Scalable Drug Discovery: Insilico Medicine

Most AI companies in healthcare promise efficiency. Insilico Medicine is aiming at a new way to originate medicines. The breakthrough is not that AI helps scientists work faster. It is that Insilico is trying to turn drug discovery itself into a repeatable, scalable engine. Its core bet is that target discovery, molecule design, and even parts of clinical prediction can be turned into a unified AI system, shrinking a process that traditionally takes years into something closer to an engineered workflow.

How It Started: From Software to Biotech

Insilico did not begin as a classic biotech company hunting one molecule for one disease. It started with Alex Zhavoronkov’s belief that large-scale biological data could be mined with AI to understand aging, disease mechanisms, and eventually therapeutics. The company says it was founded in 2014 at the Emerging Technology Centers of Johns Hopkins University, and its early work focused on applying AI to biological data rather than building a traditional wet-lab drug company from day one.

That original orientation matters. Most biotech companies begin with a biological hypothesis and then build a company around it. Insilico came from the other direction. It built tools first, then realized that the highest-value opportunity was not selling software alone, but using that software to originate and advance its own pipeline. Its internal case study explicitly describes a pivot from being mainly a software provider into a full-stack biotech company, and that move explains much of what makes Insilico strategically unusual today.

The human frustration behind the company is familiar to anyone who has watched drug R&D up close. Drug discovery is slow, expensive, probabilistic, and full of dead ends. Insilico’s answer was not to digitize one step. It was to ask whether AI could connect biology, chemistry, and development into one learning system. That is a much bigger ambition than finding a few leads faster.

The AI-Driven Innovation

What is fundamentally new about Insilico is not that it uses AI in pharma. Many companies do. The real novelty is its attempt to industrialize the entire front end of drug creation, from identifying a target to generating a candidate molecule to forecasting which programs are worth advancing. Its Pharma.AI stack spans target discovery, chemistry generation, biologics design, and clinical-trial prediction rather than treating these as isolated tools.

That matters because conventional biotech is still heavily artisanal. A lot of value depends on expert intuition, slow cycles of hypothesis formation, and serial experimentation. Insilico is trying to remove that constraint. In its own disclosures, the company says it nominated its IPF candidate ISM001-055 in under 18 months, versus an industry average of roughly 4.5 years, and moved an AI-discovered, AI-designed antifibrotic program from target identification to Phase I in 30 months. That does not eliminate biology risk, but it changes the speed and scale at which plausible programs can be generated.

The breakthrough, then, is not simply faster molecule screening. It is the creation of a reusable discovery engine. PandaOmics helps identify and rank targets from multi-omics and disease data. Chemistry42 generates and optimizes novel small molecules. InClinico predicts clinical trial outcomes. Generative Biologics extends the same logic into peptide and antibody design. Taken together, the company is building a system meant to produce multiple assets repeatedly, not a one-off moonshot.

AI is the enabler because this kind of integration was simply not realistic a few years ago. The recent stack includes deep generative models, transformers, multimodal chemistry models, reinforcement learning, structure-based design enhanced by AlphaFold-era protein prediction, and large-scale compute. Insilico’s own papers and platform descriptions show how these methods are now linked across biology and chemistry, including target nomination, de novo molecule generation, and property optimization.

This is why Insilico feels more like a market-creating shift than a productivity feature. The company is not just helping existing drug hunters do the same work a bit faster. It is trying to create a world in which drug discovery behaves more like a scalable computational platform, capable of producing many therapeutic programs across fibrosis, oncology, immunology, metabolic disease, and pain. That is a different business logic.

The Underlying Technology

Under the hood, Insilico’s platform is modular. PandaOmics is the biology engine for target discovery. Chemistry42 is the generative chemistry engine for designing small molecules with desired properties. InClinico forecasts clinical trial outcomes. The company has also added Generative Biologics for peptide and antibody work, DORA for scientific research assistance, and Nach01, a multimodal chemistry foundation model trained on billions of molecular and textual data points.

What makes this hard to replicate is not any single model. It is the accumulation of data, workflow integration, and clinical feedback loops. Insilico now has publications describing PandaOmics, Chemistry42, and its clinical prediction stack, plus a proprietary pipeline that feeds real-world validation back into the platform. In 2025, the company said its software served 13 of the top 20 global pharma companies by 2024 sales, while its broader platform had been used by more than 40 pharmaceutical companies. That kind of deployment creates commercial learning and data flywheels that are difficult for a newcomer to compress.

One level deeper technically, Insilico’s differentiator is orchestration. Plenty of firms can run AI against chemistry or protein structure. Fewer have shown a linked system that starts with disease biology, narrows to commercially tractable and druggable targets, generates molecules against those targets, and then advances selected assets into human studies. The Nature Biotechnology and Nature Medicine papers around rentosertib matter because they provide one of the clearest end-to-end demonstrations yet that this integrated workflow can produce a clinically testable asset with early efficacy signals.

The New Market

The most interesting market Insilico is creating is not simply “AI for pharma.” That market already exists. The more important shift is toward drug-discovery-as-a-platform, where pharma companies can buy software, discovery services, co-development, or out-licensed assets from the same AI-native engine. In other words, Insilico is widening the set of buyers and users who can access frontier discovery capability without having to build the entire stack internally.

The noncustomers here are not consumers. They are organizations that historically could not assemble world-class discovery infrastructure on their own. Smaller biotechs, regional pharma companies, academic groups, and even large pharmas that lack coherent AI-first discovery architecture can now access target discovery engines, chemistry generation, or pipeline assets through collaboration. That is why the company’s software business and licensing model matter as much as its proprietary drugs.

This expands the market in two ways. First, it turns internal discovery know-how into an external product. Second, it creates new behavior inside pharma: instead of waiting for biology-led discovery to produce a scarce set of candidates, partners can treat discovery more like a pipeline of computationally originated options. In 2025, Insilico said it had cumulative collaboration value of $4.6 billion and had signed more than 10 new collaborations totaling $1.3 billion during the reporting period and shortly thereafter. That signals not just scientific interest but a new commercial buying pattern around AI-native discovery.

The Strategic Landscape

Insilico’s closest competitors sit in a few different buckets. Recursion and Exscientia, now combined after Recursion’s acquisition of Exscientia, are among the most direct comparables. Recursion’s strength has long been phenomics and large-scale experimental biology. Exscientia became known for AI-designed molecules and pharma partnerships. XtalPi blends AI with robotics and computational chemistry. BenevolentAI has leaned heavily into knowledge graphs and target identification. Insilico stands apart by pushing an end-to-end narrative especially hard, with its own target-discovery engine, chemistry engine, clinical prediction layer, software subscriptions, and a proprietary pipeline that now includes 28 nominated preclinical candidates and 10 programs in clinical trials.

Market & Traction

In its 2025 results, Insilico reported $56.24 million in total revenue, software revenue growth of 23.8% year over year, and subscription customer growth of 18.3%. It said its platform serves 13 of the top 20 global pharmaceutical companies by 2024 sales, had nominated 28 preclinical candidates in total, and had 10 programs in clinical trials.

The company also continues to attract serious capital and counterparties. In late March 2026, Eli Lilly expanded its relationship with Insilico in a deal worth up to $2.75 billion, including a $115 million upfront payment. Reuters said the agreement gives Lilly exclusive rights over certain preclinical oral candidates developed with Insilico’s AI engine.

The broader market tailwind is strong as well. Grand View Research estimated the AI-in-drug-discovery market at $2.35 billion in 2025, with a path to $13.77 billion by 2033. This is gaining traction now because big pharma needs more shots on goal, more efficient R&D, and a better way to navigate biological complexity. Insilico is positioned at that intersection.

The Honest Take

The most exciting thing about Insilico is not the rhetoric around AI. It is the clinical and commercial evidence beginning to stack up around a unified platform. The main risk is that biology is still biology. AI can compress search and improve decision quality, but it cannot repeal the brutal attrition of drug development. Wired noted in 2025 that no AI-designed drug had yet reached the market, even though firms like Insilico and Recursion had pushed candidates into Phase II. 

What’s Next

Next week, we may look at another AI-native company that is not just optimizing an old workflow but expanding what can be built in the first place. 

Send a company you think belongs in that category, or share this with a founder building where the market is headed, not where it has been.