HTTP/2 Server Push was originally designed to allow servers to anticipate client needs and push subresources (like CSS or JavaScript) before the client requested them. In practice, it suffered from fatal architectural flaws: (1) Cache Blindness: The server has no knowledge of what is already in the browser's local cache, resulting in repeatedly pushing megabytes of duplicate data over mobile connections. (2) Bandwidth Contention: Pushed assets competed directly with the critical HTML document itself, actually delaying First Contentful Paint. (3) Service Worker Complexity: Pushed streams frequently raced against Service Worker cache intercepts, causing corrupted states. Google Chrome disabled Server Push by default in Chrome 106, and the IETF shifted to 103 Early Hints.

RFC 8297 defines an informational HTTP status code (103 Early Hints) sent BEFORE the final 200 OK response. When a client requests a page that requires a slow backend database query (e.g. 300 ms TTFB), the edge reverse proxy or server immediately emits a 103 response containing Link: ; rel=preload; as=style. The browser receives this hint in milliseconds, checks its own local HTTP cache, and ONLY initiates network fetches for resources it genuinely needs. The browser has full control over request prioritization, eliminating cache waste and bandwidth contention.

Yes. The HTTP/1.1 (RFC 9112), HTTP/2 (RFC 9113), and HTTP/3 (RFC 9114) specifications explicitly permit any number of 1xx Informational responses prior to the final 2xx-5xx response. A CDN edge might emit a 103 Early Hints containing static assets from edge cache within 10 ms, while the origin server might later emit another 103 containing personalized dynamic API endpoints, followed finally by the 200 OK document when server-side rendering finishes.

Edge CDNs use machine learning or static Link header inspection to generate 103 Early Hints at edge points of presence (PoPs). When an incoming request reaches Cloudflare or Fastly, the edge immediately returns a 103 response containing preloads learned from previous visits, while simultaneously forwarding the request to the customer's origin server. By overlapping origin latency (e.g. 200-500ms) with client asset preloading, 103 Early Hints routinely reduces Largest Contentful Paint (LCP) by 20% to 35% on mobile 4G/5G connections.