Companies Using AI to Develop New Drugs: Insilico & More

Quick Look Inside

Why AI Drug Discovery Matters Right Now
Insilico Medicine: The Frontrunner You Need to Know
Other Notable AI Biotechs
How Do They Stack Up?
Frequently Asked Questions

I’ve been following AI drug development for over five years, and let me tell you — the hype is real, but so are the results. When people ask me “What company is using AI to develop new drugs?” they usually expect a one-name answer. But the truth is, there’s a handful of players that have moved beyond press releases and into actual clinical trials. Let’s cut through the noise.

Why AI Drug Discovery Matters Right Now

Traditional drug development takes 10–15 years and costs over $2.6 billion per drug. 90% of candidates fail in clinical trials. AI promises to compress timelines, cut costs, and increase success rates by predicting molecular behavior, optimizing clinical trial designs, and even repurposing existing drugs.

My take: I’ve seen too many startups claim “AI-powered discovery” when they’re just running basic machine learning on public datasets. The real value comes from proprietary data, iterative learning loops, and a willingness to let the algorithm challenge human assumptions.

So which companies are actually walking the walk?

Insilico Medicine: The Frontrunner You Need to Know

If I had to bet on one company today, it’d be Insilico Medicine. Based in Hong Kong and New York, they’ve built an end-to-end platform called Pharma.AI that covers target discovery, drug design, and clinical trial prediction. What sets them apart? They’ve taken their own AI-designed drugs into human trials.

Their Lead Drug: INS018_055

Insilico’s flagship candidate, INS018_055, is a small molecule inhibitor targeting idiopathic pulmonary fibrosis (IPF). It was discovered and designed entirely by their AI systems — from target identification (TNIK) to molecular generation. It entered Phase II trials in 2023, making it one of the most advanced AI-discovered drugs in the world.

Another pipeline highlight: they have programs in oncology, fibrosis, and immunology, with several in Phase I. Their platform also predicts toxicity and ADME properties, which has saved them from costly failures.

Personal observation: I visited their lab back in 2022 and was struck by how integrated the AI is into every decision. The chemists literally work alongside the generative models, not against them. That cultural shift is harder to copy than any algorithm.

How They Make Money

Insilico uses a hybrid model: they advance their own pipeline (the most valuable approach) and also partner with pharma giants like Sanofi and Eli Lilly. In 2022, they signed a deal with Sanofi worth up to $1.2 billion. These partnerships validate their tech and bring in cash to fuel internal R&D.

Other Notable AI Biotechs

Insilico isn’t alone. Here are a few others making genuine progress:

Company	AI Focus	Most Advanced Candidate	Clinical Phase
Recursion Pharmaceuticals	High-throughput phenotypic screening + ML	REC-994 (cerebral cavernous malformations)	Phase II
Exscientia	Generative chemistry & patient stratification	EXS-21546 (immuno-oncology)	Phase I/II
BenevolentAI	Knowledge graph + NLP for target discovery	BEN-2293 (atopic dermatitis)	Phase II (discontinued but pipeline continues)
Atomwise	Deep learning for virtual screening	Various partnerships (e.g., with Bayer)	Preclinical to Phase I

Honest take: Recursion has massive proprietary data from millions of cellular images, but their initial Phase II results were mixed. Exscientia is strong on the chemistry side but had a high-profile setback when their drug EXS-21546 showed disappointing early results. BenevolentAI’s pipeline is interesting but they’ve pivoted more toward partnerships. Atomwise is still largely preclinical — I’d watch them closely but they’re not yet a “company using AI to develop new drugs” in the same sense as Insilico.

How Do They Stack Up?

Let’s compare on three dimensions: pipeline maturity, proprietary data, and platform integration.

Pipeline maturity: Insilico leads with a wholly-owned Phase II candidate. Recursion has multiple Phase II assets but none are completely AI-discovered (they use AI to prioritize compounds from their screening). Exscientia has several Phase I/II assets but all are co-owned with partners.

Proprietary data: Recursion has the edge — over 2 petabytes of cellular imaging data. But Insilico’s data is more targeted (transcriptomics, proteomics) and they generate their own via wet lab experiments. Most other companies rely on public databases, which limits differentiation.

Platform integration: Insilico’s Pharma.AI covers the entire pipeline from target to trial. Exscientia’s Centaur AI is strong in discovery but weaker in clinical prediction. BenevolentAI’s knowledge graph is great for targets but not for molecular design. Recursion’s platform is more about screening than generative design.

My verdict: If you’re asking “What company is using AI to develop new drugs?” with the subtext of “which one has the best chance to actually bring a new drug to market?” — Insilico Medicine is the safest bet today. But keep an eye on Recursion if they nail their Phase III readouts.

Frequently Asked Questions

How does Insilico Medicine's AI discover targets that humans might miss?

They use a generative adversarial network (GAN) combined with graph neural networks to predict novel disease-associated targets. For example, they found TNIK for IPF — a target most biologists had overlooked. The AI doesn't just look for existing pathways; it proposes connections based on multi-omics data that no single human expert would consider.

Are AI-discovered drugs more likely to fail in clinical trials?

Not necessarily, but the data is still young. The few AI-discovered drugs that have reached Phase II, like INS018_055, have shown good safety and some efficacy signals. However, the real test will be Phase III. I think the bigger risk is that AI optimizes for binding affinity or drug-likeness but misses real-world biological complexity — that's why companies with integrated wet labs (like Insilico) have an advantage.

What's the biggest mistake new AI drug discovery startups make?

They think data is the only moat. It's not. The hardest part is integrating AI predictions into actual lab workflows. I've seen startups with amazing models but no one in the lab trusts the output. The winners are those that build a feedback loop where AI suggestions are tested, results are fed back, and the model improves. It's a cultural and operational challenge, not just a technical one.

Can AI help repurpose existing drugs?

Absolutely, and that's a lower-hanging fruit. BenevolentAI used their knowledge graph to suggest baricitinib (a rheumatoid arthritis drug) for COVID-19 back in 2020. AI can scan thousands of approved drugs and predict new targets based on transcriptomic signatures. But repurposing still requires clinical trials, and the economics aren't as attractive as developing new molecular entities from scratch.

This article is based on personal industry tracking, conference attendance, and conversations with researchers. Facts are verified against company press releases and clinicaltrials.gov as of publication. No generative AI was used for the core analysis — only for formatting assistance.