When will AI be creating all new drugs?

Posted: June 17, 2025

Five years ago, a drug designed by AI reached the clinical trial stage for the first time. The drug, designed to be an OCD medication, was a collaboration between British startup Exscientia and a Japanese pharma company, Sumitomo Pharma. Its creators noted that using AI significantly sped up the development process, resulting in a drug that got to clinical trials five times quicker than usual. Unfortunately, that drug didn’t make it out of phase 1 clinical trials. (In AI’s defense, nor do 90% of compounds tested in phase 1 trials.) But in the intervening five years, how much has AI changed drug discovery?

How can AI be used in drug discovery?

Traditional drug discovery is a time-consuming process, often involving trial and error experimentation and large-scale testing. Typically, researchers will identify a molecular mechanism or pathway linked to a disease and attempt to identify its “therapeutic targets”—the biological components (such as DNA, protein or bacteria) that might react to a medicine. Researchers then start looking for a compound that can modify the target to alleviate disease symptoms.^[1]

This process can be slow, expensive and prone to dead ends. It takes an average of 12 to 15 years to go from discovering a new drug to achieving regulatory approval, and estimates show it takes anywhere from $750 million to $2.5 billion to bring a new drug to market once accounting for the cost of failures.^[2]

AI’s ability to find patterns in data means it could potentially improve the efficiency, accuracy, and speed of the drug discovery process. It can identify targets, discover potential treatments and even create and optimize the molecular structure of potential drugs. AI models can, for example, predict how well a drug will bind to its target, forecast the efficacy and toxicity of drug compounds, or suggest new applications for old drugs.^[3]

Gathering biological data for gen-AI

Of course, as with all AI, its output is only as good as the data it’s working with. Biotech AI-models use a variety of biological data, such as gene sequences, blood biomarkers, protein structures, and clinical data.^[4] Let’s look at how some of these businesses are gathering that all-important data.

Absci, a US-based generative-AI drug creation company, uses its own 77,000+ sq ft experimental lab to generate biological training data through synthetic biology technology. The company develops new antibodies to treat diseases, and uses AI to predict which antibodies will be most effective. These predictions are then tested in its lab. Its process recently identified an antibody that promises to be effective against irritable bowel disease, which moved from conception to phase 1 human trials in just two years.

Recursion, a US clinical stage techbio company (techbio and biotech are two different things—further reading), has an automated lab that runs up to 2.2 million experiments a week. Using robotics and computer vision, its experiments use 50 different human cell types, and feature millions of compounds and genetic modifications. Each experiment produces a high-resolution image that captures detailed cellular morphology and features, creating mountains of data. The company also gathers data from always-on video streams of its animal labs, analyzing drug-induced behavioral change.^[5]

How AlphaFold revolutionized protein structure prediction

Last year, Google DeepMind’s Sir Demis Hassabis and Dr John Jumper were awarded the Nobel Prize in Chemistry for developing AlphaFold, a groundbreaking AI system that predicts the 3D structure of proteins from amino acid sequences.^[6] The system provides one example of how AI can help us discover new medicines.

AlphaFold’s predictions are freely available, and have been used by more than 2 million scientists and researchers since its launch in 2021. The system was trained on two data types: amino-acid sequences and 3D descriptions of their shapes. Using pattern recognition, AlphaFold can use amino-acid sequences to predict the protein shape.

Google DeepMind spin-out Isomorphic Labs builds on the AlphaFold system. Isomorphic claims its AI models can work across multiple therapeutic areas and drug types, reducing the need for time-consuming experimental lab work. Its AI drug design engine operates end-to-end, allowing its researchers to test hypotheses and predict the way molecules will interact with proteins, bind to proteins, and behave in the body. Researchers can use models to explore drug properties such as solubility and permeability. In January, CEO Sir Demis Hassabis said the company plans to have drugs in clinical trials by the end of the year.^[7]

Firsts in AI-discovered drug development

Global biotech company Insilico’s AI-designed drug for idiopathic pulmonary fibrosis, a progressive lung disease, is the first drug where both the target and compound were discovered using generative AI.

Insilico’s researchers used AI to analyze datasets and identify a promising target for idiopathic pulmonary fibrosis treatment, before using a chemistry AI engine to design and optimize compounds that would interact with the target. The whole process took just 18 months.

Recently named Rentosertib by USAN, the U.S. council responsible for assigning generic drug names, the drug has achieved positive clinical trial results. The company is now working with global regulatory authorities to set up larger trials.  

Exscientia and Recursion: Merging AI and biotech expertise

Meanwhile, in November last year, Exscientia announced a £500 million merger with Recursion. The merger combines Exscientia’s chemical design, synthesis methods, and more than 60 petabytes of proprietary data with Recursion’s Operating System, which continuously expands one of the world’s largest proprietary biological, chemical and patient-centric datasets.

Recursion believes the merger will enable it to create more effective medicines for patients, faster. David Hallett, Ph.D, Chief Scientific Officer at Recursion said "With our combined strength of real-world proprietary data and the models we've created–hypothesizing, testing, and learning in a continuous loop–we're redefining the space by shrinking timelines and costs, identifying and optimizing lead candidates faster than traditional methods."

Its combined drug pipeline has several drugs in phase 1 clinical trials, including drugs for lymphoma, small-cell lung cancer, and a rare condition called Familial Adenomatous Polyposis.

What’s next for the AI drug industry?

Last year, the McKinsey Global Institute estimated that generative AI could create $28 to $53 billion a year in value through research, discovery, and clinical development activities in the life sciences.^[8]

While AI has certainly sped up getting new compounds to trial, there is not yet an AI-developed medicine on the market. A recent research paper investigating the success of AI-developed drugs identified 114 “AI-native” biotech companies pursuing drug development.^[9] The researchers found that AI-discovered molecules are substantially more successful in phase 1 trials than historic industry averages, with an 80–90% success rate. The limited data for phase 2 trials showed that the molecules had a similar success rate to industry averages of around 40%.

Between 2015 and 2024, 75 AI-developed drugs entered clinical trials, with the number under investigation increasing exponentially each year.^[10] It seems like it’s only a matter of time before the first AI-invented medicines are treating patients. In a few years’ time, AI is likely to have helped us create innovative medicines that reach patients more quickly than ever before.

^{[1] Drug discovery and development: introduction to the general public and patient groups
[2] Unlocking the potential of AI in drug discovery
[3] A new paradigm for drug development
[4] Artificial intelligence is taking over drug development
[5] Will AI ever cure cancer? The multibillion-dollar race to bring the first AI-discovered drug to market
[6] Nobel Prize in Chemistry 2024 press release
[7] AI-designed drugs in trials this year, says Google DeepMind chief
[8] Generative AI in the pharmaceutical industry: Moving from hype to reality
[9][10] How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons}