the broader "Video" category it sits inside is growing at only 0.50x, meaning the energy is concentrated specifically in these tooling and infrastructure layers rather than in video generally. Sync's pitch to make video "as fluid and editable as a word document" captures the shared ambition across nearly every company here: video is being re-architected as an editable, queryable data type instead of a finished file. This is happening now because 2024 was the year text-to-video models like Sora and Runway Gen-3 became good enough to produce usable footage, which immediately exposed the next bottleneck, stitching, editing, syncing, and understanding that raw output, and this batch of companies is built almost entirely around solving that second-order problem rather than the generation problem itself.

No comparable outside activity is connected yet.

These share a word with “Video Infrastructure” but are tracked as separate categories. The real overlap below is the actual share of Video Infrastructure’s companies also found in that category — not a guess.