No big moat in AI search

The market is likely to reward distribution and customer fit more than raw search technology. In practice, buyers like Ecosia already treat providers such as Exa, Parallel, and Tavily as replaceable enough that pricing, latency tuning, and hands on support matter more than any unique algorithm. That is why Ecosia kept a switchable architecture even while spending about $300,000 per month on Exa and routing roughly 500,000 daily queries through it.

• The hardest part to copy is not basic retrieval, it is fitting into a customer workflow. Ecosia chose Exa over Parallel even though result quality looked similar, because Exa customized pricing, helped tune latency, and acted like an engineering partner during rollout.
  1 sacra 4 sacra
• The category is crowded across different product shapes. Exa is building an AI native index and search API. Tavily leans more asset light and aggregates multiple sources. Perplexity started as an answer engine and then moved toward consumer interface ownership. That mix makes technical features easy to compare and easier to substitute.
  2 sacra 5 sacra 6 sacra
• The real wedge may move up stack. If search with citations becomes standard inside model labs and browsers, independent providers will need to win on domain specific data, enterprise controls, or workflow level products that save customers from stitching together search, extraction, and reasoning themselves.
  3 sacra 5 sacra 6 sacra

Over the next few years, AI search infrastructure is likely to look more like cloud infrastructure than a consumer monopoly. Core retrieval will get cheaper and more interchangeable, while value concentrates in who owns the user entry point, who bundles search into a larger product, and who turns generic web results into workflows for coding, research, and enterprise knowledge work.