Der Aufstieg von Suchmaschinen, die speziell für KI entwickelt wurden, bringt eine neue Welle von Veränderungen im Paradigma der Informationsgewinnung mit sich.

AI is creating a brand - new paradigm for information acquisition. Consumer - facing AI search applications such as Perplexity and ChatGPT have broken through the "ten blue links" search engine paradigm defined by Google in the early days of the Internet, enabling users to directly obtain high - quality answers generated after an AI search.

For more professional users, research - oriented Agents like OpenAI DeepResearch can conduct multiple rounds of in - depth searches and ultimately output a detailed, reliable, and professional report.

Behind AI search applications, there is also a paradigm shift, that is, search engines specifically designed for AI, which are at the forefront. Their goal is to provide a comprehensive, unbiased, and high - quality understanding of the world, enabling AI to search for things that were previously unfindable.

In the next few years, the search query volume of AI will far exceed that of humans (Source: Exa)

Recently, two companies of this type have successively received large - scale financing. Among them, Exa received an $85 million Series B financing led by Benchmark, with a valuation reaching $700 million. Lightspeed Venture Partners, Y Combinator, and NVIDIA continued to follow - on invest.

You.com received a $100 million Series C financing led by Cox Enterprises, with a valuation reaching $1.5 billion.

Search engines built for AI are leading a new round of transformation in the information acquisition paradigm

Each paradigm shift in the way of information acquisition and dissemination has brought about a significant improvement in the efficiency of human society, which in turn promotes a substantial increase in productivity. This is true for the invention of writing, the invention of paper, and the innovation of printing technology.

The invention and popularization of the Internet are the characteristics of human society entering the information age. And the application that truly improves the efficiency of human information acquisition in the Internet era is the search engine.

In the AI era, how can we further improve human efficiency, and how can the information acquisition paradigm change? The answer given by Exa and You.com is to build search engines for AI. These search engines should be faster and more accurate than the previous ones designed for humans.

As Will Bryk, the CEO of Exa, said: "Search engines like Google are designed for humans, not AI. Google is optimized for human click - through behavior, while AI needs a search engine that is powerful and precise enough to retrieve thousands of search results containing the best information. AI only pursues the highest - quality knowledge, not SEO junk information or advertisements. As we all know, garbage in, garbage out."

Richard Socher, the founder of You.com, also said: "If Bing can't find something, then SearchGPT (which relies on Bing), commonly used by AI search tools, is also helpless. Therefore, today's so - called 'AI search' mainly saves time. It is not a truly more powerful search engine."

These search engines designed for AI are used by AI Agents, and AI Agents can run in parallel. One person can manage dozens of Agents simultaneously. When the quality and quantity of searches are improved, the efficiency of information acquisition will naturally be significantly improved.

In essence, both Exa and You.com are AI - native search infrastructures. Richard Socher, the founder of You.com, wrote: "AI models are very powerful today, but they need the right infrastructure to fully realize their potential. It's like building self - driving cars before we had highways. What we at You.com are building are these highways."

However, Exa and You.com have taken different paths in terms of implementation methods, product forms, and business models.

Exa: A "pure knowledge source" built for AI

Exa was jointly founded by a pair of Harvard alumni, Will Bryk and Jeff Wang. It is an early - stage incubation project of YC.

Exa's founders: Will Bryk and Jeff Wang (Source: Exa)

The core technology of Exa is a self - developed neural network search architecture. Founder Will Bryk said: "Transformer models are usually used to predict the next word. We train our search engine to predict the next link." Its technology is based on "vector databases and embedding technology" rather than traditional keyword matching.

Its main product is a search API specifically designed for AI, positioned as the "data layer of the AI world". Specifically, it includes: the Exa Search API for quick responses; Websets for comprehensive searches; and a research API that can perform multi - step research and return structured insights.

Exa's research API achieved a high score of 94.9% in the SimpleQA benchmark test (Source: Exa)

In terms of product performance, Exa has built what it claims to be the world's fastest search API, with a response time of less than 450 milliseconds. Their Research API endpoint achieved a high score of 94.9% in the SimpleQA benchmark test.

In a nutshell, it is positioned as a "pure knowledge source" built for AI.

In terms of business model, Exa provides customers with a pay - as - you - go search service, abandoning advertisements and focusing only on quality. Currently, Exa has leading users such as Databricks, Cursor, and Notion.

Databricks uses Exa's technology to find large - scale training datasets for its own model training; Cursor uses Exa to retrieve technical documents; and Notion uses Exa to support its Notion AI function.

You.com: Building the highways of the Agent era

You.com was jointly founded by Richard Socher, the former chief scientist of Salesforce, and Bryan McCann, the former head of AI research at Salesforce. During his time at Salesforce, Socher helped build the company's large - scale Einstein AI platform.

You.com's founding team (Source: You.com)

You.com's product was not in its current form at first. In 2021, its initial version was still a traditional search engine, but it made significant changes to the UI. Its search results are presented in categories according to applications, with fewer advertisements, improving the user experience.

After the emergence of truly usable large - language models such as GPT - 3, You.com launched functions such as YouCode and YouWrite, which can generate or write papers and blog articles based on search queries or prompts.

At this time, it had not yet found its optimal form.

It wasn't until it explored its optimal positioning, building a search engine for AI rather than humans, and launching the Research Agent (in 2024) and its real flagship product ARI (Advanced Research & Insights), that its financing started to go smoothly. It received $50 million and $100 million in Series B and Series C financing in 2024 and 2025 respectively.

Now, the core technology of You.com is its "model - agnostic AI operating system", which can make any large - language model more accurate and trustworthy.

In terms of products, it has a composable API infrastructure for developers, including a series of capabilities such as web and news search, in - depth research, image search, and private RAG systems.

For enterprises, it provides end - to - end complete solutions, which are deeply customized and can directly solve business problems as complete AI applications.

For example, its core product ARI (Advanced Research & Insights) is a cutting - edge research - oriented Agent designed for enterprises. It can instantly generate beautiful PDF reports containing visual charts, annotated comparative analyses, and interactive data graphs. Its research speed is three times faster than ChatGPT, with an average response time of only three minutes, and it can handle more than 400 information sources simultaneously.

You.com's product ARI has a 76% win - rate compared to OpenAI's products (Source: You.com)

In summary, You.com has built the highways of the Agent era: a platform that can integrate multi - source data, select the best models for tasks, and deliver accurate and reliable results at an enterprise - level scale. Compared with Exa, which focuses more on being a tool behind developers, You.com not only provides tools but also directly provides products for enterprises.

Currently, You.com's customers include leading companies in the AI era such as DuckDuckGo, Windsurf, and Harvey, and it processes more than 1 billion queries per month.

In the AI era, entrepreneurs have disruptive entrepreneurial opportunities in the search field

Context is king. For enterprise customers, a simple general - purpose model can hardly bring direct benefits, especially when Agents become the main product form. For an Agent to automatically bring benefits to an enterprise, it needs high intelligence and high accuracy. Intelligence can be broken down into a deep understanding of the enterprise's business, while high accuracy solves problems similar to the "last - mile" problem in autonomous driving.

Assume that an Agent has an accuracy rate of only 95% at each step. Then, any workflow consisting of 30 steps cannot run reliably. You must increase the accuracy rate from 95% to 99.9%.

To make an Agent closer to the business, a large amount of high - quality data is required. These data refer to data that is highly relevant to the enterprise's business. This data can be collected manually to post - train the general - purpose model, turning it into a fine - tuned vertical model.

On the other hand, if an Agent has super - strong search capabilities and can search for a sufficient amount of high - quality data in real - time, both inside and outside the enterprise, to provide context for the Agent/model, then there is a high probability of achieving both high intelligence and high accuracy. This ability is very important for AI applications to penetrate into the deep waters of various vertical businesses.

The arrival of the AI era has brought disruptive entrepreneurial opportunities in the search field. Whether it is building AI search applications for humans or search tools for AI, there are still huge opportunities. Although there has already been a unicorn like Perplexity valued at tens of billions of dollars, this field is still in its early stages in terms of product experience and market penetration.

Looking further, in 2025, the general - purpose models launched by major model companies have started to become Agent - enabled. The ability to call tools has become an important direction for model development. Therefore, building the types of tools most commonly used by models/Agents when calling tools is also a good entrepreneurial direction.

In addition, for entrepreneurs, whether building search applications or other AI tools, they can either choose to build tools with extreme performance like Exa, using technology and data flywheels to build a moat, or they can build both infrastructure and self - developed applications like You.com, deeply integrating with the customer's workflow and using customer stickiness and switching costs as a moat. These choices can be made based on the characteristics of their own teams and business directions.

This article is from the WeChat public account "Alpha Startups" (ID: alphastartups). The author is the one who discovers extraordinary entrepreneurs. It is published by 36Kr with permission.