What these drugs actually cost to make

The active ingredient in a month of semaglutide costs somewhere between one and twelve cents. That fact is real, it is peer-reviewed, and on its own it is close to useless. Here is the whole picture, including the parts that cut against the easy conclusion.

Start with the drugs where the answer is simple#

Before the GLP-1 argument, look at what happens to a drug after its patent ends. These are real prices US pharmacies actually paid on invoice, from the federal government's weekly acquisition-cost survey (2026-09-02 release, prices effective 2026-08-19) — not list prices, not estimates.

A month of finasteride costs a pharmacy $1.28. A month of generic sildenafil costs $3.83, while a single Viagra tablet costs $83.63. This is the system working as designed: exclusivity ends, competitors enter, and the price collapses toward the cost of making the thing. Hold onto it, because it is the control case for everything that follows.

Now the GLP-1s#

Same survey, same week. These are still under patent, and no generic exists in the United States:

What the published research says it could cost#

Two peer-reviewed teams have estimated production cost for semaglutide. We report both as a range and do not average them, because they measure slightly different things.

The gap between the two is mostly the injector. Fifty-two devices a year at $0.30 to $2.50 each is $16 to $130 on its own — which is why the newer paper's own headline finding is that the pen costs 8 to 68 times more than the drug inside it. Any version of this argument that stops at "the active ingredient costs pennies" has skipped the larger number.

Six of these seven drugs have no published estimate at all#

Who actually makes the ingredient#

This we can answer precisely, because the FDA publishes it. Every company supplying an active ingredient to the US market files a Drug Master File. We downloaded the current edition (FDA Drug Master File list, 2Q2026 edition (received through 2026-06-30)) and counted.

The generational split runs opposite to the usual story. For the mature small molecules — sildenafil, tadalafil, finasteride — India holds the clear majority of US ingredient filings and China is secondary. For the new peptides that flips hard: China holds roughly 79% of semaglutide filings and 84% of tirzepatide. India's generics industry, on GLP-1s, is largely a customer of Chinese ingredient rather than a competitor to it.

Minoxidil is the useful counterexample: four of its ten suppliers are Italian, clustered in Lombardy, and not one is Chinese. Offshoring is not automatic even for old, cheap molecules.

Across all 10,436 FDA-registered drug establishments, 2,504 declare active-ingredient manufacture. India leads with 562, the United States has 532, China 528. 78.8% sit outside the United States.

The number almost nobody uses: audited financials#

Modelled costs are arguable. A public company's audited accounts are not. From Eli Lilly's FY2025 annual report:

Lilly spent marginally more on marketing, selling and administration ($11.09B) than on manufacturing everything it sold ($11.05B). It spent 1.2 times as much on research as on manufacturing. Novo Nordisk's FY2025 gross margin was 81%, implying a similar cost of goods near 19% of revenue. These are audited and company-reported, and they are the least arguable figures on this page.

Why cost to manufacture is not a fair price#

This is the part the viral version of this argument leaves out, and leaving it out makes the argument wrong rather than merely incomplete. The gap between marginal manufacturing cost and price funds real things:

• Research, including everything that failed. The best-sourced public estimate puts the median capitalised cost of bringing one medicine to market at $985 million — a figure that already includes the failures. Lilly's research spending exceeded its cost of sales.
• The outcome trials. The studies that proved these drugs reduce cardiovascular events cost hundreds of millions each, and they are the reason insurers cover them at all.
• Capital built years before revenue. Novo committed roughly $6 billion to ingredient capacity at Kalundborg; Lilly around $9 billion at Lebanon, Indiana. Fermentation plants are not free.
• The device. Per the research above, the pen costs 8 to 68 times the drug.
• Regulatory, safety monitoring and liability, which continue for the life of the product.

There is also a serious published critique of this whole method, funded by the pharmaceutical industry federation — which we mention because the funding is material to how you weigh it, and because it lands real hits anyway. Testing the model against actual prices found it wrong in both directions: in India nearly half of medicines sold below the projected efficient price. These are order-of-magnitude tools, not price predictions.

Which brings it back to the first table. Sildenafil at thirteen cents a tablet and finasteride at four cents show that the system does eventually deliver near-cost pricing. The real story is the timing and the gatekeeping — how long the gap lasts and who goes without while it does — not that a gap exists.

Where the country split comes from, and its limit#

The DMF list carries six fields — DMF#, STATUS, TYPE, SUBMIT DATE, HOLDER, SUBJECT — and country is not one of them. The filing counts above are read from the file. The country split is our attribution from each holder's name and corporate domicile, and it is labelled as an attribution rather than a lookup.

It is unusually legible for this molecule: most holders are named for the city they operate in — Sichuan, Shenzhen, Hangzhou, Nanjing — or are national firms nobody needs a register to place, like Dr Reddy's, Biocon, Bachem or ScinoPharm Taiwan. That makes the split reliable enough to publish. It does not make it a field in a spreadsheet, and this page said it was until we opened the file.

The US-registered holders, named#

Three entities, for the two molecules that matter most here. That is the whole American supply side of the ingredient for these drugs, as filed.

Two of the three are the same company under two entity names. So for semaglutide, an active ingredient behind a market this size has 71 filing holders worldwide and 2 of them are American.

Check our work#

Every dataset behind this page is free and machine-readable, so here are the registers themselves rather than a promise that they exist. We would rather you re-ran the query than trusted us.

The Drug Master File list is the one to start with. Every company wanting to supply an active ingredient into the US market files one, so the filing counts are read straight from it: filter the SUBJECT column and count. DMFs received through 2026-06-30, current through DMF 044443 (as stated on FDA's own page). FDA republishes it quarterly under a new address, so the direct workbook link is edition-specific while the list page stays put.

Watch: the pricing story#

Independent third-party material we found genuinely useful. Embedding is not endorsement, we have no relationship with these creators, and any commercial conflicts we know of are labelled. Video players load only when you press play, though the still image comes from Google; a podcast embed loads its player when it scrolls into view. What that means for you.

▶Play on YouTube

GLP-1 Weight Loss: Side Effects, Cost and 1-Year Update

Mayo Clinic · Health Matters podcast, academic medical center · 2026-03-25

What a year of widespread GLP-1 use taught clinicians about side effects, cost, and durability.

▶Play on YouTube

Obesity Medicine in the Age of GLP-1s

The Commonwealth Fund · guest Fatima Cody Stanford, MD MPH MPA, Harvard/MGH obesity medicine · 2025-12-03

Access, insurance coverage, and how obesity medicine actually practices now.

Common questions#

Economic and regulatory journalism, not medical or investment advice. Production-cost estimates are modelled figures with stated assumptions, not observed prices, and every one is attributed above so you can weigh it yourself. Acquisition costs and company financials are observed data. Corrections land in our changelog.