- No AI-designed drug has received regulatory approval yet.
- Over 173 AI drug programs are now in clinical trials.
- Phase II success rates match traditional drugs despite early-stage gains.
Alex Zhavoronkov spent a decade arguing that artificial intelligence could design drugs from scratch. In June 2025, his company Insilico Medicine published results in Nature Medicine that suggested he might be right. The molecule, rentosertib, had been designed by AI to treat idiopathic pulmonary fibrosis - a lung condition that slowly destroys a patient's ability to breathe.
Seventy-one patients were treated, the signals were positive. The results appeared in a peer-reviewed journal.
It was, by any reasonable measure, a milestone. And it still has not received FDA approval.
The story of AI drug discovery sits in that gap between genuine achievement and hard biological reality. What follows are five numbers that cut through the noise, and one pattern that connects them.
1. 173 AI-designed drug programs are now in clinical trials
In 2016, there were three. By 2023, sixty-seven. Now the pipeline stands at over 173 programs in clinical development, according to Drug Target Review's year-end analysis. Major pharmaceutical companies, Eli Lilly, Takeda, Novartis among them, are building dedicated AI infrastructure.
Dr. Raminderpal Singh, a drug discovery analyst writing in Drug Target Review, called it "progress without hyperbole." The phrase is apt. Pharmaceutical enthusiasm tends toward the opposite.
What counts as an "AI-designed" drug?
There is no agreed definition. Some molecules were identified by AI from existing compound libraries. Others were generated entirely by machine learning models. Still others used AI only to optimize a human-selected candidate. The 173 figure spans all these approaches, which makes direct comparison difficult.
The numbers are substantial. But zero AI-designed drugs have received regulatory approval. The pipeline is growing, but the finish line has not moved closer.
2. Phase I success runs at 80-90%, but Phase II tells a different story
That gap narrows sharply in Phase II, where AI-backed candidates succeed around 40% of the time, comparable to the traditional rate of 29-40%.
Key figure
80-90%
Phase I success rate for AI-discovered molecules, versus roughly 52% for traditionally developed drugs.
The sample sizes remain small and likely skewed. Early AI programs may have selected easier targets, a pattern familiar from the early days of monoclonal antibody development, when initial success rates reflected careful target selection rather than technological superiority.
The question is whether that early-stage advantage compounds over time, or whether it simply front-loads the same attrition. Biology, as every drug developer learns eventually, does not care how cleverly you found your molecule.
AI drug discovery is being watched by a lot of players, but it is still in the experimental stages. (Science Reader)
3. $15 billion in deals, 2% paid upfront
In 2025, AI drug discovery partnerships exceeded $15 billion in so-called "biobucks", the total potential value including milestone payments. The actual upfront payments averaged around 2% of those headline figures.
Five of the ten largest licensing deals originated from Chinese companies, including a $6 billion agreement between Dovetree and XtalPi.
Everyone acknowledges the funding environment has dried up.
Scott Schoenhaus, senior biotech analyst at KeyBanc Capital Markets
The deal structures are standard for the pharmaceutical industry, not unique to AI. But the ratio (2% real money against 98% conditional) illustrates the distance between announcement and reality.
Fortune traced this gap in detail to structural and economic barriers rather than technology limits. The algorithms work, and the business model is still forming.
4. Multiple companies have restructured or failed
BenevolentAI, once valued at over $1.5 billion, delisted from the stock exchange. Recursion, which had positioned itself as the AI-native drug company, shelved three programs. Several competitors cut 20-30% of their workforce.
These are not anomalies in biotech, where roughly nine in ten drugs fail and companies routinely collapse. But the timing stings. AI already helps analyze DNA and discover materials. This was supposed to be the moment AI arrived in medicine.
This was supposed to be the moment AI arrived in medicine.
The funding drought is part of the story. So is a harder truth, as Ardigen's analysis argues: building a drug company around AI prediction still requires navigating biology, regulation, manufacturing, and clinical trials. Zhavoronkov himself has been characteristically direct about this. The algorithm identifies a candidate in months, but getting it into a patient takes years, costs hundreds of millions, and depends on factors no model can predict.
5. The first approval could come in 2026 or 2027
Rentosertib needs a larger Phase III trial before regulators will consider it. The FDA has begun issuing guidance on AI in drug development. The EU AI Act, which the World Economic Forum notes classifies medical AI as high-risk, takes effect in August 2026.
More on AI discovery
AI Materials Discovery: 5 Things to Know
The gap between what AI predicts and what chemists can actually make has become the central drama of materials science.
→If an AI-designed drug reaches approval in the next two years, it will validate a decade of investment and reshape how pharmaceutical companies allocate research budgets. Drug Target Review's 2026 predictions describe the coming year's clinical readouts as decisive.
Zhavoronkov's molecule is the closest anyone has come to realize AI drug discovery so far. Seventy-one patients in, no approval yet, and the next trial will be larger, longer, and far more expensive.
The technology has moved from speculation to clinical evidence. Whether that evidence survives Phase III is the question the whole field is watching.
Sources
- Primary Research: AI in Drug Discovery: 2025 in Review (Drug Target Review, 2025)
- Additional Context:
- AI in Drug Discovery: Predictions for 2026 (Drug Target Review)
- The AI Drug Breakthrough Is Taking a Long Time to Arrive (Fortune)
- AI in Biotech: Lessons from 2025 (Ardigen)
- How AI Is Reshaping Drug Discovery (World Economic Forum)
Fact Check: Claim-by-Claim Verification Verified
The article accurately reports clinical trial numbers, success rates, financial deals, company challenges, and the absence of approved AI-designed drugs as confirmed by primary sources.
Commentary
- Article appropriately notes lack of standard definition for "AI-designed," spanning various AI involvement levels.
- Early AI programs may target easier indications, potentially skewing success rates; larger Phase III data needed.
- Chinese companies prominent in large deals, adding geographic diversity to investments.
Sources used for verification
Academic/Peer-reviewed:
- Rentosertib Phase IIa results - PubMed
- Nature Medicine publication - Nature
Other reliable sources:
- AI in drug discovery: 2025 in review - drugtargetreview.com
- FDA AI Guidance - fda.gov
- How AI is reshaping drug discovery - weforum.org
Fact-checked by Perplexity Sonar Pro on 2026-03-03
