Brave: the company that monetizes artificial intelligence at everyone else's expense

Thanks to its independent search index, Brave Software has become a key supplier of web data for most leading language models. After the shutdown of Bing's public API and Google's lawsuit against SerpApi, Brave has remained practically the only commercially available source of up-to-date web data on a global scale, achieving annual revenue of over 100 million dollars. The article analyzes the paradox of a company that, at a fraction of the cost, profits from those investing billions in artificial intelligence, and warns of the risks of the entire industry's dependence on a single supplier.

OpenAI reported a net loss of over five billion dollars for 2024. Anthropic projected a cash burn of three billion in 2025. Google is investing tens of billions in infrastructure for Gemini. The entire conversational artificial intelligence sector operates on the assumption that profitability will come "sometime later" — while a relatively small San Francisco firm is quietly making money off them right now.

Brave Software, the maker of the Brave browser and the independent Brave Search engine, surpassed an annualized revenue run rate of 100 million dollars in the first quarter of 2025. That is not, in itself, a staggering figure. What is staggering is where the money comes from: from the very AI companies burning billions on running and training models. Brave sells them something they cannot function without — access to up-to-date data from the web.

This article analyzes how Brave became an indispensable infrastructure supplier for most leading language models, why no one else did it, and what it means for the future of AI-assisted search.

Across the entire Western world there are only three search indexes at global scale: Google, Bing, and Brave Search. This claim is fundamental and needs to be explained.

A search index is a database of web pages that a search engine has built by crawling the internet itself. Most of the "search engines" an ordinary user knows — DuckDuckGo, Yahoo, Ecosia — do not own an index. They use results from Google or Bing through their APIs. They are essentially someone else's search results in different packaging.

Brave Search operates its own index of more than 35 billion indexed pages, with tens of millions of newly crawled pages per day. Brave owns and operates the entire technology stack — from the crawler to the API endpoint. Its origins trace back to the Munich-based firm Cliqz, founded in 2008 as 10betterpages GmbH. But the company only began working on building a private alternative to Google — its own search index and a privacy-focused browser — from 2013, under majority owner Hubert Burda Media. After Cliqz ceased operations in April 2020, the internal Tailcat project was created, which Brave acquired in March 2021.

Why does this matter? Because language models need up-to-date data from the web. Without it, they answer based solely on training data that may be months or years old. Any AI assistant that wants to answer the question "what's the weather today" or "who won yesterday's match" needs to search the web in real time. And for that it needs a search API.

In 2025, two things happened that fundamentally changed the market dynamics.

First, on May 12, 2025, Microsoft announced the shutdown of the public Bing Search API, effective August 11, 2025. The decision was final — all existing instances were completely decommissioned, and new registrations were halted. As a replacement, Microsoft offered the "Grounding with Bing Search" service within Azure AI Agents — but it returns only AI-processed summaries, not raw search results. For developers who needed structured data from the web, it was not an equivalent substitute.

The context of this decision is important. In previous years, Microsoft had repeatedly raised the prices of the Bing API — in 2023 by 257 to 900 percent depending on the tier, and for some add-on services by up to 1,000 percent. Developers who absorbed these increases now faced a complete shutdown of the service. The Register noted that Microsoft was steering customers toward its own AI products as an alternative — that is, toward more expensive, less flexible solutions within the Azure ecosystem.

Second, on December 19, 2025, Google sued the Texas-based firm SerpApi, which specializes in the automated scraping of search results. The lawsuit alleges that SerpApi circumvented the SearchGuard security system, which Google deployed in January 2025, and generated hundreds of millions of fake queries per day. In the lawsuit, Google labeled SerpApi's business model "parasitic" and sought an injunction against further activity as well as damages under the DMCA (Digital Millennium Copyright Act). Statutory damages ranging from 200 to 2,500 dollars per individual violation would, at hundreds of millions of daily queries, reach astronomical sums. In February 2026, SerpApi filed a motion to dismiss the lawsuit; the case has not yet been decided.

For the AI industry, both events had the same practical consequence: legal and reliable access to real-time web data narrowed dramatically. From a market where three paths existed (Google's API via intermediaries, Bing's API, Brave's API), de facto only one full-fledged option remained.

The Brave Search API found itself in an exceptional position: the only commercially available API at global-web scale, independent of both Google and Microsoft.

The numbers confirm it. According to Brave itself, its search API now supplies data to most of the world's ten largest language models. For some of them, Brave is the only search index backing their answers. As of February 2026, the API has over 200,000 registered developers.

Brave names specific customers only in part. From publicly available sources we know of the following: Mistral AI, Snowflake (integration into Cortex Code and Cortex Agents, announced at the BUILD London conference on February 3, 2026), Cursor, Cline, Windsurf (AI developer tools), Kagi and Swisscows (independent search engines), Chegg and Turnitin (educational platforms). Anthropic (maker of Claude) is demonstrably among the customers — Claude uses the Brave search API for answers requiring up-to-date information — yet Brave does not explicitly name it in its materials. The customer list ranges from startups to large enterprises.

On February 12, 2026, Brave introduced a new context API for language models (LLM Context API) — an endpoint designed specifically to feed large language models. Instead of classic links and snippets, it returns structured data optimized for insertion into a model's context window: cleaned text, preserved structured schema (JSON-LD), and tables with row-level precision. Response time under 600 milliseconds at the 90th percentile. Pricing: five dollars per thousand requests for search, four dollars per thousand searches plus five dollars per million tokens for the answer endpoint.

Brave bolstered the product's credibility by obtaining SOC 2 Type II security certification (October 2025) and by offering a zero-data-retention mode — enterprise customers can ensure that no queries are retained. This is a key differentiator for firms in regulated industries and for AI companies dealing with compliance with the GDPR and other regulations.

Brave operates an unusually diversified business model for a company of its size. Revenue comes from four main sources.

Brave Ads forms the backbone of revenue. It is a private advertising platform that targets ads based only on the search query, country, and device type — without personal profiles, without tracking cookies, without tracking. The BAT (Basic Attention Token) model lets users who opt in to viewing ads earn a share of advertising revenue in BAT tokens. Ads are currently not shown in AI-generated answers or in the Leo chat assistant — Josep M. Pujol, head of search at Brave, conceded in an interview with Digiday that ads "could potentially be placed in the AI answer area," but they are not yet.

The Search API is the fastest-growing segment. Brave does not disclose exact revenue, but the sharp rise in adoption over the course of 2025 and the expansion of the customer base to the enterprise level suggest the API may generate tens of millions of dollars per year. Availability on the AWS Marketplace since July 2025 and integration into Snowflake further broaden the distribution channel.

Leo Premium, at 14.99 USD per month, offers access to more powerful AI models and higher limits. The free tier offers open models. Brave does not publish subscriber counts, but given that the VPN service is their "most popular premium product" with "nearly 100,000" users, it can be estimated that Leo Premium has fewer subscribers — probably in the tens of thousands.

Brave Search Premium, at three dollars per month, offers ad-free search.

Total revenue exceeded 100 million dollars per year (Q1 2025 annualized) at 107 million monthly active browser users (February 2026) and 1.6 billion search queries per month. Brave generates over 22 million AI answers per day in its search engine.

A paradoxical situation has emerged that has no exact parallel in the technology industry.

AI companies invest tens of billions of dollars in training and running models, but they cannot function without up-to-date data from the web. That data is supplied to them by a company with annual revenue under 200 million dollars and an index it built at a fraction of the cost. Brave did not have to invest in billion-dollar compute clusters, did not have to train foundation models, did not have to deal with AI safety. It built infrastructure for crawling the web and an index — and now sells it to those who invested the billions.

An analogy presents itself from another industry: during the 19th-century gold rush in California, the most money was made not by those who panned for gold, but by those who sold shovels and tents. Levi Strauss did not come to San Francisco to look for gold — he came to sell trousers to those who were looking for it. Brave does not sell artificial intelligence — it sells data to those who are building it.

The asymmetry intensifies when we consider who the customers are. Companies like Anthropic pay Brave for web access so that their models can answer questions requiring up-to-date information. Meanwhile, Anthropic burned billions on operations in 2025 and projects positive cash flow only in 2028. Brave, which sells data to these companies, is probably profitable — or very close to it — at a fraction of their costs.

Brave is moreover actively reinforcing this dynamic. A benchmark evaluation from November 30, 2025, which Brave published on February 12, 2026, shows that Ask Brave (a chat assistant running on the open Qwen3 model with data from Brave) achieves better results than ChatGPT and Google AI Mode in answer-quality evaluation. The evaluation was conducted by Brave itself, using Claude models as judges. With this, Brave demonstrates its key thesis: data quality matters more than model size. This thesis must, however, be taken with a grain of salt — it is an internal comparison, not an independent audit.

This story does, however, have its limits and risks, which need to be named.

First, Brave somewhat overstates its position as "the only independent API." Google still offers its own search API (Custom Search JSON API) with limitations, and smaller independent indexes exist, such as Mojeek. DuckDuckGo, thanks to long-term contracts with Microsoft, retained access to Bing's data even after the public API was shut down. Brave is nonetheless the only one offering a full-fledged, commercially available search API at global-web scale without dependence on Google or Microsoft — in this specific sense, the claim is justified.

Second, the AI industry's dependence on a single supplier of search data is itself a risk. If Brave were to raise prices, change terms, or face technical problems, the impact on the entire ecosystem of AI assistants would be considerable. In essence, one form of dependence (on Google/Bing) has been replaced by another (on Brave).

Third, Brave's financial data is limited. The company is not publicly traded and does not publish audited financial statements. The "100 million dollars in revenue" figure comes from Brave press releases, not from an independent audit. The breakdown of revenue across individual segments (ads, API, subscriptions) is not publicly known. The estimate of API revenue in the tens of millions of dollars is an extrapolation based on indirect indicators.

Fourth, Brave itself benefits from AI companies using its data to create products that potentially compete with Brave Search. It is a symbiotic relationship with internal tension: the better AI assistants become at answering queries directly, the less users will need to visit any search engine — including Brave.

The position Brave has gained is strong, but not necessarily permanent. Several factors could change the dynamics.

Google may make its index available to AI companies on commercial terms — indeed, it already does so partially for select partners. Microsoft may offer direct access to the Bing index outside Azure AI Agents. New independent indexes may emerge, even though the barrier to entry is enormous (crawling billions of pages per day requires substantial infrastructure).

The most likely short-term scenario is that Brave will strengthen its position. Integration into Snowflake, availability on the AWS Marketplace, security certifications, zero data retention — these are all steps that raise switching costs for enterprise customers. The deeper the Brave API becomes embedded in the infrastructure of AI companies, the harder it will be to switch to an alternative.

The question the entire AI industry should be asking is not whether Brave is justified in profiting from the AI boom — it undoubtedly is. The question is what will happen once Brave fully realizes its bargaining position and begins to exploit it. When Microsoft raised the prices of the Bing API by hundreds of percent, customers left for Brave. Where will they go when Brave raises its prices?

Methodological note: This article draws on official posts on the Brave Software blog (brave.com/blog), press releases from Brave and Snowflake, interviews with Brave's leadership (Cerebral Valley, Digiday), Microsoft's announcement of the Bing API shutdown (learn.microsoft.com), Google's lawsuit against SerpApi (December 19, 2025, US District Court, Northern District of California, case no. 5:25-cv-10826), and coverage by The Register, Search Engine Land, and developer-tech.com. Brave's financial figures are not independently audited. The search API revenue estimates are the author's extrapolations based on indirect indicators. The comparative answer-quality evaluation comes from Brave and has not been independently replicated. Data cutoff: February 27, 2026.

Transparency of creation

The concept, structure, and editorial line of the article are the work of the author, who prepared the content outline, established the key theses, and directed the entire creative process. Generative AI (Claude Opus 4.6, Anthropic) was used as a tool for research, fact-checking, and fleshing out the author's draft.

The author verified the key findings and approved the final wording. No part of the text was published without conscious authorial control. The factual data was verified against the publicly available sources cited in the text.

This procedure complies with the transparency principles of EU Regulation 2024/1689 (AI Act). #poweredByAI