Will Bryk’s Exa has crossed $100 million in annualised revenue by building a search engine for machines rather than people. Its bigger bet is that, by 2030, AI agents will conduct 1,000 times as many searches as humans do today. If that prediction comes anywhere close to reality, the economics of internet search could change dramatically.
For decades, internet search has been built around a simple transaction: a person enters a query, receives a list of links and potentially clicks an advertisement. Google became the dominant force in that system because it could connect human attention to commercial intent.
Artificial intelligence agents threaten to change that arrangement.
An AI agent does not necessarily search once, scan a page and stop. It may conduct dozens of searches to compare products, investigate a company, write software, analyse financial information or complete a research task. It can retrieve documents, examine their contents, refine its questions and repeat the process without requiring a human to supervise every step.
That creates a new market for search infrastructure: not just finding information for people, but continuously supplying information to machines.
Exa, led by chief executive Will Bryk, wants to become one of the companies powering that market.
From almost nothing to $100 million in annualised revenue
In an announcement in October 2026, Bryk said Exa had surpassed $100 million in annualised revenue, was growing at approximately ten times a year for three consecutive years and was handling two billion search requests per month from more than 200,000 teams.
These are company-reported figures, and annualised revenue is not the same as audited revenue earned over a completed financial year. Nevertheless, the reported growth illustrates how quickly demand for AI-oriented search infrastructure is developing.
Exa previously announced a $250 million Series C funding round in May 2026 at a reported valuation of $2.2 billion. Its ambition is to build a search engine specifically designed for AI systems, rather than simply adapting a conventional search interface for a new audience.
The distinction is fundamental.
Traditional search engines primarily optimise results for people who scan headlines, follow links and decide what to read. AI agents need information that can be retrieved, processed and passed into another computational step. They may also need semantic search, access to full-page content and results that can be incorporated directly into an automated workflow.
Exa says it has spent five years building its own index, covering hundreds of billions of web documents alongside proprietary datasets.
That is the company's strategic asset. Its business is not merely selling search requests; it is attempting to control an important layer between the internet's information and the AI systems that consume it.
The $35 trillion thought experiment
Bryk's most provocative prediction is that AI agents could conduct 1,000 times as many searches as humans by 2030.
To understand the economic implications, consider the calculation.
If human search activity is approximately five trillion searches annually, multiplying that volume by 1,000 produces five quadrillion searches a year.
At a price of $7 per 1,000 requests—the published rate for Exa's standard search offering—the arithmetic produces a hypothetical annual bill of $35 trillion.
That would exceed the United States' nominal gross domestic product of approximately $30.8 trillion in 2025.
The comparison is striking, but it must be interpreted correctly. The $35 trillion figure is not a forecast that the world will actually spend that amount on AI search. It is the result of multiplying an assumed future volume by a present-day list price.
At that scale, the price itself would almost certainly change. Providers could negotiate volume discounts, build more efficient infrastructure, cache repeated results and develop cheaper retrieval methods. AI systems might also reduce unnecessary searches, reuse existing information or rely on local datasets instead of repeatedly querying the public web.
The calculation is useful for a different reason: it illustrates the enormous economic pressure that would emerge if automated search grew by several orders of magnitude.
Search becomes a cost problem, not just a ranking problem
Suppose five quadrillion searches really were conducted in a year.
To price that volume at $35 billion rather than $35 trillion, the cost would need to fall from $7 per 1,000 searches to approximately $7 per million searches—a thousandfold reduction.
That would fundamentally change the competitive landscape.
At current prices, search infrastructure can generate meaningful revenue from every request. At dramatically lower prices, companies would need to serve vast volumes while minimising the cost of crawling, indexing, storage, computing and retrieving relevant information.
The winners would not necessarily be the companies with the most recognisable consumer brands. They could be those that own efficient indexes, operate highly optimised infrastructure, minimise redundant computation and deliver useful results with fewer retrieval steps.
Search quality would still matter. A cheap system that returns irrelevant information or forces an AI agent to search repeatedly may ultimately cost more than a slightly more expensive system that finds the right information immediately.
The real contest will be about the cost of completing a useful, information-grounded task—not merely the price of one API request.
Why Google’s advertising model faces a different challenge
Google's search business is built around the relationship between information, human attention and advertising. Users search for something, advertisers compete for visibility, and Google monetises a portion of that activity.
AI agents change the interaction.
An agent retrieving information for another system does not necessarily view advertisements, click sponsored links or respond to conventional search-result layouts. It extracts relevant information and passes it to the next stage of its task.
That creates a potential gap between the economic value of a search to an AI agent and the revenue a traditional search engine can earn from the same query.
For Google, the challenge is not simply defending its existing search product. It is deciding how to serve a market in which the customer may be software, the search volume may be enormous and the traditional advertising impression may not exist.
However, it would be premature to conclude that Google cannot monetise agentic search. It can charge for APIs, sell enterprise services, integrate search into AI products and develop other commercial arrangements. Advertising may also persist in some agent experiences, although its format and effectiveness could change.
The more important question is whether Google can offer machine-oriented access at competitive prices without undermining the economics of its established business.
Microsoft's Bing decision exposes the strategic stakes
Microsoft announced that its Bing Search APIs would be retired on August 11, 2025, directing customers towards Grounding with Bing Search within its Azure AI agent services.
The decision matters because an API provides developers with access to search infrastructure without requiring them to adopt an entire vendor's AI platform. Moving towards a more integrated agent offering changes the terms on which developers access that infrastructure.
For companies building independent AI agents, this creates an incentive to find alternative providers with their own indexes, clear pricing and flexible integration options.
Exa is positioning itself to benefit from that demand.
But the market is not a simple contest between an innovative startup and two incumbents determined to protect their advertising revenue. Google and Microsoft have substantial infrastructure, distribution, data and capital. Both can develop new products, adjust pricing and compete for enterprise customers.
The advantage of an independent provider is that it can design its service around the needs of AI developers from the beginning. The disadvantage is that building and maintaining a comprehensive, current and legally usable index is an expensive, continuous undertaking.
Owning the index may become more important than owning the interface
For much of the internet era, consumers interacted with search through a website or an app. Competition focused heavily on ranking, user experience and distribution.
In an agent-driven environment, the interface may become less important than the underlying information infrastructure.
If an AI agent can search several sources automatically, compare the results and return a conclusion, users may never see the search engine that supplied the information. The provider becomes an invisible part of the software stack.
That does not make the provider less important. It may make it more important.
An AI model's capabilities are constrained by the information it can access, the freshness and reliability of that information, and the quality of the retrieval process. A sophisticated reasoning system with poor access to relevant information can still produce incomplete or outdated answers.
Bryk's central argument is that information access remains a bottleneck even as AI models become more capable. Exa's business is built around solving that problem.
The strategic implications extend beyond search companies. Publishers, research institutions, businesses and website operators may face greater automated demand for their content. Questions about crawling permissions, copyright, attribution, data licensing and compensation will become more consequential as machines retrieve information at scale.
A future in which agents conduct quadrillions of searches would require more than faster servers. It would require sustainable rules governing how the world's information is collected, accessed and used.
The risks behind the growth story
Exa's reported revenue growth is impressive, but the company's long-term prospects depend on several variables.
First, the 1,000-times prediction is a hypothesis, not an established trajectory. AI agents may generate substantial additional search demand, but efficiency improvements, caching and more capable models could reduce the number of searches required per task.
Second, search is a competitive market. Exa must compete with established technology companies, specialist search providers, open-source retrieval systems and organisations building their own indexes.
Third, the economics of crawling the web are difficult. Information changes continuously, websites restrict access, and maintaining a useful index requires ongoing investment. Search providers must also navigate legal and commercial restrictions on data collection and reuse.
Finally, two billion requests a month does not automatically translate into durable profitability. Revenue growth must be evaluated alongside infrastructure costs, customer concentration, retention, margins and the expense of maintaining and expanding the index.
These are the questions that will determine whether Exa becomes a foundational infrastructure company or a successful specialist provider in a crowded market.
The real prize is control over machine access to information
The rise of AI agents could create a new layer of internet infrastructure in which search is no longer primarily a destination for human users. It becomes a service consumed continuously by software.
That shift could redistribute economic power across the technology industry. Companies that own useful indexes and deliver reliable information at low cost could gain influence over thousands of downstream AI applications.
Exa's strategy is an early bet on that possibility. Its reported $100 million annualised revenue milestone suggests that the market already has meaningful commercial demand. Its longer-term prediction, however, remains unproven.
The $35 trillion calculation should not be treated as a spending forecast. It is a warning about the mismatch between today's search prices and the potential scale of automated demand.
If AI agents really do search at volumes far beyond human activity, the industry will have to reinvent the economics of information retrieval. Prices will fall, infrastructure will become more important, and control over the web's searchable knowledge may emerge as one of the most valuable positions in the AI economy.
The next battle in artificial intelligence may not be about which model can think best. It may be about which company can give those models the information they need—at a price the world can afford.
Comments
Post a Comment