Artificial intelligence accelerates drug discovery and could halve costs

3–5 minutes

In just 18 months, Insilico Medicine progressed from identifying a protein involved in pulmonary fibrosis to selecting a candidate molecule. Artificial intelligence guided both the choice of target—the TNIK protein—and the design of a compound capable of blocking it, named rentosertib. The drug is already in phase 3 clinical trials, expected to include 320 patients and last 52 weeks.

Why is a drug so expensive before it reaches the market?

The cost of an approved drug includes investments made in projects that failed along the way. The longer a program takes to be discontinued, the greater the accumulated expenditure. A study led by Olivier Wouters and published in JAMA estimated a median cost of about US$ 985 million per approved drug in the United States between 2009 and 2018. The overall average, including unsuccessful projects, reached US$ 2.67 billion in 2025, according to Deloitte, which tracks 20 major biopharmaceutical companies.

Where is AI already shortening research time?

Artificial intelligence helps determine what needs to be modified in the body and predict protein structure. In 2022, AlphaFold, developed by Google DeepMind, compiled more than 200 million computer-predicted structures, work that earned Demis Hassabis and John Jumper part of the 2024 Nobel Prize in Chemistry. With these predictions, researchers can compare molecules virtually, propose changes, and focus physical experiments on the most promising options.

In the case of rentosertib, the 18 months of work covered everything from the search for a biological target to the selection of a preclinical candidate. The same process, from start to first human trials, was completed in less than 30 months.

The bottleneck of clinical trials

Phase 1 assesses safety and tolerability, while phase 2 introduces patients with the disease to check for signs of clinical benefit. Only 28.9% of programs that reach phase 2 progress to phase 3, according to BIO data from 2011 to 2020. Researchers at the Boston Consulting Group observed success rates of 80% to 90% in phase 1 and about 40% in phase 2 for molecules analyzed with AI support.

Results of rentosertib

A phase 2a study with 71 participants, published in Nature Medicine, showed that the group receiving the highest dose of rentosertib had an average increase of 98.4 milliliters in forced vital capacity over 12 weeks, while the placebo group lost 20.3 milliliters. In July 2026, Insilico initiated a phase 3 trial to confirm these results in a larger sample.

How to halve costs

Kinea Investimentos simulated three scenarios based on historical approval rates. In the most favorable scenario, where 40% of candidates advance from phase 1 to phase 2, the final approval rate could reach about 18% and the cost per approved drug would fall by more than half.

Who can profit before the first drug reaches the market?

Major investments are already underway. Eli Lilly and Nvidia have committed up to US$ 1 billion over five years to a joint laboratory. In May 2026, Isomorphic Labs raised US$ 2.1 billion in a round led by Thrive Capital. Novo Nordisk has also partnered with Anthropic to apply AI in the discovery of new drugs.

Kinea estimates that about US$ 230 billion in drug sales will lose patent protection by 2030, creating pressure for pharmaceutical companies to renew their portfolios. Providers of clinical research services, equipment, and reagents may capture some of this investment before new drugs reach patients.

If more molecules can progress through clinical trials, the traditional model of drug development, which measures the path to the patient in decades, may begin to be reformulated.

Artificial intelligence does not yet eliminate all failures, but it is making the initial discovery stage more efficient, reducing the number of projects that reach the most expensive phases and increasing the chances that drugs that reach the market are truly effective.

—

Frequently Asked Questions

How much does it cost to develop a new drug?

Deloitte estimated an average cost of US$ 2.67 billion from discovery to launch in 2025, based on 20 major biopharmaceutical companies. Other studies use different methodologies and arrive at different values.

At what stage does artificial intelligence already help save time?

In the early stages of research, AI assists in the selection of biological targets, the prediction of protein structures, and the prioritization of molecules, allowing physical experiments to focus on a smaller set of promising candidates.

What is the difference between phases 2 and 3 of the trials?

Phase 2 seeks initial signs of benefit and continues to assess safety, while phase 3 involves more patients to confirm efficacy and collect more data on adverse effects.

Is rentosertib already available to patients?

Rentosertib is still in the experimental phase. Insilico initiated a phase 3 trial in July 2026 to assess its efficacy and safety in patients with idiopathic pulmonary fibrosis over 52 weeks.