The Shannon limit is the theoretical floor on how compactly any data can be compressed losslessly. Named after Claude Shannon, who proved it mathematically in 1948, the limit says: a signal with entropy of N bits per symbol cannot be reliably compressed to fewer than N bits per symbol without losing information. It's not a guideline or a heuristic — it's a hard mathematical wall.
For video, this matters because it puts a concrete ceiling on what any future codec can ever achieve. If a video signal has an inherent entropy of, say, 8 megabits per second of true information, no lossless codec will ever encode it in less than 8 Mbps. Modern lossless codecs (FFV1, ProRes 4444 XQ) come within a few percent of the Shannon limit for their content — they're nearly mathematically optimal. Further compression beyond the limit is possible only by accepting some loss of information, which is exactly what lossy codecs (H.264, HEVC, AV1) do.
For a product team, the Shannon limit explains a counterintuitive industry pattern. Lossless compression has barely improved in 30 years — there's not much room left because we're near the limit. Lossy compression keeps improving dramatically each codec generation (50 % savings H.264 → HEVC, 30 % HEVC → AV1, another 30 % AV1 → AV2) because the "limit" for lossy compression depends on what humans will notice rather than the raw information content, and that perceptual frontier keeps moving as research improves understanding of vision. The takeaway: anyone promising endless future codec improvements is essentially betting on perceptual research, not on breaking math.
