HTTP/3, defined in RFC 9114 (June 2022), is HTTP with its transport replaced. The request/response model, headers and methods are nearly identical to HTTP/2, but the bytes run over QUIC streams instead of TCP. The key gain for streaming is per-stream loss recovery: when a client multiplexes many segment fetches on one connection (typical for low-latency HLS or DASH), packet loss on one stream no longer stalls the others. That alone can cut rebuffer ratios over lossy networks by single-digit percentages.
For streaming pipelines HTTP/3 also brings two operational wins. 0-RTT resumption means a returning client can send its first segment request in the same flight as the connection handshake, shaving 100–300 ms off startup on warm connections. And connection migration — QUIC's ability to keep a session alive when the client changes IP — keeps mobile viewers connected through Wi-Fi-to-cellular handoffs without resyncing the player.
The catch is consistency. Not every middlebox passes UDP cleanly, so HTTP/3 is always deployed alongside HTTP/2 fallback. CDNs negotiate HTTP/3 via the Alt-Svc header, the player upgrades silently, and on networks where QUIC fails the second connection falls back to TCP. By 2026 most large OTT services see 60–80 % of byte traffic over HTTP/3 with the rest on HTTP/2, with no observable user-facing difference except slightly better tail latencies.
