Who owns the market data industry

How it works

A price on a screen has passed through five layers, each owned by a different kind of company.

The venue generates data as a by-product of matching orders and sells it twice: as a proprietary direct feed, and into the consolidated tape its regulator requires. The vendor — a terminal, a feed handler, an API — licenses from many venues, normalises, and resells with a workflow around it. The identifier layer decides whether two firms agree that this instrument is that instrument. The index layer turns prices into a benchmark and licenses it to funds, which pay on assets rather than usage. The ratings layer sells an opinion regulation makes expensive to ignore.

Money moves up the chain: the fund pays the index provider, the bank pays the vendor, the vendor pays the venue, and nearly everyone pays for identifiers whether or not they see the line item. Defensibility moves the other way. The layer a competitor can rebuild in a year is the terminal; the layers nobody can rebuild are the ones whose whole value is that everybody already agreed on them.

The Reuters correction

Most writing on this subject gets one name wrong, so here is the chain in order.

Thomson Reuters sold a 55 per cent interest in its Financial & Risk business to a Blackstone-led consortium, with Canada Pension Plan Investment Board and GIC alongside; the sale closed on 1 October 2018 for roughly 17 billion dollars gross, and the business was renamed Refinitiv. LSEG's 2025 annual report records the next step in its glossary: Refinitiv "was founded in 2018" and "became a subsidiary of London Stock Exchange Group as of 29 January 2021".

Reuters News did not go with it. Thomson Reuters' 2025 annual report states that Reuters supplies news and information services to "the Data & Analytics business of LSEG" through 1 October 2048, and recorded 398 million dollars under that agreement in 2025 — "the current minimum annual value", adjusted for inflation and currency. Against Reuters segment revenue of 853 million dollars, LSEG is close to half the news agency's business.

So Reuters is a news agency with one very large customer, and Refinitiv is a retired brand inside LSEG. Anything headed "Bloomberg versus Reuters" is describing 2017.

The map, and what counts as market share

The share figures that circulate come from Burton-Taylor International Consulting, a paid research product; what is open is second-hand paraphrase, so nothing below is taken from there. The alternative is to build the comparison from what the companies file themselves.

The denominator is the revenue eight reporting companies attributed, in their own 2025 segment disclosures, to financial data, index and ratings businesses; trading, clearing, listings, mortgage technology and automotive data are excluded. LSEG is converted from sterling at the 2025 average of 1.3192 dollars per pound (Federal Reserve G.5A). FactSet's year ends 31 August, the rest 31 December.

Roughly 41.6 billion dollars, with the top three at 68.8 per cent of it. Every figure comes from the filing cited below; the arithmetic is the only part that is ours.

Each is defended in a different layer. S&P Global sells research, commodity price assessments, indices and ratings — the last two are the position. Moody's sells ratings and risk analytics; the ratings half is defended. LSEG sells terminals, feeds and FTSE Russell benchmarks, and only the benchmarks are hard to displace. ICE's data position is fixed income evaluated pricing, where there is no exchange print to compete with. MSCI is almost entirely index licensing, and Nasdaq's smaller index business sits in the same layer. FactSet sells a workstation and feeds, with CUSIP underneath. Morningstar sells fund data, PitchBook and, through Morningstar DBRS, ratings.

Bloomberg is not in the table, and that is the point. Bloomberg L.P. is privately held, files no annual report and publishes no revenue. Adding a third-party estimate to a column of audited segment numbers would make the whole column an estimate — which is why every published market-share figure here is one. Two others sit outside for the same reason: SIX Group, owned by its member banks, whose Financial Information unit reported net operating income of 408.2 million Swiss francs for 2025 within a group total of 1,496.5 million; and Infront, the European terminal challenger, privately held since it left the Oslo Stock Exchange in 2021.

One layer does have a share published by a regulator. The SEC's NRSRO statistics show 2,192,543 outstanding credit ratings across ten registered agencies at 31 December 2025: S&P holds 1,077,798 (49.2 per cent), Moody's 684,055 (31.2 per cent) and Fitch 274,291 (12.5 per cent) — 92.9 per cent between three firms, counted rather than estimated. It exists because registration under Section 15E makes the count public record.

What it costs

By layer, with the price-setter named. Margin measures differ — operating margin for some, adjusted EBITDA for others — because that is how each reports.

Exchange data. The venue sets the price and files it with its regulator — in the US under Regulation NMS, where the SEC may suspend or disapprove it. ICE reported 1,031 million dollars of data and connectivity revenue inside its Exchanges segment for 2025; Nasdaq, 804 million of Data & Listing Services revenue. A challenger cannot replicate this: the data is a by-product of order flow, and the order flow is the business.

Vendor subscriptions. The vendor sets the price but pays the layers below it. LSEG's Data & Analytics division took 3,978 million pounds in 2025 at a 40.7 per cent adjusted EBITDA margin, after 821 million of cost of sales for "purchased content and royalties, including news, specialist data and exchange data". S&P Global's Market Intelligence, the largest single data business in the table, returned a 20.2 per cent operating margin; FactSet's whole-company figure was 32.2 per cent. This is the competitive layer, and its margins say so.

Index licensing. The provider charges funds on assets under management, so revenue compounds with markets rather than headcount. MSCI's Index segment: 1,786.8 million dollars at a 76.4 per cent adjusted EBITDA margin. FTSE Russell: 954 million pounds at 66.6 per cent. S&P Global's Indices segment: 1,839 million dollars at a 69.1 per cent operating margin. A challenger would have to persuade asset owners to rewrite mandates naming a specific benchmark — a governance process, not a sale.

Ratings. The agency sets the price and the issuer pays, because investor mandates and capital rules are written around the output. S&P Global's Ratings segment: 4,549 million dollars at a 66.2 per cent operating margin. Moody's Investors Service: 4,317 million of segment revenue and 2,746 million of adjusted operating income, 63.6 per cent. Replicating that needs NRSRO registration and then decades of accepted track record.

Identifiers. FactSet's 10-K describes CUSIP Global Services as "the exclusive issuer of CUSIP and CINS identifiers globally" and the official ISIN numbering agency for the US. FactSet paid S&P Global 1.925 billion dollars for it in 2022 — for a numbering scheme. The margin is not the useful figure here; the price paid is.

What the concentration is made of

Switching costs across a desk. Replacing a vendor means reconciling every downstream report, model and compliance artefact built on its conventions. The cost is not the licence; it is the year of parallel running.

Ticker and definition consistency. Two firms confirming a trade must agree on the instrument, its corporate action history and the adjustment convention. Buying from the same vendor as your counterparties is a way of buying agreement, which is why the reference data layer holds when the front end changes.

The messaging network. The most cited reason a Bloomberg Terminal survives a budget review has nothing to do with data: the buy side is contractually reachable on it, and network effects are not priced out of existence by a cheaper feed. See what to use instead of a Bloomberg Terminal.

Regulatory entrenchment. Ratings carry the NRSRO designation, an SEC registration created by the Credit Rating Agency Reform Act of 2006 that determines which opinions certain rules accept. Exchange data rights run through Regulation NMS: venues supply core data to the consolidated tape, file their fees, and sell richer proprietary feeds alongside.

Index and identifier licensing. Both license a vocabulary rather than a file — permission to use a name other people already recognise, which is why a cheaper substitute does not substitute.

What is moving

A European consolidated tape, being built now. Under the MiFIR review, ESMA selected Ediphy (fairCT) as the first consolidated tape provider for bonds on 3 July 2025 and authorised EuroCTP for shares and ETFs on 27 July 2026, with a transition to 30 September 2026 before the service starts and a five-year term; a third selection, for OTC derivatives, has followed. It is the first structural change to European market data economics in a decade, and it is not yet running.

The US rule is upheld and still phasing in. The SEC adopted the Market Data Infrastructure Rule in December 2020 (Release 34-90610), adding competing consolidators and self-aggregators alongside a broader definition of core data. Exchanges challenged it and lost — the D.C. Circuit denied the petitions in Nasdaq Stock Market LLC v. SEC on 24 May 2022 — but a second case that summer vacated the SEC's CT Plan Order on 5 July 2022, striking the single-plan governance reform while leaving the rule standing. Implementation moved into the 2024 Regulation NMS amendments, challenged in turn and upheld on 14 October 2025; the SEC's order of 31 October 2025 then pushed compliance for the amended pricing increment and access fee caps to November 2026. That order still describes consolidated information as disseminated "through an exclusive securities information processor".

Open identifiers exist and are unevenly adopted. The Legal Entity Identifier is free to look up and issued through GLEIF's accredited issuers; the active population passed 2.93 million at the end of 2025. It is also the case against assuming open beats incumbent on merit: the same report puts the renewal rate at 56.7 per cent and good standing at 61.7 per cent. FIGI, the instrument-level open standard maintained by the Object Management Group, is free to use and still does not appear on the confirmations that matter. Which identifier to build on works through what each one costs.

LLM interfaces erode the terminal, not the licence. What a model answering in natural language competes with is the user interface — screens, shortcuts, function codes — the one layer a competitor could always rebuild. The layers underneath are untouched, and the incumbents are already writing the contracts that say so: LSEG's 2025 annual report describes launching "MCP infrastructure" under its LSEG Everywhere strategy, and states that its AI-ready content "will also be accessible to licensed ChatGPT and Claude users via MCP". The load-bearing word there is licensed. The chat window is new; the entitlement check behind it is the one that was behind the terminal.

What you can do about it

Layers you will pay for either way. Real-time exchange data, if you need it live and attributable: the fee is set by the venue and passed through by whoever you buy from, so shopping vendors changes the markup, not the base — and the venue's schedule is public, which is what why real-time stock data is so expensive reads. Index data, if your workflow names a benchmark, where historical membership is the expensive half. Ratings, if you are in credit.

Layers you can route around. Delayed and end-of-day prices carry a fraction of the licensing weight, and a surprising amount is published free by regulators and central banks — Tiingo sells equity history back to 1962 for tens of dollars a month, and the rest of market data APIs is priced in the same range. Fundamentals and filings come from issuer disclosure rather than a venue, and analytics you can compute from prices you already licensed. So can the terminal interface: Koyfin and the rest of fundamentals and research exist because the screens were always the replaceable layer.

Ask any cheap vendor which layer its price comes out of. Several publish the answer already, and where research terminals get their data collects what they say. A provider redistributing exchange data under its own licence — Databento is explicit about it, billing historical usage by the gigabyte — has a cost base that scales with what it pays the venues, and pricing that is durable because it is real. A provider whose price implies it pays nobody is either serving data whose licence does not cover your use, or one enforcement letter from repricing. That is the durability question in a cheap tool, and it is not a question about quality.

Design for the switch, and know where you have no leverage. The vendor layer is competitive precisely because its customers can leave: put an abstraction between your code and your provider so that leaving is a configuration change — OpenBB does this across roughly thirty providers — and you are buying from the one layer where being a small customer still works. But if what you need is the same number, at the same instant, as the counterparty you are confirming with, you are buying agreement rather than data, and agreement is sold by whoever everyone else already bought it from.