The AI Drug Discovery Landscape

AI drug discovery has evolved from a speculative concept to a major sector of the pharmaceutical industry. Billions of dollars in venture capital and pharma partnership deals have flowed into the space, dozens of AI-discovered molecules have entered clinical trials, and every major pharmaceutical company now has an AI strategy. This article maps the key players, partnerships, and milestones shaping the field.

Market Overview and Investment Trends

Investment in AI-driven drug discovery grew dramatically through the early 2020s. Total funding for AI drug discovery companies has reached billions of dollars annually, spanning venture capital, IPOs, and partnership deals. Notable funding rounds include Xaira Therapeutics' launch in 2024 with over $1 billion, Recursion Pharmaceuticals' public market capitalization and subsequent acquisition of Exscientia, and continued investment in companies like Insilico Medicine, Absci, and Generate:Biomedicines.

However, the sector has also experienced volatility. Several AI drug discovery companies saw significant stock price declines from their IPO peaks, and the market has become more discerning about which companies have genuine technological differentiation versus those that primarily repackaged standard computational chemistry with an AI label.

Key Company Categories

The AI drug discovery landscape can be roughly organized into several categories based on primary focus:

Structure prediction and molecular modeling: Companies and groups focused on predicting protein structures and molecular interactions. This space is shaped by academic and quasi-academic efforts (AlphaFold from DeepMind, RoseTTAFold from the Baker lab, Boltz from MIT) alongside commercial platforms like Schrodinger, which combines physics-based modeling with machine learning.

Generative chemistry: Companies that use AI to design novel small molecules. Insilico Medicine (Chemistry42 platform), Recursion (following its acquisition of generative chemistry capabilities), and Relay Therapeutics are prominent players. These platforms generate and optimize molecular structures for target binding, selectivity, and drug-like properties.

Biologics and protein design: Absci, Generate:Biomedicines, and the Baker lab's Institute for Protein Design (whose spinouts include Monod Bio and others) focus on AI-driven design of therapeutic proteins, antibodies, and other biologics.

Phenomics and cellular biology: Recursion Pharmaceuticals has built one of the largest biological datasets in the world through automated high-content imaging of cells treated with genetic and chemical perturbations. Their platform uses computer vision and large-scale machine learning to identify drug targets and candidate molecules from cellular phenotypic data.

Multi-omics target discovery: BenevolentAI, Insilico Medicine (PandaOmics), and others use knowledge graphs and multi-omics analysis for target identification, as covered in our target identification article.

Major Pharma Partnerships

Partnerships between AI drug discovery companies and large pharmaceutical companies have been a defining feature of the landscape, providing both validation and funding. Notable deals include:

• Recursion and Roche/Genentech — a multi-year partnership focused on neuroscience and oncology
• Insilico Medicine and Sanofi — a partnership focused on using Insilico's AI platform for target discovery and molecule design
• Absci — applying their generative AI and wet-lab platform to antibody discovery programs through multiple pharma partnerships
• Eli Lilly — has formed multiple AI partnerships and built significant internal AI capabilities
• Novartis — has invested in AI across drug discovery, including partnerships with Isomorphic Labs (Alphabet's drug discovery company)

These deals typically involve upfront payments, research funding, and milestone payments tied to preclinical and clinical progress. The large headline values in press releases represent maximum potential payouts if all milestones are achieved — actual realized values will be lower.

Clinical Milestones

The ultimate test of AI drug discovery is whether it produces medicines that work in patients. Several programs have reached clinical trials:

Insilico Medicine's INS018_055 is a small molecule inhibitor targeting TNIK (Traf2- and Nck-interacting protein kinase) for idiopathic pulmonary fibrosis (IPF). Both the target and the molecule were identified using AI (PandaOmics for target discovery, Chemistry42 for molecule generation). The compound advanced to Phase 2 clinical trials, making it one of the most advanced fully AI-discovered and AI-designed drug candidates.

Recursion Pharmaceuticals has built a portfolio of clinical-stage programs across oncology and rare diseases. Their approach uses phenomics-based screening to identify drug candidates, and several compounds have entered clinical trials.

Relay Therapeutics focuses on using motion-based analysis of protein dynamics (combining cryo-EM, molecular dynamics simulations, and machine learning) to design precision medicines. Their lead program, RLY-4008 (lirafugratinib), a selective FGFR2 inhibitor for cholangiocarcinoma, received FDA Breakthrough Therapy designation.

It is important to note that "AI-discovered" encompasses a spectrum — from programs where AI played a minor supporting role to those where AI was central to both target and molecule selection. The industry has not yet converged on standards for what qualifies as an AI-discovered drug.

Open-Source vs. Proprietary Models

A significant tension in the field is between open-source and proprietary approaches. Key structure prediction models (AlphaFold2, RoseTTAFold, ESMFold, OpenFold, Boltz-1) are open-source, enabling broad academic and commercial adoption. Many molecular property prediction tools (Chemprop, RDKit, DeepChem) are also open-source. However, AlphaFold3's code was initially released under restricted terms, sparking debate about the future of openness in the field.

Companies like Recursion and Insilico argue that proprietary data (biological datasets, experimental results, clinical data) is more valuable than proprietary algorithms — the models are important, but the data they are trained on is the real competitive advantage. This perspective aligns with the broader AI industry trend where access to high-quality training data differentiates performance more than model architecture alone.

What Comes Next

The field is at an inflection point. The question is shifting from "can AI design drug candidates?" (it can) to "can AI-designed drugs succeed in the clinic at higher rates and lower cost than traditional approaches?" This will be answered over the next several years as current clinical programs mature.

Key trends to watch include: the integration of foundation models into end-to-end drug discovery workflows; the convergence of AI with automated wet-lab platforms (self-driving labs); expanding applications to complex modalities like cell and gene therapy; and the increasing importance of regulatory science — how agencies like the FDA will evaluate AI's role in drug development submissions. The companies that succeed will likely be those that combine strong AI capabilities with robust experimental validation, deep biological expertise, and the clinical development infrastructure needed to bring a drug to patients.