Perplexity Built on Bing Infrastructure

This showed that the fastest way to build an AI search product was to sit on top of someone else’s search stack, not replace it. Perplexity could focus on the part users actually felt, turning a messy page of links into a short cited answer, while Bing handled the expensive plumbing of crawling the web, keeping pages fresh, ranking results, and returning relevant documents for the model to read.1 2 3 4 5

• A web crawler and index are hard to fake. Microsoft describes Bing based search products as combining a fresh web index with ranking and answer systems, and its search APIs expose those results to developers. Perplexity used that existing layer instead of spending years building web coverage from scratch.2 3 4
• That choice made Perplexity more of a product company than a core infrastructure company in its first phase. The differentiation was the workflow, ask a long question, fetch live pages, compress them into one answer, and show citations, rather than owning the raw retrieval layer itself.1 5
• The tradeoff is that dependency buys speed early, but caps defensibility later. As more companies get access to similar models and search feeds, advantage shifts toward user habit, interface, vertical workflows, and eventually proprietary retrieval or agent behavior, which is why newer AI search players now market their own search APIs and indexes.5 6 7

The next step in this market is moving from answer formatting to owning more of the stack. Perplexity’s path points toward deeper control over retrieval, browsing, and task completion, because once the same models and search backends are widely available, the winners are the products that turn search into repeated work, not one off answers.5 6 7