The industry runs on manuscripts it does not pay for and quality control it does not pay for. Thirteen years of data on 4,221,735 articles a year show why that structure has proved so hard to dislodge.
Five commercial publishers account for roughly half of the world's scientific output and earn operating margins above Apple's. This piece measures the structure directly, on the full OpenAlex corpus from 2012 to 2025, and pairs it with the industry's own accounts. The central finding is that open access did not decentralise the market: the two publishers that grew fastest under it added their volume on top of the incumbents rather than taking share from them, and then raised their prices toward incumbent levels.
Begin with the cost structure, because it is the whole argument. A scientific publisher acquires its raw material at no cost and its quality control at no cost, then sells the result to the same institutions that supplied both.
| Input | Cost to publisher |
|---|---|
| Manuscripts | nil |
| Peer review | nil |
| Typesetting & hosting | real, and falling |
| Distribution | the revenue line |
In February 2026 RELX, the London-listed parent of Elsevier, reported £2,714m of revenue and £1,035m of adjusted operating profit in its scientific, technical and medical division — a margin of 38.1%, up from 37.4% the year before.1 That is roughly £2.8m of profit every day of the year, from an activity whose two most expensive inputs are donated to it. 1RELX restated its 2024 segments, moving print and a small healthcare portfolio out of STM. Figures here use the restated basis so the two years are comparable.
The four largest commercial journal publishers together took in $7.8bn in their latest fiscal years and kept $2.7bn — a blended margin of 34%.
Trade publishing is a hard business, running on high-single-digit margins. Scientific publishing is a different animal, and the useful comparison is not to other publishers but to the most profitable companies in the world.
| Company | Operating margin | Basis |
|---|---|---|
| Microsoft | 45.6% | FY2025 (June) |
| Elsevier | 38.1% | FY2025, adjusted |
| Taylor & Francis | 36.6% | FY2025, adjusted |
| Wiley Research | 33.2% | FY2026, adj. EBITDA |
| Alphabet | 32.1% | FY2024 |
| Apple | 32.0% | FY2025 (Sept) |
| Springer Nature | 28.2% | FY2025, adjusted |
| Walmart | 4.4% | FY2025 (Jan) |
Elsevier's 38.1% sits above Alphabet's 32.1% and Apple's 32.0%. Taylor & Francis runs at 36.6%, Wiley's research segment at 33.2%, Springer Nature — the most open-access-heavy of the four, and the one carrying the most debt from its private-equity years — at 28.2%. Only Microsoft, with a near-monopoly in enterprise software, clears the publishers comfortably. Walmart, a genuinely competitive business, earns 4.4%.
Margins like these in a mature industry normally attract entrants until they are competed away. That has not happened, and the reason is structural. It begins with how concentrated the market became.
The canonical measurement is Larivière, Haustein and Mongeon's 2015 study of 45 million Web of Science documents.2 In the natural and medical sciences the top five publishers held about 20% of papers in 1973, 30% by 1996, and 53% by 2013. That number is now more than a decade old, and the decade it missed was the eventful one. 2Larivière, Haustein & Mongeon (2015), PLOS ONE 10(6). No equivalent Web of Science census has been repeated at that scale since.
What follows re-measures it on OpenAlex, which indexes far more of the world's journals than Web of Science ever did. Every article is counted once, against the group that owns the journal it appeared in: 2,909,960 articles in 2012 rising to 4,221,735 in 2025.
| Year | Physical | Life | Health | All fields | Social |
|---|---|---|---|---|---|
| 2012 | 500 | 745 | 894 | 517 | 225 |
| 2013 | 512 | 712 | 859 | 508 | 218 |
| 2014 | 553 | 743 | 886 | 531 | 213 |
| 2015 | 621 | 754 | 916 | 561 | 205 |
| 2016 | 730 | 821 | 933 | 610 | 215 |
| 2017 | 817 | 801 | 931 | 631 | 195 |
| 2018 | 758 | 772 | 898 | 591 | 180 |
| 2019 | 765 | 770 | 852 | 586 | 184 |
| 2020 | 834 | 765 | 859 | 611 | 183 |
| 2021 | 879 | 794 | 819 | 621 | 192 |
| 2022 | 996 | 859 | 859 | 691 | 215 |
| 2023 | 1,026 | 913 | 917 | 724 | 213 |
| 2024 | 1,096 | 983 | 970 | 778 | 225 |
| 2025 | 1,130 | 1,026 | 996 | 813 | 247 |
Across all fields the index rose from 516.9 to 812.6, an increase of 57%. But the aggregate conceals the more interesting fact: the physical sciences went from 500.2 to 1130.2, while the social sciences went from 225.3 to 246.8 — thirteen years, essentially flat. Whatever consolidated this market operated on laboratory science and left the seminar disciplines alone.
Concentration can rise because the leader grows or because the pack thins. Here it was the leader.
| Year | Elsevier | Springer Nature | Wiley | MDPI | Taylor & Francis | IEEE | Wolters Kluwer | ACS |
|---|---|---|---|---|---|---|---|---|
| 2012 | 17.4% | 10.1% | 8.0% | 0.3% | 3.9% | 1.2% | 2.5% | 1.4% |
| 2013 | 17.1% | 10.2% | 7.8% | 0.4% | 3.9% | 1.1% | 2.6% | 1.4% |
| 2014 | 17.7% | 10.5% | 7.8% | 0.4% | 3.8% | 1.3% | 2.9% | 1.4% |
| 2015 | 18.4% | 10.8% | 7.8% | 0.6% | 3.7% | 1.3% | 2.8% | 1.4% |
| 2016 | 19.6% | 10.9% | 7.6% | 0.7% | 3.9% | 1.4% | 2.9% | 1.5% |
| 2017 | 20.4% | 10.9% | 7.1% | 1.1% | 3.6% | 1.6% | 2.7% | 1.5% |
| 2018 | 19.2% | 11.0% | 7.3% | 2.0% | 3.4% | 1.6% | 2.7% | 1.6% |
| 2019 | 19.0% | 10.9% | 7.2% | 3.0% | 3.6% | 1.8% | 2.7% | 1.7% |
| 2020 | 19.5% | 10.8% | 7.2% | 4.2% | 3.4% | 2.0% | 2.7% | 1.6% |
| 2021 | 19.0% | 11.5% | 7.1% | 6.0% | 3.3% | 1.9% | 2.5% | 1.6% |
| 2022 | 19.8% | 11.9% | 7.3% | 7.7% | 3.1% | 1.9% | 2.5% | 1.7% |
| 2023 | 21.0% | 11.9% | 6.7% | 7.3% | 3.0% | 2.1% | 2.4% | 1.8% |
| 2024 | 22.5% | 12.6% | 6.6% | 5.8% | 3.0% | 2.3% | 2.2% | 1.9% |
| 2025 | 22.8% | 13.2% | 6.8% | 5.9% | 3.1% | 2.3% | 2.2% | 2.0% |
Elsevier, already the largest publisher in 2012 at 17.4% of world output, finished 2025 at 22.8% — it added more share over the period than any challenger won. Wiley is the only member of the big five to lose ground.
Acquisition did much of the rest. Wiley bought Blackwell in 2007, Hindawi in 2021 for $298m, and in mid-2026 Emerald Publishing and its roughly 485 journals for about $452m.3 Each purchase adds titles that already have reputations, editorial boards and subscribers; the buyer inherits a captive customer base rather than building one. 3Wiley valued Emerald at roughly seven times EBITDA after expected cost synergies — consolidation is attractive precisely because a journal's production cost falls when it moves onto a larger platform.
Analysts describe the arrangement as a triple-pay system. The same public purse funds the research, pays for its publication, and pays again to read it. No one in the chain sees the whole bill.
Three further features turn a cheap business into an extraordinarily profitable one.
Demand is inelastic, and the buyer is not the user. A researcher's career depends on publishing in and reading specific journals whose prestige has no substitute. The library that pays does not choose the journals; the faculty who choose do not see the price. Each journal is a small monopoly — there is exactly one Cell — and the bundle is a portfolio of them.
The working-capital cycle runs backwards. Subscriptions are paid at the start of the year for content delivered across it, so the publisher holds its customers' cash before incurring the cost of serving them. RELX converted 99% of adjusted operating profit into cash and has £2.25bn of buybacks planned for 2026.
The product is a database, and databases compound. A back catalogue does not depreciate the way unsold books do; a 1994 paper is still cited, downloaded and licensed. Elsevier is 28% of RELX revenue but 31% of its profit.
The open-access movement set out to remove payment three. It largely succeeded — 53% of Springer Nature's primary research articles were open access in 2025 — and in doing so it enlarged payment two. Under subscriptions, revenue was capped by the number of libraries in the world. Under the author-pays model it scales with the number of articles accepted.
If MDPI and Frontiers, the two publishers that grew fastest under that model, had been disrupting the incumbents, their gains would have come out of the big five's share. They did not.4 4Independent estimates put global APC spending at $0.91bn in 2019 and $2.54bn in 2023 — a 2.8-fold rise in four years. Subscription revenue did not fall by anything close to that.
| Year | Big five | Big five + MDPI & Frontiers | Gap |
|---|---|---|---|
| 2012 | 41.8% | 42.3% | 0.5 pp |
| 2013 | 41.4% | 42.1% | 0.7 pp |
| 2014 | 42.0% | 42.8% | 0.8 pp |
| 2015 | 42.9% | 44.0% | 1.0 pp |
| 2016 | 44.3% | 45.6% | 1.3 pp |
| 2017 | 44.3% | 46.0% | 1.7 pp |
| 2018 | 43.1% | 45.9% | 2.8 pp |
| 2019 | 42.9% | 46.8% | 4.0 pp |
| 2020 | 43.2% | 48.6% | 5.4 pp |
| 2021 | 43.1% | 51.1% | 8.0 pp |
| 2022 | 44.3% | 55.0% | 10.7 pp |
| 2023 | 44.7% | 54.1% | 9.4 pp |
| 2024 | 46.6% | 53.9% | 7.4 pp |
| 2025 | 47.7% | 55.4% | 7.7 pp |
The big five gained 5.9 percentage points over the period, ending at 47.7%. The same five plus MDPI and Frontiers gained 13.1, ending at 55.4%. A second layer of concentration was added on top of the first.
Open access did not decentralise scientific publishing. It added two more large publishers to it.
And then the entrants broke. MDPI published 9,374 articles in 2012 and 289,401 ten years later, a 31-fold rise that briefly made it the third-largest publisher of scientific articles in the world. It then fell 21% in two years. Frontiers peaked at 115,298 and fell 45%.
| Year | MDPI | Frontiers | Hindawi | BioMed Central |
|---|---|---|---|---|
| 2012 | 9,374 | 6,526 | 23,459 | 26,610 |
| 2013 | 11,715 | 9,385 | 26,386 | 29,830 |
| 2014 | 13,941 | 12,144 | 30,049 | 34,910 |
| 2015 | 18,280 | 14,352 | 23,541 | 34,983 |
| 2016 | 23,772 | 18,943 | 18,297 | 28,090 |
| 2017 | 36,669 | 19,958 | 16,303 | 29,711 |
| 2018 | 65,302 | 28,199 | 20,285 | 32,101 |
| 2019 | 106,486 | 32,168 | 18,199 | 35,030 |
| 2020 | 160,424 | 45,471 | 23,733 | 39,700 |
| 2021 | 232,223 | 79,557 | 29,131 | 45,303 |
| 2022 | 289,401 | 115,298 | 36,730 | 46,181 |
| 2023 | 273,458 | 78,920 | 16,580 | 47,210 |
| 2024 | 229,642 | 63,598 | 7,365 | 59,009 |
| 2025 | 248,487 | 75,252 | 7,567 | 74,891 |
Two independent measurements are available: the price publishers advertise, and the money institutions actually paid.
| Publisher | 25th pct | Median | 75th pct | 90th pct | Journals | Coverage |
|---|---|---|---|---|---|---|
| Oxford University Press | $3,016 | $3,534 | $4,017 | $4,368 | 406 | 89.3% |
| Wiley | $2,630 | $3,140 | $3,810 | $4,430 | 1,715 | 91.4% |
| Elsevier | $2,095 | $3,000 | $3,525 | $4,100 | 2,587 | 92.4% |
| Springer Nature | $2,890 | $2,990 | $3,490 | $4,190 | 2,409 | 76.5% |
| Frontiers | $1,897 | $1,900 | $2,950 | $2,950 | 109 | 95% |
| MDPI | $1,082 | $1,515 | $1,949 | $2,328 | 296 | 99.5% |
Oxford University Press has the highest median list price at $3,534, ahead of Elsevier at $3,000; MDPI is cheapest at $1,515. At the top of the range the numbers stop resembling costs altogether: Nature lists $12,850 and Cell $11,400, against a median of $2,040 for a US-published open-access journal and $950 worldwide. These are not the cost of producing an article. They are the price of a brand that academic careers require.
The realized prices are the more revealing series, and they run in one direction.
| Year | MDPI | Frontiers | Springer Nature | Elsevier | Wiley | Oxford UP |
|---|---|---|---|---|---|---|
| 2015 | €1,297 | €1,268 | €1,730 | €2,576 | €2,406 | €2,759 |
| 2016 | €1,290 | €1,703 | €1,658 | €2,781 | €2,144 | €2,567 |
| 2017 | €1,265 | €1,898 | €1,569 | €2,820 | €2,064 | €2,477 |
| 2018 | €1,177 | €1,980 | €1,644 | €2,904 | €2,083 | €2,532 |
| 2019 | €1,286 | €1,999 | €1,773 | €2,590 | €2,175 | €2,657 |
| 2020 | €1,517 | €1,999 | €1,990 | €2,182 | €2,160 | €2,680 |
| 2021 | €1,668 | €1,999 | €2,114 | €2,190 | €2,369 | €2,695 |
| 2022 | €1,876 | €2,496 | €2,390 | €2,260 | €2,400 | €2,619 |
| 2023 | €2,103 | €2,793 | €2,595 | €2,413 | €2,520 | €3,064 |
| 2024 | €2,458 | €2,892 | €2,653 | €2,318 | €2,560 | €2,927 |
MDPI's median payment rose from €1,297 to €2,458, up 90%. Frontiers went from €1,268 to €2,892, up 128%. Springer Nature, the incumbent, went from €1,730 to €2,653. The challengers' prices rose considerably faster than the incumbents', from a lower base, toward them.
“Academic publishing” is not one market. It is at least twenty-six, and they are not alike.
| Field | HHI | Big five | Big seven | Articles |
|---|---|---|---|---|
| Energy | 2,268 | 66.1% | 72.1% | 49,214 |
| Chemical Engineering | 2,143 | 62.8% | 68.2% | 17,228 |
| Materials Science | 1,863 | 64.0% | 69.6% | 174,085 |
| Chemistry | 1,429 | 54.2% | 59.4% | 73,331 |
| Engineering | 1,278 | 52.1% | 61.1% | 630,289 |
| Environmental Science | 1,254 | 55.4% | 66.6% | 231,489 |
| Biochemistry, Genetics and Molecular Biology | 1,200 | 56.0% | 66.8% | 255,484 |
| Earth and Planetary Sciences | 1,131 | 54.0% | 63.4% | 64,805 |
| Nursing | 1,119 | 52.5% | 63.7% | 24,682 |
| Immunology and Microbiology | 1,109 | 53.2% | 65.1% | 50,246 |
| Neuroscience | 1,062 | 56.0% | 65.7% | 72,134 |
| Medicine | 1,059 | 53.0% | 60.9% | 1,004,134 |
| Dentistry | 992 | 50.0% | 60.0% | 25,453 |
| Pharmacology, Toxicology and Pharmaceutics | 970 | 51.6% | 60.7% | 18,234 |
| Veterinary | 965 | 48.8% | 63.8% | 7,324 |
| Physics and Astronomy | 902 | 36.9% | 40.7% | 109,156 |
| Agricultural and Biological Sciences | 844 | 45.8% | 59.1% | 201,245 |
| Mathematics | 830 | 48.9% | 53.5% | 54,894 |
| Computer Science | 745 | 37.0% | 44.8% | 213,478 |
| Economics, Econometrics and Finance | 609 | 43.1% | 48.3% | 77,420 |
| Psychology | 533 | 46.5% | 53.7% | 130,174 |
| Decision Sciences | 526 | 40.9% | 47.0% | 33,639 |
| Health Professions | 503 | 40.8% | 45.7% | 102,363 |
| Business, Management and Accounting | 352 | 32.1% | 36.8% | 96,021 |
| Social Sciences | 179 | 27.3% | 30.4% | 386,782 |
| Arts and Humanities | 108 | 17.3% | 19.5% | 105,910 |
At the top, energy research: an index of 2267.7 with the big five taking 66.1% of everything published — above the level at which US agencies call a market highly concentrated. At the bottom, arts and humanities, at 108.1 and a big-five share of 17.3%. A 21-fold spread. Aggregate figures for the industry average a highly concentrated chemistry market with a fragmented humanities one and describe neither.
Every concentration estimate depends on which articles are counted. OpenAlex's great virtue — that it indexes the long tail of the world's journals — is also what pushes its estimates below Larivière's. Restrict the universe and the picture changes sharply.
oa_status:gold; DOAJ venues are journals listed in the Directory of Open Access Journals.| Universe (2019) | HHI | Big five | Big seven | Articles |
|---|---|---|---|---|
| All journal articles | 586 | 42.9% | 46.8% | 3,498,696 |
| DOAJ-listed venues | 323 | 28.0% | 39.3% | 826,412 |
| Gold open access | 846 | 31.9% | 59.4% | 503,107 |
| At least 10 citations | 1,129 | 57.5% | 64.9% | 1,464,803 |
On 2019 articles that went on to earn at least ten citations, the big five published 57.5% — against 42.9% of all 2019 articles. The index on the same restriction is 1128.5 against 585.9. The long tail that deflates the headline number is overwhelmingly a tail of low-citation work; measured on research that is actually read, the market looks much closer to Larivière's.
And in the gold open-access universe — the author-pays segment that grew out of the movement to break publisher power — the big five plus MDPI and Frontiers published 69.7% of everything in 2024, against 53.9% across all journals. The part of the market built to disperse concentration is the most concentrated part of it.
None of what follows has yet dented the margins in Figure 1. Each is a symptom of the same strain: a volume-driven revenue model bolted onto quality control that runs on unpaid goodwill.
Paper mills. When revenue scales with acceptances, the incentive to accept grows. Wiley's Hindawi is the clearest demonstration: bought as a ready-made volume engine, its guest-edited special issues were infiltrated at industrial scale by mills selling authorship on fabricated manuscripts. In 2023 alone Hindawi retracted more than 8,000 articles, pushing the world past 10,000 retractions in a single year for the first time. The cumulative count exceeded 11,000, Wiley wrote off $35–40m of revenue, and the brand was retired.5 Our own series puts the collapse at 80%, from 36,730 articles at peak to 7,365. 5Retraction Watch also counts at least 28 mass resignations of journal editorial boards since 2020, mostly in disputes over charge levels or pressure to raise acceptance rates.
The lesson is not that Wiley was uniquely careless. It is that a volume model with free quality control will be gamed by whoever can generate volume cheapest — and generative models have made fabricated manuscripts very cheap.
The lawsuit that named the mechanism. In September 2024 researchers filed a class action against six publishers, alleging they had agreed to fix the price of peer review at zero and to bar simultaneous submission. A federal judge dismissed it in January 2026, finding no plausible agreement: the practices are universal because every publisher independently benefits from them. That is legally decisive and economically beside the point. An industry where every firm reaches the same rules without needing to coordinate is the textbook description of a stable oligopoly.
Funders as buyers. The US National Institutes of Health moved in 2025 to cap the publication costs chargeable to its grants. It matters less for its dollar value than for who is making it. For thirty years the counterparty across the table was the university library, which has no leverage over what its faculty publish. A funder that pays for the research, mandates public access, and then limits what it will reimburse is the first buyer in the chain able to change author behaviour rather than merely complain about it.
The fourth payment. Since 2024 the publishers have found a new customer. Taylor & Francis licensed its corpus to Microsoft for $10m up front plus recurring payments, and Informa booked more than $75m of non-recurring data-licensing revenue in 2024. Wiley reported $49m of AI licensing revenue in fiscal 2026 and more than $110m cumulatively. The authors, who signed over copyright decades ago for nothing, were neither consulted nor paid.
The absolute level is low by antitrust standards. An all-field index of 812.6 is not, on its face, an oligopoly; US agencies do not call a market concentrated below 1,000. The case rests on the field-level numbers, where several disciplines clear the highly-concentrated threshold outright; on the restricted universes, where measured concentration roughly doubles; and on what an index cannot see — bundling, transformative agreements, and the fact that journals are not substitutes. A chemist cannot publish in a cheaper journal the way a shopper buys a cheaper detergent.
A third of journal articles cannot be attributed to a publisher. In 2025, 30.5% of journal articles in OpenAlex carry no publisher on the journal at all, and are excluded from every figure here. They skew small, regional and non-English, so including them would push concentration down — by an amount nobody measuring this corpus can quantify.
Comparability with Web of Science is imperfect. Our 2013 big-five share in the health sciences is 49.8% against the 53% Larivière's team reported for the natural and medical sciences. The gap is the venue tail, and Figure 9 is the demonstration of it. It is explained rather than tuned away.
What survives all three is the direction, which is the part that matters: concentration in the sciences rose over thirteen years, the entrants that were supposed to reverse it did not, and the price of publishing openly converged upward.
Universe. OpenAlex works with type=article,
is_paratext=false, a primary location whose source is a journal, and a non-null
publisher on that journal. The journal restriction matters: 8.7% of records typed as articles
have a repository rather than a journal as their primary location, and counting those would
attribute preprints to whoever hosts the server.
Publisher unit. Each journal's owning group, resolved through the OpenAlex
publisher hierarchy to its root. Two corrections were needed first: OpenAlex orders its
lineage array self-first and root-last, and 449 publishers carry a null hierarchy
level that a naive root filter silently drops. Eleven further cases are fixed in a documented
overrides file — IEEE is split across five separate entities, Oxford University Press is filed
under the University of Oxford, and one small Japanese society is credited with 3.3 million
articles it did not publish.
Truncation. The OpenAlex API returns at most 200 groups per query. Unseen publishers are all smaller than the smallest one visible, which bounds their total possible contribution to the index at under 4.0 points across every cell in the analysis. Concentration ratios and the big-five and big-seven shares are exact, since they use the true denominator and the largest publishers are always visible.
Verification. Every headline number was recomputed by an independent path — 108 assertions checked against raw cached API responses rather than the analysis pipeline, with no mismatches. Four publisher-year counts were confirmed against a different API endpoint. List prices were spot-checked against publicly posted values.
Financial figures are drawn from the companion piece on the industry's accounts, which generates them from company results releases; they are reproduced here rather than re-derived.