Bright Data Is Building a Toll Road for the Machine-Readable Web

The modern web has a small practical joke built into it. Its pages are public enough for a shopper to compare two hotel rooms, yet often hostile to a machine asked to compare two million. The first requests work. Then the rate limits arrive. A CAPTCHA pops up. The price changes by country. A retailer redesigns its product page and yesterday's neat selector returns an empty cell. What looked like a data project becomes a small operations department.

Bright Data sells relief from that department. The Israeli company provides the machinery for finding, reaching, rendering, extracting and cleaning public web information at scale. Its customers can rent residential, mobile, ISP or datacenter proxy access; call an API that handles blocking; run code inside managed browsers; buy a ready-made dataset; or ask Bright Data to operate the collection process. Increasingly, an AI agent can do the asking through a model-context server or command-line tool.

That breadth is the point. A decade ago, the company was Luminati Networks, a proxy infrastructure business founded in 2014 by Ofer Vilenski and Derry Shribman. EMK Capital acquired it in partnership with the founders in 2017. It added an unblocker, automated collection, datasets and retail analytics, then adopted the Bright Data name in 2021. Each step moved it farther from selling a raw connection and closer to selling a completed answer.

From a page to a row

Bright Data's simplest explanation is also its most revealing: turn an unstructured web page into a structured table a machine can process. That transformation hides three different jobs. First comes access - reaching a public page from the right location without the request being rejected. Next comes interpretation - rendering JavaScript, clicking through a flow or identifying the useful fields. Finally comes delivery - returning validated records in a format that can land in an application, warehouse or model pipeline.

Consider a retailer monitoring a competitor's prices. The posted number may depend on location, login state, inventory and promotion. A financial researcher may need the same product history every morning, with consistent fields, even after the page design changes. An AI research agent has an even shorter fuse: it needs current context during a conversation, not a batch export next Thursday. Bright Data's promise is to absorb those variations behind a stable interface.

Using pre-collected datasets in conjunction with live scraping using APIs was key to increasing customer value.Nathan Ondracek, CTO and co-founder of Raylu

Raylu, an AI research platform for private-market investors, offers a compact example. Its case study says a combined pipeline of historical datasets and live collection cut delivery of a market landscape from two hours to 15 minutes, while expanding the number of core sources. The result matters less as a universal benchmark than as a sketch of the customer job: use old data for coverage, new data for freshness and one provider to keep both moving.

One platform, several ways to buy

The product catalog follows the customer's willingness to operate machinery. At the lowest layer, proxy networks give technical teams direct control over routing. Web Unlocker accepts a target and deals with blocking, while Browser API hosts browsers for Puppeteer, Selenium and Playwright scripts. Web Scraper APIs return structured records from popular sites. The datasets marketplace removes collection from the customer's workload altogether.

Scraper Studio, launched in 2026, compresses the stack further. A user describes the public site and desired schema; the service builds a hosted endpoint and is designed to repair collection logic when the site changes. The feature is a tidy expression of Bright Data's strategy. The customer does not really want proxies or selectors. The customer wants a dependable API that happens to survive the internet.

A meter attached to every layer

Bright Data's business model changes with the amount of work it assumes. Proxy and browser products can be metered by traffic. Some scraping APIs and the Web Unlocker charge for successful requests, aligning the bill with usable output. Dataset pricing can depend on the number of records and refresh schedule. Large managed projects arrive through sales with custom scope, delivery and support. A referral program and thousands of solution, cloud and affiliate partners widen distribution.

The company says annual recurring revenue passed $300 million in 2025, up from the $100 million revenue milestone announced in 2021. It reports more than 20,000 customers and 450-plus team members. The customer base reaches across e-commerce, finance, travel, advertising, cybersecurity, market research and academia. Public examples include Bitget, Remazing, Kingston Brass, Faber and Post for Rent, plus an automation project that collected daily news, commodity prices and exchange rates for ThyssenKrupp.

The difference is the stack

Bright Data competes in several neighborhoods at once. Oxylabs, Decodo, NetNut, SOAX and Webshare sell proxy infrastructure. Zyte, Apify, ScraperAPI and ScrapingBee simplify extraction. Newer agent-oriented products such as Firecrawl compete for the developer who wants clean web context with a small amount of setup. A narrow tool may be cheaper, friendlier or perfectly adequate for a single source.

Bright Data's argument is consolidation. The same vendor can supply the route, browser, unblocker, parser, stored dataset and managed feed, with enterprise support across them. Its integrations with AWS, Snowflake, Databricks and IBM watsonx place the output where large customers already keep data and build AI. Typed Python and JavaScript SDKs, a CLI and a Model Context Protocol server bring the other end of the pipeline closer to developers.

Scale is useful here because web collection is a long tail of failure modes. It also creates a feedback loop. More varied workloads expose more blocks and page structures; those lessons can improve shared infrastructure. The company's patent count - stated as more than 5,500 granted claims - reflects how seriously it has treated that engineering layer. Still, breadth brings complexity. A buyer with one modest task may not need the whole machine.

The compliance desk beside the tollbooth

Web-data infrastructure lives in a permanent argument about what public means. A page can be viewable without a login while its operator objects to automated collection. A record can be public and still contain personal information. An AI company may be allowed to retrieve data without being free to use it for every conceivable purpose. No proxy rotation trick resolves those questions.

Bright Data makes compliance part of its sales case. It introduced customer identity checks early, publishes acceptable-use rules, employs compliance leadership and describes its IP sourcing as consent-based. Those controls are also commercially necessary: banks, research institutions and model labs need vendors that can survive procurement and audit review, not merely return a successful HTTP response.

The legal boundary remains contested. X sued Bright Data in 2023 over the scraping and sale of public information. A federal judge dismissed the claims in 2024, rejecting the idea that X could use the asserted theories to create broad control over public data. Bright Data also prevailed in litigation with Meta. The outcomes strengthened the company's access argument, but they did not convert all public scraping into a blanket permission slip. Privacy, copyright, contract and purpose still matter case by case.

Our mission is about ensuring that web data is accessible, transparent, and ethically sourced.Or Lenchner, chief executive officer

The Bright Initiative gives the mission a public-interest outlet. More than 750 nonprofit organizations, universities and public bodies receive pro-bono tools and expertise for work involving health, human rights, discrimination, labor and the environment. It is useful community work, and it also demonstrates that web collection is not limited to price bots and lead lists.

Now the customer has no hands

The arrival of AI agents changes the interface more than the underlying headache. A person can notice a cookie dialog, revise a search or decide that a result looks stale. An agent needs those capabilities expressed as reliable tools. Bright Data's Web MCP offers scoped functions for search, extraction and browser navigation; its 2026 launch added custom tool groups, one-click connections and observability. The company says these configurations can reduce the irrelevant tool descriptions that consume an agent's context window.

This is where Bright Data fits in the current AI market: below applications and model orchestration, but above raw network access. It supplies live context when a model's training cutoff is not enough. It also sells bulk text, image, audio and video for model development. The company says 14 of the top 20 LLM labs use its infrastructure, though it does not publicly identify the entire group.

The strategic risk is that every layer is attracting competition. Proxy companies are adding AI features. Scraping platforms are shipping agent connectors. Model providers and cloud platforms can build retrieval tools of their own. Bright Data's defense is accumulated infrastructure, operational knowledge and the ability to serve a customer that starts with one API call and ends with a petabyte-scale feed.

There is a lesson in the company's climb. Sell the primitive, watch what customers repeatedly assemble around it, then package the repeated work. Luminati sold the route. Bright Data added the vehicle, the map, the cargo inspection and, finally, a courier that speaks to robots. The web is still messy on the other side of the gate. That is why the gate has value.

Keep exploring

See the products in action, browse developer examples or follow the company where it publishes updates and demonstrations.