The Economics of Drug Development
Why a single medicine costs billions and takes a decade, why most fail, and how AI and biology are rewriting the maths.
Bringing one drug to market costs on the order of $2 billion and takes over a decade, because the price of success must pay for thousands of failures. Drug development isn't expensive because chemistry is dear; it's expensive because most attempts don't work.
What this argues
- Attrition is the business~93% of drugs entering trials fail.
- Failure sets the priceOne approval funds thousands of dead ends.
- Productivity fell for decades'Eroom's Law': costs rose as odds fell.
- The model rewards blockbustersPatents, pricing and scale, not volume.
- AI and biology may bend the curveThe first credible reversal in decades.
What has changed
Updated November 2026
US drug pricing is being renegotiated deal by deal
Pharmacy Times (3 Feb 2026) counted 16 most-favoured-nation agreements with the administration, including Pfizer, AstraZeneca and Eli Lilly. Terms include price cuts of up to 85% on some drugs, direct-to-consumer sales and participation in the TrumpRx platform. Bloomberg Law reported Pfizer's deal brought a three-year grace period from pharmaceutical tariffs. This is the price-policy dial in the deck moving in practice.
Oral obesity pills reached the market
- The FDA approved Eli Lilly's oral GLP-1 orforglipron (Foundayo) around 1 April 2026 (BioSpace, 1 Apr 2026).
- Pricing: $25 a month with commercial coverage, $149 a month self-pay through LillyDirect.
- Novo Nordisk's oral Wegovy entered the US market in January 2026 and reached nearly 3,100 patients in its first week.
- Foundayo carries a boxed warning for thyroid C-cell tumours; no head-to-head trial against oral Wegovy exists.
An AI-discovered drug entered Phase III
Insilico Medicine dosed the first patient on 10 September 2026 in GENESIS-IPF-3, a randomised, placebo-controlled Phase III trial of rentosertib in idiopathic pulmonary fibrosis (about 320 patients, 47 centres in China, primary endpoint annual FVC decline). It remains investigational. Separately, Isomorphic Labs raised $2.1 billion in May 2026 and its first human trials were reported as expected by end-2026, slipped from an earlier target (press reports, Jan-May 2026; not confirmed from company sources).
FDA approval pace stayed steady
The FDA's 2026 novel approvals list showed 47 approvals through 20 October 2026 (latest: bepirovirsen, Hibsago), already above the 46 CDER approved in all of 2025 (RAPS, 13 Jan 2026). More than half of the 2025 approvals were for rare disease indications.
Obesity leadership consolidated around Lilly
Mounjaro earned $22.965 billion in 2025 versus $20.1 billion for Ozempic (BioSpace, Apr 2026). A Truist estimate cited in November 2025 put peak worldwide revenue for Lilly's Mounjaro, Zepbound and Foundayo franchise at $101 billion, an analyst assumption rather than a result.
Not refreshed: phase success rates and cost-per-stage figures, the patent-cliff revenue estimate, the obesity-market chart (analyst estimates), Eroom's Law data, biotech return statistics.
The Pipeline
From molecule to medicine
A decade-plus, stage by stage
- Discovery: screen for a promising molecule
- Preclinical: cells, then animals
- Phase I: first humans, safety
- Phase II/III: efficacy, then scale
- Approval: regulators judge; launch
The four stages every drug must survive
| Stage | What happens |
|---|---|
| 01 Discovery | Screen thousands of molecules for a promising hit. |
| 02 Preclinical | Test safety in cells and animals, years of work. |
| 03 Clinical trials | Three phases in humans: safety, then efficacy, then scale. |
| 04 Approval & launch | Regulators judge; manufacturing and marketing begin. |
Where the money and risk concentrate
| Phase | What it tests | Time and cost |
|---|---|---|
| Phase I | 20-100 healthy volunteers. Is it safe? | ~1-2 years, ~$60M |
| Phase II | 100s of patients. Does it work, and at what dose? The great killer. | |
| Phase III | 1,000s of patients. Definitive proof. | ~$350M-$1B, years long |
Of 10,000 compounds, one reaches patients
View the data
| Compounds | |
|---|---|
| Compounds screened | 10000 |
| Preclinical | 250 |
| Phase I | 20 |
| Phase II | 10 |
| Phase III | 6 |
| Approved | 1 |
Source: Industry data
6.7%: the odds a drug entering human trials ever gets approved
Down from ~10% a decade ago. Put differently: a candidate that has already survived years of preclinical work still has a ~93% chance of failing. Drug development is the highest-stakes filtering process in business.
Where most drugs and most money die
- Cheap, thousands of tries
- The first cull
- The graveyard: efficacy fails
- Costly, late deaths
- The survivors
The Economics of Failure
Why success costs billions
Each phase costs more than the last
View the data
| Cost ($M) | |
|---|---|
| Preclinical | 100 |
| Phase I | 60 |
| Phase II | 130 |
| Phase III | 350 |
Source: JAMA; industry estimates
The $2 billion isn't the cost of one drug, it's the cost of the nine that failed to produce it.
Direct cost of a single approval can be ~$700M. The rest is the graveyard it's built on.
How the numbers actually work
Source: RAND; Deloitte (2024-25).
Why time is the hidden enemy
| The patent | The consequence |
|---|---|
| Life: 20 years from filing | Recoup fast: high prices while protected |
| Filed early: before trials even start | Blockbuster or bust: must earn billions quickly |
| Left at launch: often <10 years | Cliff: revenue collapses at expiry |
| Then: generics erase the price | Refill: the pipeline, endlessly |
Phase II is where drugs go to die
View the data
| Success rate | |
|---|---|
| Phase I to II | 52 |
| Phase II to III | 29 |
| Phase III to NDA | 58 |
| NDA to approval | 91 |
Source: BIO Clinical Development Success Rates
The four failure modes, in order of cruelty
- Lack of efficacy: the drug simply doesn't work, the largest single killer (~40-50%).
- Safety / toxicity: it works, but harms, the second great cull.
- Commercial: it works and is safe, but can't earn back its cost.
- Strategic / operational: trials fail to recruit, or priorities shift.
Source: Nature Reviews Drug Discovery.
Eroom's Law
Moore's Law, backwards
Drug R&D got LESS productive for 60 years
A chart on a log scale shows drugs per $1B of R&D falling in a steady line from 1950 through 1970, 1990, 2010 and 2024.
Source: Industry analyses; log scale.
~9 years: the time it took for drug R&D productivity to HALVE, repeatedly
For six decades, each dollar of pharma research bought fewer drugs. That is the exact inverse of the semiconductor story, and why so much capital now bets on AI finally breaking the pattern.
The four forces behind Eroom's Law
| Force | Why |
|---|---|
| The 'better than the Beatles' problem | Each new drug must beat existing ones, a rising bar. |
| Low-hanging fruit gone | Easy targets were picked decades ago. |
| Regulatory caution | One thalidomide justifies ever-stricter trials. |
| Throw money at it | Big budgets funded waste, not just discovery. |
Where the industry is placing its bets
A two-axis map of scientific tractability against market size places therapeutic areas in four quadrants (small and easy, the gold rush, small and hard, big and brutal). Obesity / GLP-1, Immunology and Oncology sit in the top-right gold rush. Rare disease sits at the top-left, and Antibiotics near the centre line. Neuro / Alzheimer's sits at the bottom-right, big and brutal.
Source: Analysis of therapeutic areas by scientific tractability and market size.
The Business Model
How pharma actually makes money
~$200 billion of annual pharma revenue faces patent expiry this decade
The looming 'patent cliff': blockbusters from the 2010s losing protection, exposed to generics that cut prices ~80% overnight. Refilling that hole is the industry's single greatest strategic pressure.
Who does what in the drug economy
| Player | Role |
|---|---|
| Big pharma | Funds late trials, markets, and sells at scale. |
| Biotech | Does risky early science; sells or partners winners. |
| Academia & NIH | The basic research most drugs originate from. |
| CROs | Run the trials pharma outsources. |
| Payers & PBMs | Insurers and middlemen who set what's actually paid. |
How the winners pay for everything
| Lever | How it works |
|---|---|
| Blockbusters | A single >$1B/yr drug can carry a whole company. |
| Patent protection | Temporary monopoly to recoup and profit. |
| Price discrimination | The US subsidises the world's R&D via high prices. |
| Lifecycle management | Reformulate and defend before the cliff. |
| M&A | Big pharma buys pipelines it can't grow. |
| Biologics moat | Hard-to-copy drugs resist generic erosion. |
America is the world's drug-pricing engine: ~4% of the population funds roughly half of global pharmaceutical profits, and thus most of its R&D.
Every other country's cheaper drugs are, in effect, subsidised by the American patient. Since late 2025 the White House has struck most-favoured-nation pricing agreements with drugmakers (16 by February 2026), including Pfizer, AstraZeneca and Eli Lilly, trading price cuts and TrumpRx participation for tariff relief.
How the same pill is priced across borders
| United States | Rest of world | |
|---|---|---|
| Price | Highest on Earth | 50-80% lower |
| Negotiation | Fragmented, weak | Single-payer leverage |
| Role | Funds global R&D | Free-rides on US pricing |
| Politics | Perennial outrage | Access as a right |
What's Changing
Can the curve finally bend?
For sixty years, biology got harder to monetise. The bet of the 2020s is that AI and new biology finally make it get easier.
If Eroom's Law reverses, it would be one of the most valuable inflections in the economy.
What might reverse Eroom's Law
| Tool | What it does |
|---|---|
| AI drug discovery | Models now predict protein structure and drug candidates in silico. |
| GLP-1s | Obesity drugs, a genuine blockbuster category reborn. |
| Gene & cell therapy | One-time cures, not lifelong treatments. |
| mRNA platforms | COVID proved a programmable drug factory. |
| Real-world data | Trials augmented by health records. |
| Biomarkers | Targeting the patients a drug will actually help. |
AlphaFold: 200M+ protein structures predicted, a problem once measured one at a time
What took a PhD years now takes a model seconds. AI hasn't yet delivered a blockbuster from scratch, but it has demolished a bottleneck that constrained biology for fifty years. By September 2026 the first AI-discovered drug with a novel target, Insilico's rentosertib for pulmonary fibrosis, had entered a Phase III trial, though it remains investigational and unapproved.
One drug class is rewriting pharma's map
View the data
| Obesity-drug market | |
|---|---|
| 2021 | 3 |
| 2023 | 9 |
| 2025 | 30 |
| 2028E | 85 |
| 2030E | 130 |
Source: Compilation of analyst estimates · values read from the chart, approximate
Does AI actually change the economics?
| The optimist case | The skeptic case |
|---|---|
| Cheaper discovery: AI slashes the early stages | Trials still dominate: discovery is <30% of cost |
| Fewer failures: better targets, better patients | Biology is hard: prediction is not a working drug |
| Faster trials: digital and adaptive designs | Regulation unchanged: safety bar won't move |
| New biology: undruggable targets in reach | Hype cycle: we've heard this before |
Miraculous science, mediocre stocks
Source: Deloitte; BIO.
Risks & Debates
The honest chapter
Is pharma a broken model or a misunderstood one?
| The critique | The defence |
|---|---|
| Prices too high: Americans pay most | Failure is real: someone must fund the 93% |
| Innovation theatre: tweaks over breakthroughs | US funds the world: price controls cut R&D |
| Marketing > research: some firms spend more selling | Cures are permanent: value beyond the price tag |
| Access: cures the insured can't afford | Risk deserves reward: no profit, no pipeline |
Six risks a serious investor tracks
| Risk | What it means |
|---|---|
| Patent cliffs | $200B+ of revenue faces generic exposure this decade. |
| Price regulation | US reform could compress the profit engine. |
| Clinical failure | The base rate that never stops biting. |
| Pipeline gaps | Buying growth gets ever more expensive. |
| AI disappointment | A decade of promise, few approved drugs yet. |
| Political targeting | Drug prices are a perennial campaign issue. |
What to Watch
Signposts, not predictions
Four dials worth watching
| Dial | Signal | What it tells you |
|---|---|---|
| Phase success | Trial odds | Is attrition improving? |
| AI approvals | The proof | An AI-discovered drug on shelves |
| R&D per drug | Productivity | Does Eroom's Law bend? |
| Price policy | US reform | The profit-engine dial |
Five signposts for the next 24 months
- Does an AI-discovered drug get approved? The field's Sputnik moment.
- Do Phase II success rates rise? Where better targeting would show first.
- How hard do patent cliffs hit? $200B+ of revenue is exposed.
- Does US price reform advance? It would reset global R&D economics.
- Do GLP-1s keep expanding? The category rewriting pharma's map.
Reference
Glossary & sources
Plain-language glossary
| Term | Meaning |
|---|---|
| Preclinical | Testing in cells and animals before humans. |
| Phase I/II/III | Human trials: safety, efficacy, then scale. |
| Attrition | The rate at which candidate drugs fail. |
| Eroom's Law | The decades-long fall in R&D productivity. |
| Blockbuster | A drug earning over $1 billion a year. |
| Patent cliff | The revenue drop when protection expires. |
| Biologic | A drug made from living cells, hard to copy. |
| PBM | Pharmacy benefit manager, a pricing middleman. |
A new medicine is a lottery ticket that costs a decade and two billion dollars to buy, wins once in fifteen tries, and expires the moment it pays out. That it happens at all is a wonder; that it happens for profit is the model.
The five conclusions to take away
- Attrition is the whole story: the model is priced failure.
- Winners fund the graveyard: one approval carries thousands of deaths.
- Eroom's Law was the curse: sixty years of falling productivity.
- The US subsidises the world: high prices fund global R&D.
- AI attacks the wrong end (so far): discovery is cheap; trials are the prize.
Sources & further reading
- Deloitte: Measuring the return from pharmaceutical innovation (2024).
- RAND Corporation: drug development cost study (2025).
- BIO: Clinical Development Success Rates.
- JAMA Network Open: clinical trial costs analysis.
- Scannell et al.: 'Diagnosing the decline in pharmaceutical R&D efficiency' (Eroom's Law).
- Company disclosures: pharma R&D and pipeline data.
- Figures marked 'estimate' are compilations of the above.
Sources cited
Deloitte · BIO · RAND · JAMA · Nature Reviews Drug Discovery · Scannell et al. · Company disclosures · Industry data and analyses · Analyst estimates · FDA Novel Drug Approvals 2026 (Oct 2026) · RAPS, Jan 2026 · BioSpace, Apr 2026 · Insilico Medicine press release, Sep 2026 · Pharmacy Times, Feb 2026 · Bloomberg Law, 2025-26
Keep reading
Research, not investment advice. Figures as of , from the sources cited.
Legal information and disclosures
These are my own views as of the update shown on this page. They can change without notice, and I have no duty to update them. Past performance does not indicate future results, and any investment can lose money.
This is for education only. It is not investment, legal, tax, medical or accounting advice, and it is not an offer to sell or a solicitation to buy any security in any jurisdiction. It does not consider your goals or circumstances.
Figures from company filings, public agencies and other third parties are believed to be reliable, but I have not independently verified them and cannot guarantee them. Forecasts and anything marked as an estimate rest on assumptions that may prove wrong.
Please do not copy or republish this material without my written consent.