The External Data Market Map 2026: 145 Companies, Four Routes

Every company that runs on data eventually needs data it does not have: competitor prices on other people's websites, provider records spread across thousands of hospital pages, filings, listings and job posts. The industry calls this external data.

Most market maps of this space list tools, such as scrapers, proxies and AI search APIs. In practice, a buyer does not start with a tool. They start with a question about where the data should come from.

Do we buy it ready-made, license it from whoever owns it, build the collection ourselves, or have someone build it for us? This external data market map is organised around that choice, with 145 companies across the four routes, what each one costs you, and what AI changed in each during 2026.

The External Data Market Map, 2026

Download the full-resolution map (PNG), the print version (PDF) or the full company list (CSV), which has every company's website, category and owner. You are welcome to share the map, with credit to Forage AI.

What are the four ways to get external data?

There are four ways to get external data: buy it ready, license it, build it yourself, or have it built. Each box on the map is one of them, and the two lines at the top of each box matter more than the names underneath.

"You get" is what the route is good at. "You give up" is the cost it carries, and that cost stays with the route even when you pick a better vendor inside it. A better data vendor still sells you its own schema, and a better proxy network still leaves your team fixing the scraper every time a site changes. Inside each box, categories group companies by what they sell, in alphabetical order, and the italic "AI shift" line sums up what AI changed in that route this year.

In practice, each route fails in a predictable way, and we see the same patterns across the teams we work with. A global asset manager bought firmographic data from a major B2B vendor, then found in an audit that 40% of records had no website and 30% had a dead or wrong one. That is the cost of buying ready-made: the data was never built for their use.

A secondary-ticket marketplace had a strong developer build scrapers for the three biggest ticketing sites in a few days. Then the team counted roughly 500 sites in its market. That is the cost of building: the first sources are fast, and the long tail is where the time goes.

Having it built carries its own risk, and it is worth naming. A broadcaster came to us after its previous data vendor went out of business mid-contract and left it racing to replace the feed. When someone else runs the pipeline, their staying power becomes part of your risk.

So the useful question is not which route your company uses. It is which route each data need belongs on. List your needs one per line, mark each against the "you give up" column, and move any need whose cost keeps hurting.

External data companies by category

The tables below cover all 145 companies on the map, one table per route. Each row says what the company is best used for and what it sells, in plain terms. Companies sit in alphabetical order within each category, and the full list with websites and owners is in the CSV download above.

1. Buy it ready: data vendors and marketplaces

Data vendors and marketplaces that sell data someone else has already collected. Pick from here when a standard dataset answers your question. 45 companies in 7 categories.

2. License it: content licensing and AI data brokers

Marketplaces, edge networks and brokers that sell the right to use content its owner controls, mostly for AI training and retrieval. 21 companies in 3 categories.

3. Build it yourself: data collection tools and infrastructure

The infrastructure, APIs, platforms and open-source frameworks your own team uses to collect external data itself. 54 companies in 7 categories.

4. Have it built: managed data services

Providers that build and run a custom data pipeline, or create a custom dataset, to your spec. We sit in the first category. 25 companies in 3 categories.

What changed in the external data market in 2026?

Four shifts stood out in the research behind this map, and each one moved a route forward without removing the trade-off that route carries.

Data vendors moved inside AI assistants. The ready-made route is being rebuilt around AI agents. ZoomInfo now offers GTM.AI, which lets agents in tools such as Claude, ChatGPT and Copilot query its company data directly. S&P Global and Anthropic announced in July 2025 that Capital IQ financials and earnings call transcripts would be available inside Claude. Apollo, Similarweb and Regrid have each shipped MCP servers of their own, and Snowflake Marketplace, with more than 750 providers, now lists agent products next to its datasets. Buyers get faster access to the data, but the schema is still the vendor's.

Licensing became a priced market. In July 2025, Cloudflare, which serves roughly a fifth of all web pages, began blocking AI crawlers by default on new sites and launched pay per crawl, which charges AI bots for access. In July 2026 it began moving to pay per use, where payment follows content appearing in an AI answer. Cloudflare also bought the licensing marketplace Human Native in January 2026. Microsoft opened a Publisher Content Marketplace pilot in February 2026, and the Really Simple Licensing standard for machine-readable AI terms reached version 1.0 in December 2025. Licensing now has prices and plumbing, though it still covers only the content whose owners opt in.

Building got cheaper to start and riskier to run. Investors put heavy money into the build route. Parallel Web Systems raised at a valuation of about $2 billion in April 2026. Exa raised $85 million at a $700 million valuation in September 2025, and Nebius agreed to buy Tavily for $275 million in February 2026. Firecrawl's open-source repository has passed 189,000 GitHub stars. Over the same period, the risks of running collection yourself became concrete. Google sued SerpApi in December 2025 over the scraping of its search results. In July 2026 the FBI seized domains belonging to the proxy provider NetNut over allegedly botnet-sourced IP addresses, which its parent company disputes. Several build tools now promise extraction with no maintenance at all. That promise is worth testing against your own sources over several months, not in a demo. We have also watched legal teams stop in-house scraping outright, at a financial-software company that had both the engineers and the budget to build.

Consolidation folded smaller names into bigger ones. Many companies that appeared on last year's lists now sit inside someone else. Oxylabs bought ScrapingBee, Elastic bought Jina AI and RealPage bought Cherre, while H1 absorbed Ribbon Health, Consumer Edge absorbed Earnest Analytics and EXL bought iMerit. In human data, Meta's purchase of 49% of Scale AI in June 2025 showed buyers that the owner of a data provider can become a neutrality question. The map lists every acquired company under its acquirer, because that is who you would be signing with.

Read together, the four shifts point the same way. Every route got better at what it was already good at, and the choice between routes is still a choice about which cost you would rather carry.

Where Forage AI fits in the external data market

We build and run custom web and document data pipelines for companies whose product or analysis depends on external data. That puts us in route four, under managed web data services, next to providers we compete with every week. We listed them on the same terms as ourselves: alphabetical order, the same chip, no ranking. The diamond marks us as the author of the map and nothing more.

We do think the map shows why route four exists. Companies rarely start there. They tend to arrive after another route stopped working for one specific data need: a dataset that did not fit, a scraper that kept breaking, or a licence that covered too little. If that describes a source you are dealing with now, our team can usually tell you on one call which route it belongs on, including when the answer is not us.

How we built the external data market map

We started from the buyer's decision rather than from a list of tools, defined the four routes, and then researched each route's categories from scratch, using each company's own website, funding announcements and acquisition news.

• Scope: companies that sell data the buyer does not own, sell the right to use it, or sell the means to collect it. Storage, warehousing and analytics tools are out of scope.
• Checks: on 7 October 2026 we confirmed a live website and a current commercial offering for every company, along with its 2026 brand name and owner.
• Placement: each company appears once per product line it sells separately. Bright Data appears three times because its proxies, datasets and managed service are separate offerings.
• Exclusions: we left out companies we could not verify, products that now exist only as a feature inside another platform, and NetNut, whose domains were seized in July 2026.
• No paid placement: no company paid or asked to be listed.

The map is not exhaustive, and a market that consolidates this quickly will date it. If your company belongs on it, or a listing is out of date, tell us through our contact page and we will review it for the next edition.

Verdict: which external data option should you choose?

The external data market is how companies get data they do not own, and in 2026 it splits into four options. Choose by the cost you can carry, not by the vendor:

• Buy it ready when a standard dataset answers your question and you can accept the vendor's schema.
• License it when you need specific owned content, such as news, books or media, cleared for AI use.
• Build it yourself when collection is a core skill and your sources are few and stable.
• Have it built when you need custom data from many or difficult sources and do not want to staff its upkeep.

In practice, most teams use two or three of these at once. Check each data need against its route once a year, because the source that was quick to build in January can be the one breaking every week by December.

Frequently asked questions

What is the external data market?

It is every way companies get data they do not generate themselves: buying datasets from vendors, licensing content from its owners, building their own collection, or paying a provider to build and run it for them. It spans data vendors and marketplaces, licensing intermediaries, scraping and AI search infrastructure, and managed data services.

What is the difference between buying and licensing data?

Buying usually means paying for access to a dataset that a vendor has already collected and packaged. Licensing means paying a rights holder, such as a publisher or an archive, for permission to use content it owns, often for AI training or retrieval and always on its terms.

When does a managed data service make more sense than building in-house?

When you need custom data from many or difficult sources, refreshed often, and keeping collectors running would pull engineers away from your product. Teams with a few stable sources and collection skills in-house often do well building. Our build vs buy guide walks through the decision.

Can I use this map in my own work?

Yes. Share or embed the map and the company list freely, with credit to Forage AI and a link to this page.

Related articles

• Top AI Training Data Providers (2026): Buyer's Guide: a closer look at the AI data side of routes one, two and four.
• How to Choose a B2B Data Provider: for teams buying company and people data ready-made.
• What Is Managed Web Data Extraction?: how route four works in practice.
• When to Outsource Data Extraction: 7 Signs It's Time: how to tell when a need has outgrown the build route.