There is a test almost nobody applies to a market-size estimate: add up what the companies in that market say they sold, in filings they can be sued over, and see whether the total fits underneath the estimate.
We ran it across three therapeutic markets this month. In all three the arithmetic came back wrong, and in a specific direction — the disclosed revenue of a handful of products is larger than what several research houses say the entire global market is worth.
CAR-T: five products, three companies, more than the whole market
Worldwide net product revenue for approved CAR-T therapies, FY2025, as disclosed by their manufacturers:
| Product | Company | FY2025 (USD m) |
|---|---|---|
| Yescarta | Gilead | 1,500 |
| Tecartus | Gilead | 344 |
| Carvykti | Johnson & Johnson / Legend Biotech | 1,900 |
| Breyanzi | Bristol Myers Squibb | 1,358 |
| Abecma | Bristol Myers Squibb | 427 |
| Sum | 5,529 |
That is $5.53 billion, and it is a floor rather than a total: Novartis does not break out Kymriah in its FY2025 brand table, and Autolus's Aucatzyl was approved after the others. Both are positive numbers we could not add.
Against that floor, four published base-year market sizes:
| Published base | Disclosed FY2025 revenue as a multiple of it |
|---|---|
| Research and Markets (Dec 2025), $3.87B for 2024 | 1.43x |
| Coherent, $3.99B for 2025 | 1.39x |
| Global Market Insights, $4.3B for 2024 | 1.29x |
| Research and Markets (Jul 2025), $4.6B for 2024 | 1.20x |
The Coherent figure is for 2025 itself, so this cannot be dismissed as a stale base year. And the error compounds rather than staying put: a 30% growth rate applied to a base that is 1.3x too low does not give you a forecast that is 1.3x too low — it gives you one that is 1.3x too low and anchored to the wrong starting point, in a market where the published rates themselves span 14.6% to 43.3%.
A side casualty of the same arithmetic: the widely repeated claim that CD19 accounts for 86.5% of CAR-T revenue. The two BCMA products alone — Carvykti and Abecma — are $2.33B of the $5.53B, or 42%. The 86.5% figure is inconsistent with the filings and should not be repeated.
Read the CAR-T cell therapy report
GLP-1 receptor agonists: a 24x spread in a market with no definitional excuse
Published 2025 base-year estimates for GLP-1 receptor agonists run from $2.6 billion to $62.86 billion — a 24.2x spread. Usually a spread that wide is a definition problem: two houses counted different things and both are internally consistent. Not here. "GLP-1 receptor agonist" is a pharmacological class with a countable, approved, branded membership, and two companies hold essentially all of it — both of which report brand-level revenue in audited filings.
Novo Nordisk's FY2025 GLP-1 medicines total DKK 234,549 million, which at the DKK 6.62 per USD average rate disclosed in the same filing is $35.43 billion. Eli Lilly's tirzepatide brands — Mounjaro $22,965 million and Zepbound $13,542 million — add $36.51 billion. Two companies, $71.94 billion of disclosed GLP-1 revenue, excluding Trulicity's full year, the legacy exenatide-class products and all Chinese domestic semaglutide.
Three of five published market sizes are below that figure. One, at $2.6 billion, is smaller than a single quarter of Zepbound. Only one house's back-solved base — Grand View's implied $69.8 billion — reconciles with what the two sellers actually report.
This cuts the other way too, and we say so in the report: a 17.5% rate applied to the correct base implies roughly $189 billion by 2032, which is 2.63x the two firms' current GLP-1 revenue against Lilly's own $80-83 billion all-product guidance. The run-rate does not yet support it. A market can be larger than the estimates say and still be forecast too aggressively.
Read the GLP-1 receptor agonists report
Bispecific antibodies: two estimates below two products
Published 2025 estimates for bispecific antibodies spread 3.49x, from $8.93 billion to $31.17 billion. Hemlibra and Vabysmo — Roche's haemophilia and ophthalmology bispecifics — reported CHF 4,754 million and CHF 4,102 million respectively, which at the Swiss Federal Tax Administration's 0.83053 rate is $10.66 billion between them. Add $3.79 billion of disclosed oncology bispecific revenue and you reach $14.45 billion.
Two of the published estimates are below Hemlibra and Vabysmo alone. At the other end, $31.17 billion is 2.16x all the bispecific revenue we could find disclosed anywhere. One house — Astute, at $14.4 billion — lands within 0.3% of the bottom-up sum, which is the kind of agreement that should make a reader trust a number.
The structural finding underneath: 74% of disclosed bispecific revenue is non-oncology. Most published commentary on this market is about T-cell engagers in cancer. The money, today, is in haemophilia and the eye.
Read the bispecific antibodies report
Why this keeps happening
Three markets, three inversions, one mechanism. A top-down market model starts from a total and apportions it; a bottom-up sum starts from what was sold and adds it. When the two disagree, the sum is the harder number — it comes from documents with an auditor's name on them and a legal consequence for being wrong.
The published estimates are not fraudulent. They are mostly built from scope decisions made before the disclosures existed, then carried forward through editions, and almost never re-grounded against the filings that appear each quarter. The forecast rate gets revised; the base rarely does.
Which is why the base year is where we spend our time. Every headline in our reports is a named publisher's own figure, advanced at that publisher's own rate, and then divided through by something countable — product revenue, an approval register, a trial count, a certified-site count — before we publish it. When that division produces an impossible number, we say so, and we say which figure we are therefore not using.
Sources
- Gilead Sciences, FY2025 results, 10 February 2026 (Yescarta, Tecartus).
- Bristol Myers Squibb, Q4 2025 Form 8-K Exhibit 99.1 product revenue table (Breyanzi, Abecma).
- Johnson & Johnson, FY2025 reporting (Carvykti).
- Novo Nordisk, Q4 2025 Form 6-K sales tables and the DKK/USD average rate disclosed in the same filing.
- Eli Lilly and Company, Q4 2025 results release (Mounjaro, Zepbound).
- Roche, FY2025 reporting (Hemlibra, Vabysmo); Swiss Federal Tax Administration CHF/USD rate 0.83053.
- Published market estimates as cited in the individual reports, each with its publisher, date and stated scope.
Every figure above is reproduced from a named filing or a named published estimate; the full source lists, including what we examined and chose not to publish, are in the research notes behind each report.