Drug Discovery & R&D

Outlook

Comparability caveat: high growth, low measurement consistency across publishers — some studies count only third-party AI software/services, others include discovery services, biosimulation, preclinical testing, infrastructure, and portions of clinical development.

For underwriting: the most conservative comparable set is Mordor Intelligence globally and Grand View Research for India; broader estimates are better read as an addressable-market ceiling. North America (not US specifically) was 44.9% of the global market in 2025 under MarketsandMarkets, or 52.8% under Grand View Research’s different methodology — these are North American, not US-specific, shares.

AI Maturity

Assessment: commercially proven in research workflows, clinically emerging, but not yet proven at pharmaceutical-approval scale.

This segment should not be assessed using medical-device-style FDA/CDSCO “AI clearances” — drug-discovery models are generally research tools, and regulatory approval attaches to the resulting drug, not the discovery algorithm. FDA’s Jan 2025 draft AI guidance focuses on AI-generated information supporting safety/effectiveness/quality decisions; discovery-stage and internal operational applications are outside its stated scope.

Evidence of scaled adoption:

• FDA’s CDER reported experience with 500+ submissions containing AI components (2016–2023) — spans nonclinical, clinical, post-marketing, manufacturing uses, so this is not “500 AI-discovered molecules”
• Insilico Medicine: first patient dosed in a Phase III trial of rentosertib, Sept 10, 2026 — a materially stronger validation point than a partnership/funding announcement, though the asset remains investigational, not approved
• Recursion: five clinical programs advancing at end of 2025; described results in familial adenomatous polyposis as first clinical validation of its operating system; received $134M across five accepted Sanofi program milestones (as of Feb 2026) — meaningful platform/translation signals, but don’t establish systematic Phase III success improvement

There is no “health systems live in production” metric here since primary buyers are pharma/biotech companies, CROs, and research institutions, not hospitals. Production evidence is better measured through recurring software revenue, repeat pharma collaborations, milestone acceptance, clinical-stage assets, and regulatory submissions — on those measures, the segment has moved beyond pilot-only adoption, but broad clinical/approval-level validation remains incomplete.

Key Players

Recursion Pharmaceuticals — combines automated wet-lab experiments, high-content cellular imaging, phenomics, ML, computational chemistry, clinical development in “Recursion OS”; differentiated by the feedback loop between proprietary experiments and model training, not a software-only design model. 2024 combination with Exscientia (all-stock merger, ~$688M valuation, Exscientia shareholders to own ~26% of combined business) added precision chemistry and automated small-molecule synthesis. Expected $100M+ annual synergies, ~$850M combined cash at Q2 2024. As of Feb 2026: $754M cash and equivalents, five advancing clinical programs, $134M received via five Sanofi milestones. Publicly listed, Nasdaq: RXRX.

Schrödinger — integrates physics-based molecular simulation with ML for protein-structure analysis, virtual screening, binding-affinity prediction, lead optimization, toxicity prediction; monetizes via licensed/hosted software, collaborative discovery programs, and an internal therapeutics pipeline. FY2025: $255.9M total revenue ($199.5M software, $56.4M drug-discovery), 74% software gross margin, but $103.3M net loss; ~$402.3M cash/restricted cash/marketable securities at year-end. Publicly listed, Nasdaq: SDGR, not PE-controlled.

Insilico Medicine — end-to-end platform: PandaOmics (target discovery), Chemistry42 (generative molecular design), inClinico (clinical-trial prediction); also runs an internal therapeutic pipeline, monetizing via software, collaborations, asset licensing, milestones, potential royalties. Strongest clinical maturity in the pure-play group — first Phase III patient dosed for rentosertib, Sept 10, 2026. 2021: $255M Series C led by Warburg Pincus. March 2026 Eli Lilly collaboration: $115M upfront + up to $2.63B contingent development/regulatory/commercial milestones — headline value is contingent, not contracted revenue. Publicly listed, Hong Kong Stock Exchange: 3696, IPO raised ~$293M (Dec 2025). Historical sponsors: Warburg Pincus, Qiming Venture Partners, OrbiMed, B Capital and others; post-IPO ownership percentages not available in reviewed sources.

Isomorphic Labs — AI-first drug-design company from the Google DeepMind ecosystem; developing IsoDDE, an integrated engine modeling biological interactions, identifying binding opportunities, designing molecules, advancing proprietary/partnered programs. May 12, 2026: $2.1B Series B led by Thrive Capital with Alphabet, GV, MGX, Temasek, CapitalG, UK Sovereign AI Fund — a scale-capital signal, not clinical validation; pipeline described as progressing “towards the clinic,” not an approved medicine. Privately held, institutionally backed; no investor ownership percentages disclosed.

Peptris Technologies (India) — AI-enabled preclinical drug discovery using unsupervised learning, generative AI, molecular-language models to generate novel molecules, predict physicochemical/target-interaction properties, identify repurposing opportunities, rescue clinically tested molecules; asset-licensing/co-development model, outsources wet-lab work to CRO/biotech partners. Feb 2026: ₹70 crore (~$7.7M) Series A co-led by IAN Alpha Fund and Speciale Invest with Tenacity Ventures, BYT Ventures. Private, VC-backed; no disclosed stakes, revenue figure, or clinical-stage drug identified — evidence base is preclinical and partnership-led, not approval-led.

Certara — downstream end of AI-enabled R&D: biosimulation, model-informed drug development, quantitative pharmacology, regulatory science; supports dose selection, PK/PD modeling, clinical-trial design, regulatory submissions. Not a pure generative-molecule-design company but highly relevant to the investable R&D-software layer. At EQT’s 2017 acquisition, served 1,200 commercial companies, 250 academic institutions, regulatory agencies across 60 countries; transaction valued Certara at $850M. Publicly listed, Nasdaq: CERT. Nov 2022: Arsenal Capital Partners invested $449M (~30M shares at $15/share), increasing diluted ownership from ~4% to ~22%.

Porter’s Five Forces

PE Firms, Stakes, and Strategic Vision

Warburg Pincus — Insilico Medicine — led $255M Series C, June 22, 2021, stake undisclosed; a Warburg Pincus managing director joined the board. Thesis: scale an end-to-end platform into a clinical-stage biotech company, creating value through software + partnered discovery programs + internally owned therapeutic assets — growth equity, not a traditional buyout.

Insight Partners — Dotmatics — investment began 2017, supported 14 strategic acquisitions per Insight’s account. 2021: Insight-backed Insightful Science agreed to acquire Dotmatics, combining into a cloud-first scientific R&D platform serving 1M+ scientists, generating $100M+ ARR (2021 purchase price and Insight’s precise stake undisclosed). Siemens subsequently agreed to acquire Dotmatics from Insight Partners for $5.1B on April 2, 2025. Thesis: buy-and-build across laboratory data management, bioinformatics, scientific applications; migrate users to an integrated cloud platform; position the asset as the data/workflow layer required for AI-enabled R&D. The Siemens exit demonstrates strategic buyers may pay more for recurring, workflow-embedded R&D infrastructure than for binary-risk drug pipelines — a genuinely useful proof point for this sub-sector’s PE thesis.

EQT + Arsenal Capital Partners — Certara — EQT acquired Certara from Arsenal July 2017 at an $850M enterprise value; Arsenal retained an undisclosed minority stake. EQT supported next-generation technology, international expansion, complementary acquisitions; Certara IPO’d in 2020. Nov 2022: Arsenal invested $449M (~30M shares at $15/share) from EQT-controlled funds, taking expected diluted ownership to ~22%. Thesis: expand biosimulation/model-informed-drug-development adoption, deepen regulatory-science capabilities, consolidate complementary software/services — supported by recurring pharmaceutical workflows and diversified customers rather than single-asset dependence.

Blackstone Growth — DNAnexus — led $200M financing, March 2022, stake undisclosed. At the time: 12,000+ users across 48 countries, customers including 7 of the top 10 pharma companies, 65+ petabytes stored. Thesis: scale the governed data infrastructure pharma companies need before AI models can be trained/deployed reliably — an adjacent-enabler investment, not a direct bet on an AI-designed therapeutic.

PE-market conclusion: traditional control buyouts remain limited in pure-play AI drug discovery — many companies are loss-making, clinically immature, dependent on contingent milestones. PE activity is more established in R&D software, biosimulation, biomedical data infrastructure, and scientific workflow platforms, where recurring revenue, customer retention, and buy-and-build strategies are more compatible with PE underwriting.

Regulatory Context

United States

• Relevant bodies: FDA’s CDER, CBER, and (for combination products/software-based trial endpoints) CDRH.
• FDA draft guidance (Jan 2025) covers AI models generating information supporting regulatory decisions on drug/biologic safety, effectiveness, or quality — proposes a risk-based credibility assessment tied to specific context of use, informed partly by 500+ AI-component submissions (2016–2023) and 800+ external comments on a 2023 discussion paper.
• Critical investment distinction: discovery-only applications (initial target ID, molecular screening) not producing evidence supporting a regulatory decision are outside the draft guidance’s stated scope. Once AI output influences nonclinical evidence, patient selection, trial design, endpoint analysis, manufacturing controls, or safety evaluation, documentation of data provenance, model validation, context of use, and change control becomes important.
• As of July 2026, the Jan 2025 guidance remained in draft per contemporary regulatory reporting — creates near-term advantage for platforms already maintaining auditable data lineage, version control, reproducibility, and context-specific validation.
• FDA “clearance” counts for SaMD should not be used to assess this market — a generative chemistry model is not ordinarily cleared as a medical device merely for supporting discovery; the resulting drug proceeds through normal investigational/marketing-approval pathways.

India

• Principal regulator: CDSCO, led by the Drugs Controller General of India, under the Drugs and Cosmetics Act and applicable rules.
• AI-designed drugs are not subject to a separate CDSCO approval pathway solely for using AI in discovery — once a candidate enters human testing, it follows India’s standard clinical-trial and new-drug requirements (including SUGAM portal applications for manufacture/import/clinical-trial permissions).
• 2025–2026 developments: CDSCO invited comments on revised 2025 Guidelines on Similar Biologics (potentially relevant to AI-enabled biologics, not AI-specific); CDSCO initiated procurement for a Digital Drug Regulatory System in 2026 (confirms digitization intent, does not itself establish a new approval pathway for AI-discovered drugs); IndiaAI and ICMR reportedly entered a 2026 strategic partnership combining AI infrastructure with biomedical-research capabilities (source is a market-research publication, not an official agreement text — treat as not independently verified).
• Core India regulatory gap: not prohibition, but absence of a detailed AI-model credibility framework comparable to the FDA draft. Sponsors targeting international commercialization will likely need to design governance/validation to FDA, EMA, and ICH expectations even where Indian rules don’t expressly require the same AI-specific documentation.

Risk Factors

1. Clinical translation failure despite faster discovery — AI may improve target ranking or molecule generation without improving human efficacy; an investor can underwrite shorter preclinical timelines while still facing the same Phase II/III attrition as conventional biotech. Even the leading disclosed example, rentosertib, remained investigational when its Phase III trial began Sept 2026.
2. Milestone headline values may overstate economic scale — partnerships often advertise multibillion-dollar potential values, but only a small portion is paid upfront (Lilly-Insilico: $115M upfront against up to $2.63B contingent). If programs are discontinued, most headline value may never be realized.
3. Weak model moat and rapid commoditization — open-source models, cloud infrastructure, and foundation-model advances can erode software-only platform value. Sustainable differentiation requires proprietary experimental data, wet-lab feedback, workflow integration, validated therapeutic assets, or regulatory-grade data governance.
4. Data-rights and reproducibility failure — models trained on externally sourced chemical/genomic/patient/assay data may face unclear usage rights, inconsistent annotations, population bias, or irreproducible lab inputs. Inability to trace training data and model versions may lead pharma customers to exclude outputs from regulated development decisions under FDA’s context-of-use credibility approach.
5. Capital-intensity mismatch with PE return requirements — vertically integrated AI-biotech combines software development, wet-lab infrastructure, and therapeutic clinical trials; can remain cash-flow negative for extended periods requiring repeated equity financing. Schrödinger, despite $255.9M 2025 revenue, reported a $103.3M net loss — leverage-based underwriting is risky unless revenue is recurring and substantially independent of milestone receipts.

PE Attractiveness Verdict

Selective. AI in drug discovery has crossed from pilot-only experimentation to scaled pharmaceutical workflows, regulatory submissions, and genuine late-stage clinical programs, but has not yet demonstrated repeatable approval-level superiority over conventional discovery. The most attractive PE targets are not binary-risk “AI drug companies” — they are workflow-embedded R&D software, biosimulation, laboratory-data infrastructure, specialist AI-enabled CRO services, and platforms with recurring revenue, high switching costs, and auditable regulatory-grade data. The Dotmatics and Certara precedents show credible PE value creation through consolidation, cloud migration, international expansion, and strategic exit. Pure-play therapeutic platforms are better suited to venture, growth equity, or structured milestone financing unless they already possess validated clinical assets, diversified pharma partnerships, and sufficient capital to absorb trial failure.