Toward an AI-Native Internet: Rethinking the Web Architecture for Semantic Retrieval

TL;DR

The paper identifies the inefficiencies of a document-centric Web for AI-driven semantic retrieval and proposes an AI-Native Internet where servers expose semantically chunked information and a Web-native semantic resolver enables Internet-wide discovery of relevant sources before fetching fine-grained chunks. It details design goals and three architectural components—Query Processing, Web Sources, and Semantic Resolver—and provides motivational experiments showing substantial data-transfer reductions (e.g., ~74%–87%) with comparable accuracy, in both centralized and decentralized settings. The work highlights open challenges in governance, provenance, incentives, and deployment, arguing for embedding semantic retrieval as a foundational Internet primitive rather than a retrofit atop HTML. If realized, this architecture could enable scalable, bandwidth-efficient AI applications and agents with standardized semantic primitives and open, auditable discovery across the Web.

Abstract

The rise of Generative AI Search is fundamentally transforming how users and intelligent systems interact with the Internet. LLMs increasingly act as intermediaries between humans and web information. Yet the web remains optimized for human browsing rather than AI-driven semantic retrieval, resulting in wasted network bandwidth, lower information quality, and unnecessary complexity for developers. We introduce the concept of an AI-Native Internet, a web architecture in which servers expose semantically relevant information chunks rather than full documents, supported by a Web-native semantic resolver that allows AI applications to discover relevant information sources before retrieving fine-grained chunks. Through motivational experiments, we quantify the inefficiencies of current HTML-based retrieval, and outline architectural directions and open challenges for evolving today's document-centric web into an AI-oriented substrate that better supports semantic access to web content.

Figures (4)

• Figure 1: Basic architecture for AI-Native Internet
• Figure 2: Queries (out of 200) that were rephrased, decomposed, or left unchanged by GPT-5-mini.
• Figure 3: Number of questions with relevant context retrieved with centralized vector datastore
• Figure 4: Questions with relevant context present in the decentralized model