Executive Summary

In March 2026, Eli Lilly & Company (“Lilly”) announced a landmark collaboration with Insilico Medicine—an artificial intelligence (AI)-driven biotech—to co-discover and develop new medicines. The deal commits $115 million upfront plus up to $2.63 billion in development, regulatory and commercial milestones, with tiered royalties on future sales (^[1]) (^[2]). Insilico will use its proprietary Pharma.AI platform (including deep-learning tools like PandaOmics and Chemistry42) to design multiple novel small-molecule drug candidates for indications designated by Lilly. In return, Lilly gains exclusive global rights to develop, manufacture and market the AI-discovered compounds (^[3]) (^[2]).

This collaboration is one of the largest AI-drug deals to date, underscoring how transformative AI has become in pharma R&D strategy. By leveraging generative AI and big data, companies like Lilly aim to dramatically accelerate discovery and reduce costs. Lilly’s investment deepens its multi-pronged AI strategy (e.g. partnerships with NVIDIA for a $1B “co-innovation lab” and internal supercomputing efforts (^[4]) (^[5])) and aligns with broader industry trends: the past year saw multiple biopharma giants sign nine-figure AI partnerships and even multi-billion-dollar pacts (for example, AbbVie’s $65M/$1.95B deal with Gilgamesh in 2024 (^[6])).

This report examines the historical context of AI in drug R&D, the specifics of the Lilly–Insilico agreement, and the state of AI-driven pharmaceutical innovation. We analyze how generative AI platforms work, profile Insilico’s progress and pipeline, and assess the implications for overall pharma R&D strategy. Drawing on data from industry reports, peer-reviewed articles, and expert commentary, we detail case studies and evidence-based insights into how AI commercialization is reshaping R&D portfolios, organizational priorities, and competitive positioning in the industry. We also discuss regulatory, ethical, and future considerations for AI-enabled drug development. Ultimately, the Lilly–Insilico collaboration exemplifies a strategic inflection point: pharmaceutical R&D is increasingly pivoting from traditional chemistry-based pipelines toward AI-assisted, data-driven discovery that promises faster timelines and novel therapies (^[7]) (^[8]).

Introduction and Background

The pharmaceutical sector traditionally has relied on long, labor-intensive R&D cycles, where developing a single new drug can take a decade or more and cost $1–3 billion. AI and machine learning (ML) methods have been explored for years (e.g. quantitative structure–activity relationships, computational docking), but recent advances in deep learning have dramatically expanded AI’s role. In particular, generative AI models – which can create novel molecular structures with desired properties – have emerged as a groundbreaking approach to drug discovery (^[9]) (^[10]). These models can ingest vast biological and chemical datasets, identify new drug targets, and propose candidate molecules in silico, exploring chemical spaces far beyond human capability. As Nai Huang (NVIDIA) and industry analysts note, AI promises to “revolutionize” drug research by enabling scientists to “explore vast biological and chemical spaces in silico before a single molecule is made” (^[11]) (^[8]).

This Lilly–Insilico partnership reflects a confluence of trends: pharmaceutical companies, striving to populate pipelines with more productive leads, are increasingly collaborating with AI firms. Recent years have seen a sharp rise in AI–pharma deals. Nature News (2025) documents that AI-enabled drug collaborations now regularly include nine-figure upfront payments and can reach multi-billion-dollar structures (^[12]) (^[6]). For example, in 2024 Novartis struck ~+$1.07B deals with Generate:Biomedicines, and Roche agreed $1.05B with Dyno Therapeutics (^[13]). Pharma-industry analytics report that AI/ML-related biopharma deals approached $10 billion in 2024, with Eli Lilly and Novartis each leading with multiple megadeals (e.g. a $1.75B pact between Lilly and Isomorphic Labs) (^[14]). In this context, the $2.75B Lilly–Insilico collaboration (announced March 2026) not only reflects Lilly’s commitment but stands among the largest AI-era deals.

Key players: Eli Lilly is a leading global pharmaceutical company known for breakthrough medicines (e.g. insulin analogs, cancer therapies, and notably the obesity/diabetes blockbuster Mounjaro/Zepbound). Lilly’s R&D spends on the order of $5–6 billion annually and the company has intensively pursued new modalities and technologies to boost its pipeline. Insilico Medicine (co-founded 2014 by Alex Zhavoronkov) is an AI-centric biotech headquartered in Hong Kong (with operations in Boston, Shanghai, etc.) that has developed its proprietary Pharma.AI suite of AI platforms for target discovery (PandaOmics), small-molecule design (Chemistry42), and drug development optimization (^[15]) (^[16]). Insilico claims a deep pipeline of AI-designed drug candidates; as of early 2026 it reported over 40 programs (24 publicly disclosed, 12 IND-approved) spanning fibrotic, oncologic, metabolic, and inflammatory diseases (^[17]). Notably, Insilico’s lead candidate, a novel TNIK inhibitor for idiopathic pulmonary fibrosis (IPF) called rentosertib (formerly INS018_055), was advanced into Phase II trials after its discovery through AI-driven workflows (^[9]) (^[18]).

For Lilly, the Insilico deal is a continuation of an AI strategy initiated in 2023–2025.Lilly had already licensed Insilico’s software in 2023, and in Nov. 2025 it announced a >$100M research pact with Insilico to jointly advance novel compounds (^[19]). Lilly is simultaneously building major compute infrastructure (a supercomputer “AI factory” with NVIDIA) and launching its own AI partnerships. The company’s CEO, David Ricks, publicly emphasizes that integrating Lilly’s clinical data and expertise with cutting-edge AI could “reinvent drug discovery” (^[8]). As McKinsey highlights, generative AI broadly offers the “once-in-a-century opportunity” to generate $60–110 billion annually for pharma via faster discovery, more efficient trials, and smarter commercialization (^[10]). Lilly appears to be acting on that insight: it joined NVIDIA in Jan 2026 in a $1 billion joint “co-innovation lab” to pair Lilly scientists with NVIDIA engineers on next-generation discovery platforms (^[4]) (^[20]).

This report provides a detailed analysis of what the Lilly–Insilico deal means for pharma R&D. We present: (1) the deal specifics and strategic context; (2) Insilico’s AI technology and accomplishments; (3) Lilly’s broader AI initiatives and pipeline needs; (4) industry-wide data and case studies on AI-driven drug development; (5) a discussion of implications for R&D strategy (including organizational, regulatory, and competitive factors); and (6) a thorough conclusion on future directions. Throughout, claims and data are backed by authoritative sources, including industry publications, academic reports, and company releases. We emphasize evidence-based analysis over hype, while exploring multiple perspectives (pharmaceutical executives, technology experts, regulatory scholars, and market analysts).

The Lilly–Insilico Collaboration

Deal Overview and Terms

On March 29, 2026, Lilly and Insilico announced a strategic discovery and development collaboration valued at up to $2.75 billion (^[21]) (^[2]). Under the agreement, Lilly will pay $115 million upfront to Insilico, with contingent milestone payments of up to approximately $2.63 billion tied to development, regulatory approvals, and sales thresholds of the resulting drug candidates (^[1]) (^[2]). In addition, Insilico will earn tiered royalties on any future drug sales from the funded programs. In exchange, Insilico grants Lilly an exclusive global license to develop, manufacture, and commercialize any “potentially best-in-class, novel oral therapeutics in preclinical development for certain indications” arising from the collaboration (^[3]). Lilly will lead the selection of research programs and direct the clinical development, leveraging its extensive disease expertise and regulatory capabilities, while Insilico provides its AI-driven discovery engine and laboratory automation to generate new drug candidates (^[1]) (^[22]).

Notably, Insilico’s announcement indicates the focus will be on high-unmet-need diseases, though the companies have not publicly detailed specific targets (^[23]). The collaboration “deepens Lilly’s connection” with Insilico, building on prior engagements: the two firms first signed an AI licensing agreement in 2023 (giving Lilly access to Insilico’s Pharma.AI software), and a $100M+ research pact in November 2025 (^[24]) (^[22]). According to Lilly, the deal allows Lilly to “identify promising candidates faster” by harnessing generative AI (^[25]). (Andrew Adams, Lilly’s head of molecule discovery, commented that using Insilico’s advanced AI platforms accelerates the search for novel compounds.)

The immediate market reaction highlighted the size of the deal: Bloomberg reports that Insilico’s Hong Kong–listed shares jumped up to 15% on the news (^[26]). This underscores investor recognition that the agreement—“worth up to $2.75 billion,” to use the exact phrasing of sources (^[26])—marks a major bet on AI. By comparison, a 2024 analysis by PharmInt noted that top AI-drug partnerships were approaching $1–2B each, making Lilly’s commitment especially large (^[13]). In sum, the deal terms align incentives and resources: Lilly essentially outsources early drug design to Insilico (paying mostly on success) while gaining exclusive rights to any breakthroughs, whereas Insilico gains a major client and funding to validate its AI platform on real-world programs.

Table 1 below summarizes the key financial terms and scope of the Lilly–Insilico deal, placed in context of other recent pharma–AI agreements:

^ [Note: Some figures above (denoted ^) are drawn from public reports and may include equity investments or aggregated milestone caps. See sources for specifics.]

Table 1 illustrates that Lilly’s Insilico deal (bottom row) stands out by its sheer scale ($2.75B+ in total potential value). While different deals target various modalities (e.g. biologics, gene therapies, oligonucleotides), the Lilly pact is notable for being focused on small-molecule drug discovery via generative AI – a testament to confidence that AI can produce whole new chemical entities. The structure (modest upfront, large contingent payments) is typical for early-stage, high-risk collaborations, aligning with Lilly’s focus on paying for successful outcomes (^[1]) (^[2]).

Project Focus and Integration

Under the agreement, Lilly selects which disease indications to pursue, leveraging its therapeutic area experts. Insilico then uses its platforms to generate and optimize compounds against targets relevant to those diseases. According to the announcement, only preclinical-phase novel molecules are in scope: Lilly obtains exclusive rights to those Insilico discovers and produces (^[3]) (^[2]). While specific targets or disease areas were not disclosed, typical Insilico focus areas (fibrosis, oncology, metabolism, inflammation) and Lilly priorities (cardiometabolic disease, neurodegeneration, etc.) suggest possible intersections. Lilly’s team will “lead selected research programs” and shepherd molecules through IND (Investigational New Drug) applications and clinical trials using Lilly’s global development capabilities (^[1]) (^[22]).

This essentially makes Insilico an outsourced discovery engine. Lilly solves one of its key R&D challenges—early target identification and lead generation—with advanced AI tools, while retaining all benefits if a drug succeeds. For Insilico, the payoff lies in milestone payments and royalties as Insilico’s molecules progress. Importantly, Insilico agreed to apply its Pharma.AI software suite (including PandaOmics for target discovery and Chemistry42 for molecule design) to Lilly’s projects (^[15]) (^[16]). According to Insilico, these tools have already been used to nominate dozens of candidates rapidly (see next section). Nothing in the announcement suggests a transfer of ownership of Insilico’s AI technology; rather, Lilly gains usage rights and a pipeline feed.

Financially, Insilico is incentivized by both milestones (tied to development/regulatory endpoints) and royalties. GlobalData (via Pharmaceutical Technology) notes that Lilly’s commitment extends to royalties on product sales (^[2]). Hence Insilico participates in the long-term success of any approved drugs, aligning interests. As Insilico stated, the collaboration “give [s] Lilly an ability to identify promising candidates more quickly” (^[25]), while positioning Insilico to benefit handsomely if a valuable therapy emerges.

Insilico Medicine: AI Drug Discovery Pioneer

Company and Platform Overview

Insilico Medicine is a biotech that rose to prominence by applying deep learning to pharmaceutical discovery. Founded in 2014, it developed the concept of an “end-to-end AI drug discovery platform”. In January 2024, Insilico went public on the Hong Kong Stock Exchange (stock code 3696), raising ~$291 million (^[30]). Lilly participated as a “cornerstone investor,” acquiring ~$5 million of Insilico shares in the IPO (^[31]). After going public, Insilico’s share price more than doubled, reflecting market excitement about AI drug pipelines (^[32]). However, Insilico’s 2025 results showed a large accounting loss (driven by the IPO’s fair-value adjustments) and a drop in revenue, signaling the financial strain in scaling a novel platform (^[33]).

Insilico’s technology stack, called Pharma.AI, includes several components developed in-house:

• Chemistry42: A generative chemistry engine for de novo small-molecule design. It integrates 40+ AI models (GANs, autoencoders, flow models, reinforcement learners, even transformer-based language models for molecules) into a multi-agent reinforcement-learning pipeline (^[16]). Chemistry42 rewards generated molecules that meet drug-like criteria (target activity, ADME properties, synthetic accessibility) and applies medicinal-chemistry filters (e.g. PAINS and toxicity alerts) to exclude undesirable structures (^[16]) (^[34]). Insilico published an application note (in J. Chem. Inf. Model., 2023) detailing Chemistry42’s architecture – noting that it launched in 2020 and has since been used in >20 pharma collaborations (^[35]) (^[16]). This platform is central to Insilico’s claim of generating “novel small molecules with optimized properties” (^[36]).

• PandaOmics: An AI-driven target discovery engine that analyzes multi-omics (genomics, transcriptomics, proteomics, clinical) datasets to rank promising biological targets and pathways. In practice, Insilico often uses PandaOmics first to find a disease mechanism or protein of interest, and then feeds that target to Chemistry42 to design compounds. For example, in their lead fibrosis program, PandaOmics parsed aging and tissue-damage data to propose a previously unstudied kinase (codenamed “Target X”) (^[37]), which was then pursued with Chemistry42.

• Automation and Robotics: Insilico has also invested in AI-controlled lab robotics to expedite synthesis and testing of AI-predicted compounds. Its “Humanoid Lab” concept (announced 2025) envisions robotic platforms that can perform experiments continuously, learning chemical procedures from human scientists (^[38]). This is intended to shorten the design–test loop. While specific progress on the robotics is not fully disclosed, the company’s automation labs (e.g. LifeStar2 facility) are highlighted in company imagery.

• inClinico (AI for Development): Insilico mentions an inClinico module to predict clinical trial success probabilities, potentially recommending trial design optimizations. Details are limited, but the tool aims to link preclinical predictions with likely human outcomes.

The core idea is an “end-to-end AI workflow”: from target to candidate to preclinical development. Insilico’s leadership asserts that AI adds the most value in the earliest R&D stages (“zero to preclinical candidate”) (^[39]). Indeed, company data indicate dramatically shortened timelines: since 2021, Insilico has nominated 28 preclinical candidates using its AI methods (^[7]). They claim an average of 12–18 months per program (from project start to candidate nomination) (compared to a typical 3–6 years by traditional methods (^[7])) and required synthesizing only dozens to a couple hundred molecules per program (versus screening millions in older paradigms (^[7])). Whether these accelerated metrics fully hold in practice will be seen over time, but they illustrate the company’s value proposition: massive reduction in discovery timelines and resources. A 2023 Insilico blog announced that INS018_055 (rentosertib), their first generative AI-derived molecule, had entered Phase II, noting it had gone from in silico concept to clinical trials in roughly 30 months (^[18]) (about half the usual time). While that is a company press release and should be read with caution, it demonstrates a concrete milestone in AI-driven discovery (^[9]) (^[18]).

Insilico’s Pipeline and Progress

By early 2026, Insilico reports over 40 distinct drug programs in its pipeline. Of these, 24 are publicly disclosed, and 12 programs have already received IND clearance (allowing clinical trials) (^[17]). The division of indications is broad: the portfolio includes fibrosis, oncology, metabolic disorders, inflammation, neurology, and more. The company’s farthest-advanced asset is rentosertib (renamed ISM001-055 or REN) for idiopathic pulmonary fibrosis (IPF) (^[40]). This small-molecule inhibitor targets TNIK kinase; it demonstrated safety in Phase I trials and has completed a Phase IIa study in China (published in Nature Medicine, 2025) (^[40]). Additional trials for IPF (including inhaled formulations) are underway or planned.

Insilico has spun out several programs in partnership or licensing deals (below). Notably, it also engages with larger pharma in technology partnerships (see next section for details on deals). The total pipeline beyond disclosed programs suggests Insilico has dozens of AI-nominated projects in discovery and preclinical stages. In-house, Insilico highlights an accelerated “AI-driven bottom-up philosophy” where target discovery and lead design happen in parallel, contrasting with sequential conventional R&D (^[39]). The eventual clinical viability of these candidates is still unfolding, but Insilico’s early data (e.g. multiplex potency across fibrosis-related targets (^[41])) and partnerships (below) indicate industrial interest.

Insilico Collaborations and Case Deals

Table 2 summarizes selected notable collaborations and internal programs of Insilico, showcasing how it has already engaged with pharma partners (many outside Lilly).

^ [†] The table includes representative deals. Some collaborations (like Fosun or Lilly deals) specify milestones but often keep total Milestones confidential beyond maximum figures. ††Insilico as of early 2026 states it has nominated 28 preclinical candidates since 2021 (^[7]); about half are now in or entering clinical trials. Many internally discovered candidates target fibrosis, cancer, metabolic, and inflammatory pathways.

These examples illustrate Insilico’s role as an AI-platform provider as well as originator of drug candidates. Its high-value partnerships (Merck, Servier, Exelixis, etc.) suggest confidence in its approach. For instance, the Exelixis deal (2022) gave Exelixis exclusive rights to an Insilico-designed USP1 inhibitor (XL309) for certain cancers (^[42]). Exelixis has since paid milestone payments (a $10M clinical trigger in 2024) and is advancing the program in Phase I (^[42]). Similarly, in early 2025 Insilico and Menarini (via Stemline) signed a new oncology pact with up to $550M total – an unusually large sum for preclinical assets (^[45]). These deals are anchored around Insilico’s generative chemistry tech; Menarini’s collaboration explicitly leverages Insilico’s AI for in silico molecular design.

Collectively, Insilico’s track record shows repeated validation of AI-derived molecules: at least two candidate programs (REN for IPF, KIF18A inhibitor for cancer) have moved into clinical testing based on Insilico’s AI platform. The Lilly collab is the latest and largest, designed to blend Lilly’s development muscle with Insilico’s AI pipeline generation.

Eli Lilly’s AI-Driven R&D Strategy

A Rapid Evolution Toward AI

Eli Lilly has long invested in cutting-edge R&D, and in recent years its focus has turned sharply toward AI integration. Historically, Lilly’s R&D portfolio centers on diabetes, oncology, immunology and neuroscience. Its success with peptide drugs (insulin analogs, GLP-1 agonists) and novel classes (e.g. PCSK9 inhibitors) positioned it for strong sales, resulting in over $30 billion in revenue by 2025. With that success, Lilly is well-capitalized to experiment with new technologies.

Beginning around 2020–2021, Lilly quietly built in-house AI capabilities. For example, Lilly’s Indiana-based Institute for Genetic Medicine (headed by Andrew Adams) began applying AI tools to target discovery. In early 2023, Lilly announced that it had teamed with leading ML developers and data providers, reflecting growing interest in AI. The pace accelerated notably in late 2024 and 2025.

Prominent moves include:

• Insilico Partnership (2023–2026): Lilly first licensed Insilico’s Pharma.AI in late 2023 (software-only deal, details undisclosed) and renewed with the $100M research pact in Nov. 2025 (^[22]) (estimated over $100M in total payments). These prior agreements gave Lilly evaluation access and small-scale collaboration experience. The Mar. 2026 deal greatly expands that scope and financial commitment.

• NVIDIA and the AI Co-Innovation Lab (2026): In Jan. 2026, Lilly and NVIDIA announced a joint $1 billion over 5 years initiative to create an AI co-innovation lab in the San Francisco Bay Area (^[4]) (^[20]). This lab will co-locate Lilly biologists and chemists with NVIDIA’s AI engineers, building “foundation and frontier models for biology and chemistry” using NVIDIA’s BioNeMo platform (^[4]) (^[49]). Lilly CEO Ricks said this collaboration could “reinvent drug discovery as we know it”, leveraging “world-class data” and compute (^[8]). The co-lab is intended to integrate Lilly’s “wet labs” directly with AI-driven “dry labs” for continuous learning (24/7 experimentation with AI feedback) (^[49]). This builds on Lilly’s announcement in late 2025 of an in-house AI supercomputer (“AI Factory”) – claimed to be the industry’s largest PB-level system for drug R&D (^[50]). At JPMorgan 2025, Lilly had already disclosed a roadmap to use this compute for training models that predict molecules and prioritize candidates (^[51]).

• Lilly TuneLab Platform: Lilly has also launched an internal ML platform, TuneLab, to share Lilly’s models with external biotech partners (^[52]) (^[53]). For example, Lilly invited startups like Circle Pharma (macrocycle drugs) and Insitro (AI-centric biotech) to use Lilly’s molecular-prediction models via TuneLab (^[52]). This openness serves to crowdsource model improvements and acquire more data (partners provide training data). TuneLab’s use of open-models like NVIDIA’s Clara (for life sciences) demonstrates Lilly’s interest in being a data integrator across the field (^[53]).

• Multiple AI Collaborations: Beyond Insilico and NVIDIA, Lilly recently inked several AI collaborations in late 2025/early 2026, each focusing on different modalities. For instance, Lilly signed a >$1B biotech deal with Creyon Bio (AI-designed RNA therapeutics) (^[28]), and joined Manulife’s Project B – involving GSK and Pfizer – to create AI tools for drug discovery. Other deals include pacts with Benchling, Chai Discovery, Revvity, Schrödinger, and others (^[51]). The pattern is clear: Lilly is pursuing an “open innovation” strategy – partnering with specialized tech firms to accelerate discovery rather than relying solely on internal labs.

All these moves are anchored by Lilly’s commercial success. Notably, strong sales of Mounjaro (tirzepatide) have fueled Lilly’s $1 trillion market capitalization (as of early 2026) (^[54]). Financial strength enables Lilly to allocate unprecedented sums toward AI initiatives. For context, a recent industry summary [South China Morning Post] notes Lilly had signed 16 AI-focused deals in 2025 alone (including 7 licensing agreements worth >$2B total), reflecting an aggressive pivot toward tech-driven R&D (^[28]). Lilly is even applying AI beyond discovery – exploring “agentic AI” for experiment planning and digital-twin simulations in manufacturing (^[55]) (^[56]).

Rationale and Implications

Lilly’s leadership sees AI as the next frontier. Jensen Huang (NVIDIA CEO) explicitly cited Lilly at the Davos 2026 World Economic Forum as a prime example of “pharma giants making the leap” to AI platforms (^[11]). Internally, Lilly executives note that traditional R&D hit productivity limits; everyone screens millions of compounds and most fail. AI can “find the needle in the haystack faster,” as one biotech investor puts it (^[57]). Lilly aims to use AI to shorten timelines (e.g. from target identification to lead by years), reduce attrition (filter out dead ends early), and expand chemical diversity (design molecules humans have not considered).

The Insilico deal advances these strategic goals by delivering a turnkey AI engine. Instead of Lilly building new algorithms from scratch, it buys into Insilico’s mature platform and talent. For Lilly’s R&D planning, this means shifting resources: rather than allocating hundreds of chemists and biologists to brute-force screening or manual synthesis, Lilly can redeploy those to later stages and to validating AI-generated hits. In effect, Lilly treats Insilico’s AI as an “advanced vendor” that generates candidate series that enter Lilly’s drug pipeline. This could make Lilly’s early R&D portfolio sparser (in terms of programs Lilly scouts itself) but potentially higher-quality, with each program accelerated by AI insight.

However, it also means greater reliance on external technology. Lilly must now manage technical integration and IP share. The structure (milestones & royalties) distributes risk (Lilly pays only if milestones met). Strategically, Lilly positions itself not just as a traditional biotech, but as a technology-driven R&D orchestrator. Its value proposition to shareholders increasingly depends on translating AI investments into marketable drugs.

We will discuss in later sections how this strategy compares with other companies’ approaches and what it means for pharma R&D more broadly. But already, as of 2026, Lilly’s aggressive building of AI infrastructure (hardware, software collaborations, talent acquisition) signals it views AI not as a transient experiment but as a core R&D competency. The Insilico agreement, with its massive financial commitment, is arguably the flagship of this strategy.

AI in Pharma: Industry Context and Data

To understand the Lilly–Insilico deal’s significance, we review the broader landscape of AI-enabled drug discovery. AI partnerships span from startups to big tech, and cover small molecules, biologics, genomics, and trials. Multiple sources document rapidly accelerating investment:

• Deal Volume and Value: A recent Nature news feature reports AI drug deals commanding nine-figure upfronts and multi-billion milestones, across modalities (small molecules, biologics, ADCs, oligonucleotides) (^[12]) (^[27]). Aris Persidis (executive in biotech analytics) notes that “machine learning remains in vogue”, with companies sparing no expense for integrated AI platforms. An analysis by Pharmint shows global AI/ML pharma deals reached nearly $10 billion in 2024, with Eli Lilly and Novartis leading the count of major deals (^[14]). Ondeal lists confirm multiple megadeals: for example, Merck KGaA partnered with Biolojic Design for ~€346M in milestones (antibody engineering) (^[58]), and GSK has deals with Vertex (Exscientia) and Recursion for hundreds of millions in AI-driven pipelines.

• Payment Structures: These partnerships typically feature heavy back-loaded payments. The upfront cash often covers initial research, while milestones (linked to clinical or sales targets) account for the bulk of value. For instance, the Gilgamesh–AbbVie pact had just $65M upfront and ~$1.95B in milestones (^[6]). Similarly, Insilico’s collaboration offers only $115M initially, with $2.63B contingent (^[1]) (^[2]), reflecting Lilly’s strategy to pay for success. Equity components sometimes appear (Lilly took a small equity stake in Insilico’s IPO (^[32])), but cash remains the core.

• Pharma Players: Nearly all top pharmas are now active in AI. Roche has partnered with Dyno and others, Novartis with Generate:Biomedicines, Pfizer with Recursion, and J&J with BenevolentAI. AstraZeneca acquired an AI startup (Moderna’s ex-team, RaNA Therapeutics) for RNA design (2022), and, as of Jan 2026, AZ agreed to buy Modella AI for oncology pipeline design (^[59]). GSK has taken stakes in firms like Exscientia and supported AI platforms. Even beyond R&D, AI is used for trial recruitment and manufacturing optimization. The cumulative effect is an industry embracing AI across the board. A recent McKinsey report estimates AI (particularly generative AI) could contribute $60–110 billion per year in pharma value, chiefly by speeding discovery and development (^[10]).

• Clinical Outcomes: As of early 2026, no AI-originated drug has yet been approved in humans. The field is still proving itself: most AI-derived candidates are in early trials. Insilico’s INS018_055 is often cited as the first fully AI-designed molecule reaching human trials (Phase I in 2021, Phase II in 2023 (^[9])). In parallel, Exscientia’s AI-designed epilepsy drug (FINLAY-FR195) advanced to human testing (San Carlos, UK). However, major regulatory bodies (FDA, EMA) have not yet confronted AI-specific challenges beyond standard drug evaluation. Skeptics caution that AI must still prove compound quality and safety, and that “AI hype” may overpromise.

• Regulatory Considerations: Some analysts (e.g. Ast Reg reports) warn of a “regulatory tightrope” upcoming (^[60]) (^[61]). AI models often behave as “black boxes,” making it hard to rationalize a molecule’s origin for regulators. A recent International Bar Association review notes that the opacity of deep learning poses challenges for FDA/EMA expectations of explainability (^[61]) (^[62]). Additionally, biases in training data or dynamic model updates require new oversight frameworks (^[63]) (^[64])—not mission-critical for Lilly–Insilico itself but part of the future ecosystem.

In sum, Lilly’s deal arrives at a moment when AI drug development has proven enough conceptually to command major deals, but is still at an early inflection point in terms of clinical validation. The scale of investment ($2.75B) signals that Lilly expects AI-derived programs will soon reach clinics and ultimately patients – fundamentally altering the R&D pipeline.

Case Studies and Examples

To illustrate the real-world dynamics of AI in pharma R&D, we consider several case studies beyond Lilly–Insilico.

1. Insilico–Lilly (March 2026)

As detailed above, this collaboration is among the first AI-centric discovery deals of its size. It exemplifies co-development of AI-generated small molecules, combining Insilico’s generative chemistry with Lilly’s clinical expertise. Key learning points:

• Outputs to Date: Nothing concrete has yet been announced on specific compounds or targets. The agreement covers “selected research programs” – presumably still in early stages. Insilico’s pre-existing candidates (e.g. fibrosis and oncology drugs) are unlikely included, as the language suggests new programs initiated by Lilly’s choice. Thus, commercialization is still years away; success depends on whether any of Insilico’s AI candidates can demonstrate efficacy in IND-enabling studies.

• Strategic Significance: For Lilly, the deal is both a research partnership and a message to the market that it is leaning heavily into AI. It also raises questions: Will Lilly eventually acquire Insilico outright (as speculated by some analysts) if collaboration succeeds? Or will Insilico remain independent? (Alex Zhavoronkov did not commit in earnings calls (^[65]).) Another implication is pipeline focus: Lilly may now have an influx of AI candidates in areas outside its traditional disease fortes (if Insilico’s systems highlight novel chemistry against niche targets). This could diversify Lilly’s pipeline risk profile, but also stretch its resources into new therapeutic domains.

• Comparison to Other Lilly Programs: Lilly’s internal AI efforts (NVIDIA lab, AI supercomputer) suggest it sees Insilico not as replacing its computational teams but as augmenting them. While Lilly’s own models might eventually design molecules, partnering with Insilico saves development time and brings external expertise.

2. AbbVie–Gilgamesh (May 2024)

In mid-2024, AbbVie partnered with Gilgamesh Pharmaceuticals in a novel psychedelic-inspired AI deal (^[27]). Gilgamesh’s Neuroplastogen platform uses AI/machine learning to design molecules related to ibogaine, a known neuroplasticity agent. AbbVie paid $65M upfront and could pay up to $1.95B (milestones) (^[27]). This deal targeted CNS diseases (depression, PTSD, anxiety). It illustrates two industry trends:

• AI for New Modalities: Most big deals involve small molecules; this one involves psychedelic-modified compounds. It shows AI can unlock non-traditional modalities.

• De-Risking Innovation: AbbVie saw value in Gilgamesh’s AI to “de-risk” entering the psychedelic space. By paying a capped upfront and large milestones, AbbVie hedged on Gilgamesh delivering viable candidates. If AI design fails, AbbVie’s total cost is limited.

This deal’s structure is similar to Lilly’s: upfront plus milestone-laden. It also highlights that large pharmas are willing to bet enormous sums on AI-found leads. If an AI-driven candidate were to become an approved CNS drug, it would likely yield tens of billions in revenue (justifying the high milestones).

3. Novartis–Generate:Biomedicines (June 2024)

Generate is a biotech using AI to engineer biologics. In mid-2024, Novartis agreed to pay ~$1.07B in milestones (plus an upfront) to license AI-designed multispecific antibodies from Generate (^[66]). Unlike Lilly–Insilico, this involve protein therapeutics (antibodies, ADCs) rather than small molecules. The lesson here is that:

• Firms outside small-molecule space also leverage AI.
• Upfronts may be smaller (Novartis’s on upfront not publicly stated, likely tens of millions), with most payment in milestones.

While Novartis’s deal details are not fully public, it reflects the same theme: integrate AI to tackle complex targets (in this case, “challenging oncology targets” with AI-designed ADCs). It broadens our understanding of pharma R&D strategy: AI is not limited to chemists’ work but applies to antibody engineering, hitching onto digital biology.

4. Roche–Dyno Therapeutics (March 2024)

Dyno applies machine learning to engineer gene therapy payloads (AAV vectors) and synthetic biology tools. Roche’s 2024 deal with Dyno was ~$1.05B (incl. ~€50M upfront) for rights to AI-developed AAV capsids targeting the central nervous system (^[66]). This deal, like Novartis’s, is outside the small-molecule realm. It shows:

• Diverse AI Uses: Even in gene therapy, AI can optimize the biological components (vectors) for better tissue targeting and manufacturability.
• Large Pharma Is Interested: Roche paid Novartis-like money for Dyno’s enzyme/protein tech. If successful, it could yield safer, more efficient gene therapies for neurodegenerative diseases.

5. AstraZeneca–Moderna (2021–25)

Though outside drug discovery per se, a relevant example is AZ’s 2021 collaboration with Moderna to explore mRNA therapies using AI and high-performance computing. AZ invested in AI modeling for mRNA design. More recently (Jan 2026), AstraZeneca agreed to acquire MediMab (ex- Moderna), an AI-driven startup in oncology (^[59]). These moves show the line between pharma and tech is blurring. AZ’s acquisitions and partnerships signal that big drugmakers are integrating AI companies either by deals or M&A to accelerate their pipelines.

6. BenevolentAI–J&J (2019–Present)

Johnson & Johnson’s Janssen division partnered with BenevolentAI (UK) around 2019 for AI-driven drug discovery. They paid $65M upfront and $1.9B in milestones for two neuroscience projects. In 2022, J&J announced termination of one partnership (less successful) but continued the other. This saga is a cautionary tale in AI R&D: not every project yields results, and exiting can happen. It highlighted early challenges of validating AI candidates. Nevertheless, J&J credited AI insights even where compounds fell short. For J&J’s R&D strategy, it underscored that AI projects carry risk, but those companies persisted in AI investment. (See, e.g., media reporting on BenevolentAI deals in 2019–2022.)

Data Analysis: AI Impact Metrics

Beyond deals, what data quantify AI’s influence on R&D? Few longitudinal metrics exist yet, but we note:

• Cost and Time Reductions: Studies suggest AI can significantly cut early discovery costs. McKinsey’s analysis estimates generative AI can speed up target-to-hit processes, potentially reducing discovery costs by ~20–40% (^[10]). Insilico’s claims (12–18 months per candidate vs ~60 months tradition (^[7])) align with a multi-fold speedup. A broader report estimates say AI could shorten some lead optimization with 40-70% fewer compounds tested (since AI screens virtual libraries) – but actual drug cost savings per successful asset will only be apparent once AI-derived drugs reach approval.

• Portfolio and Pipeline Metrics: Some consultancies track how many drug candidates list “AI” involvement. GlobalData’s Pharma Intelligence Center reports that, by late 2025, Lilly added 7 AI-licensed projects (total >$2B) to its pipeline (^[28]). Meanwhile, other firms: GSK has co-pipelines in AI-designed small molecules and antibodies; Pfizer has deals with Recursion for in vivo phenotyping; and Sanofi is using AI for selective kinase inhibitors. These pipelines remain “incubating” – only time (and trial results) will prove their quality.

• R&D Productivity: Industry R&D productivity (new NME approvals per R&D dollar) had plateaued for decades. Early indications from analysts say AI could “bend the cost curve”, but quantitative evidence of improved approval rates or higher candidate quality is not yet public. Anecdotally, leaders like Andrew Adams argue that AI has become part of a “lab-to-API” pipeline, but emphasize that clinical validation is crucial.

• Investment Trends: Venture funding in AI biotech boomed from a few hundred million per year pre-2020 to several billion in 2021–2023. Startups like Insilico, Exscientia, Recursion became unicorns. In 2025, even private companies like Isomorphic Labs (Alphabet’s spinout) attracted multiple billion-dollar deals (^[67]). According to industry trackers, AI drug startup valuations reached record highs by 2024, though some cooling occurred as investors awaited proof-of-concept.

We summarize key data points:

• Deals: ~US$10B in AI/ML pharma deals in 2024 (^[14]). Lilly’s own 2025 AI deals totaled well over $2B (Creyon ~$1B, two Insilico deals $0.2B+ (^[2]) (^[28])).
• Licenses by year: Lilly – 16 AI deals in 2025 (^[28]); GSK – 10+ new AI alliances in same period (source: FiercePharma interviews).
• Upfront vs milestones: Typical ratio is small upfront (~5–15%) vs total deal (~85–95%) (^[12]) (^[6]). Lilly–Insilico’s $115M vs $2.63B fits this pattern (~4.3% upfront).

Implications for Pharma R&D Strategy

The Lilly–Insilico deal and its peers carry significant implications for how pharma companies organize and execute R&D. Below we discuss several strategic considerations:

1. Portfolio Management and Pipeline Allocation

Historically, big pharma diversified across many programs, often de-risking by maintaining large portfolios (the “law of large numbers”). Integrating AI shifts this calculus in two ways:

• Concentration on High-Value AI Programs: Lilly’s willingness to pay up to $2.75B implies confidence in fewer, high-potential projects rather than many low-probability ones. If AI shortens early-phase time, Lilly can afford to cycle through more programs in a given timeframe – focusing R&D headcount on advancing AI-screened leads. In practice, this may lead to leaner in-house discovery teams focusing on validation and optimization, while partnering covers ideation.

• Changing Risk Profiles: In traditional R&D, early failures (target invalidation or lead failure) eat most costs. With AI, the risk shifts earlier but in a more controlled manner: a drug candidate failing IND due to poor ADME/etc. would have cost Lilly less (since Insilico bore early discovery costs). Conversely, if AI-derived leads consistently fail in early clinical phases (for lack of efficacy), Lilly would have fronted less cost but the pipeline value would erode. In either case, R&D budgeting must adapt to different risk timing and cashflow.

• Open Innovation and Externalization: Partnerships like this effectively outsourced discovery to Insilico. Lilly’s R&D strategy thus becomes more oriented toward managing an ecosystem: licensing in tech from AI firms, incubating startups via TuneLab, and potentially acquiring AI companies in the future. This is a departure from older models where discovery was largely internal. Other pharma companies face similar strategic decisions: devote internal resources to AI (hire ML scientists) vs. collaborate or acquire external AI startups.

2. Talent and Organizational Structure

Implementing AI drug discovery requires new skills. Lilly (and peers) will need to hire data scientists, bioinformaticians, software engineers, and AI researchers – roles not traditionally found in pharma R&D. Organizing these new teams poses challenges: how to integrate AI teams with chemists and biologists? The Insitro CEO notes a “culture clash” between engineers and life scientists (^[68]), requiring companies to create interdisciplinary teams. Lilly’s formation of the co-innovation lab is one approach: co-locating domain experts and AI experts in a startup-like setting (^[4]).

For Lilly specifically, existing groups like its computational biology and genetic medicine institutes will likely grow. Meanwhile, Lilly’s leaders must manage career incentives: Will chemists and biologists be enthusiastic to see AI handling molecules? In some cases, jobs may shift from synthesis to assay evaluation or AI supervision. Internally, monitoring and interpreting AI outputs (balancing trust vs scrutiny) will be a new duty. Lilly has also created Lilly Labs as a semi-autonomous unit focused on AI and digital innovation, potentially to provide flexibility.

3. Data Strategy and Collaborations

AI thrives on data. Lilly must leverage its vast internal knowledge base (billions of historical data points from past R&D) by training algorithms on its own data to complement Insilico’s models. There are trade-offs: proprietary data can yield better models, but using them within Insilico’s platform requires careful IP/data arrangements. The co-lab announcement suggests Lilly will generate “large-scale data” to train new models (^[55]).

Moreover, Lilly participates in precompetitive data-sharing initiatives (e.g. DREAM Challenges) to augment data diversity. It may also purchase or license datasets (e.g. genomic databases) for AI training. The Insilico deal itself likely included confidentiality rules on data sharing. A strong data pipeline strategy will be critical: without high-quality training data, generative AI cannot outperform known chemistry.

4. IP and Commercialization

AI brings nuanced IP issues. Can a company patent a molecule first created by an algorithm? Generally, yes, as long as there is human inventorship in the claim. Lilly will patent any promising Insilico-designed leads, but its intellectual capture depends on how agreements are structured. In this case, Lilly gets exclusive commercialization rights, suggesting it will own or control resulting IP. For Insilico, its reward is in milestones and royalties, not ownership.

Commercially, if a successful AI-derived drug emerges, Lilly will reap revenue but also must pay Insilico a share. This may require creative financial forecasting and valuation methods: how to set the fair royalties for something originally “invented” by AI. Public markets are paying attention: valuation agencies (like Simply Wall St) are already analyzing how this pact affects Lilly’s stock and pipeline value (^[69]).

5. Regulatory and Ethical Oversight

One cannot ignore how regulators will view AI-derived drugs. Lilly will need to ensure that Insilico’s discovery process is well-documented and scientifically sound. The IBA report warns that regulators may soon demand transparency on AI mechanisms (^[61]). In practice, Lilly’s regulatory submissions for an Insilico-originating drug will likely focus on rigorous preclinical and clinical evidence; the “AI” origin may get only a sidebar mention unless regulators explicitly ask about computational methods. Inside R&D strategy, Lilly must maintain compliance by perhaps overlapping traditional target validation steps with AI prediction to show mechanistic rationale.

Ethically, agents like ChatGPT have shown people are uneasy with “black box” decisions (^[70]). Lilly’s marketing and communications departments will need to consider how to present an AI-derived drug. Will they say “AI-designed molecule” in public materials? Possibly not in patient-facing info, but among the scientific community, novel AI origin stories may become selling points (illustrating innovation).

6. Future R&D Directions

Looking forward, the Lilly–Insilico deal foreshadows broader industry shifts:

• Platformization of Drug Discovery: If successful, similar deals may become commonplace. Pharma R&D groups might be organized around platform assets (AI models), just as tech companies organize around software. Lilly might expand this deal to biologics (if Insilico or another partner adapts AI for proteins) or genomics (e.g. AI-driven gene editors).

• Competitive Pressure: If Lilly’s AI pipeline yields breakthrough drugs faster, competitors will be compelled to match. This could spark a new “arms race” in R&D investment and partnerships. Companies without such deals may struggle to keep pace, potentially leading to M&A or greater focus on AI capability (e.g. smaller biotechs offering AI-as-a-service to big pharma).

• Drug Repurposing and Precision Medicine: Although the Lilly–Insilico deal focuses on novel molecules, once AI is in use, companies are likely to extend it to repurposing existing drugs or finding biomarkers. R&D strategy may increasingly tie discovery to clinical data, enabling more personalized drug development. Lilly could, for example, use AI to match its pipeline candidates to genomic or phenotypic patient subsets, improving trial success rates.

• Collaborative R&D Ecosystems: Lilly’s TuneLab initiative indicates a move toward open innovation. In the future, large pharmas might form consortia sharing AI tools for pre-competitive targets (e.g. Alzheimer’s disease through shared models), while competing in late-stage development. This could reshape how preclinical research is funded and conducted across the industry.

Overall, the implications of the Lilly–Insilico $2.75B pact are profound. It represents a new model of R&D spend: a large portion of discovery is now encapsulated in partnerships with tech-driven companies. If AI delivers even a fraction of its promised speedup, pharmaceutical R&D might shift from cost-of-failure toward fee-for-service research, with traditional discovery benches giving way to hybrid labs run by AI and robotics. This may both increase efficiency and require pharma companies to become comfortable managing AI pipelines as core assets.

Future Outlook and Conclusions

The Lilly–Insilico collaboration marks a milestone in the industry’s march toward AI-driven drug discovery. This report has examined its context and consequences: Lilly is effectively doubling down on AI by outsourcing early-stage chemistry to an AI specialist, funded through one of the largest-ever biotech partnerships. We have detailed how Insilico’s generative AI platform can propose dozens of candidates rapidly; how Lilly’s broad AI strategy (supercomputers, partnerships, platform sharing) supports such deals; and how this pattern of megadeals is representative of a wider sector trend (^[28]) (^[10]).

While the ultimate success of Insilico’s AI-derived drugs remains to be proven in clinical trials, the evidence suggests pharma R&D strategy is already changing. Key takeaways include:

• Acceleration: AI shows promise to greatly accelerate early R&D. Insilico claims to nominate candidates in 1–1.5 years vs. 5+ years traditionally (^[7]). Should this hold true, drug pipelines could be replenished much faster. Lilly is betting that candidates generated by AI will fill meaningful portions of its future pipeline.

• Cost Efficiency (Potential): By testing far fewer virtual compounds, companies can save millions in synthesis and screening. For Lilly, paying Insilico primarily on milestones means shifting R&D expenditure risk from fixed salary budgets to conditional payments. If Insilico delivers, Lilly gets a free trial; if not, Lilly’s loss is limited to the upfront fee.

• Risk Diversification: Pharma can diversify risk by investing in multiple AI platforms. Lilly, for example, works with Insilico, NVIDIA (hardware/software), and offers TuneLab to other AI firms. If one platform underperforms, others may succeed.

• Competitive Landscape: Early adopters (Lilly, Novartis, Roche) may gain a competitive edge. Weaker R&D returns are already driving big players to seek innovation externally. Smaller companies, especially biotech startups, will increasingly focus on AI as their core value proposition, hoping to partner with (or be acquired by) big pharma.

• Regulatory Evolution: Regulators are aware of AI’s growing role. We expect new guidelines on AI in drug development, similar to how agencies have worked on cell/gene therapy frameworks. Companies will need to advocate for clear pathways to approval of AI-derived molecules, potentially engaging in dialogues with FDA/EMA about best practices for algorithmic drug design documentation.

• Clinical Success is Key: Ultimately, the market and healthcare outcomes will judge these strategies. Not all AI-engendered candidates will succeed, as was the case with early biotech-era drug engines. But each failure will feed more data back into the AI, potentially improving future predictions. Lilly’s financial structure (large milestone pools) implies it sees a viable path: if even one or two AI-derived drugs are approved, the revenues could easily justify the upfront portfolio.

In conclusion, the $2.75B Lilly–Insilico deal is a landmark indicator that AI-based drug discovery is entering a commercialization phase. It symbolizes an industry transitioning from pilot projects to committed integration of AI into core R&D. We anticipate that within 5–10 years, multiple FDA analytics reviews will cite machine learning both in the methods of discovery and in post-hoc analyses of pharmacology. Companies that embed AI tools effectively into their pipelines may dramatically boost productivity—gaining first-mover advantage in rapidly meeting patient needs. For R&D strategy, this means allocating significant resources to AI, forming specialized R&D arms or labs, and rethinking drug development as a hybrid human–machine workflow.

Tables and figures included (above) highlight deal comparisons and Insilico’s deal portfolio; these quantitatively frame how this collaboration fits into a sprawling new ecosystem of AI-driven innovation (^[1]) (^[6]). As an academic perspective, it’s clear that Lilly’s approach is scrutinized by investors, scientists, and regulators alike. Early technical and clinical results will shape the narrative: success could validate massive AI investments, while setbacks will necessitate strategic recalibration.

Future Directions: Looking ahead, Lilly and Insilico will likely fine-tune their collaboration. Possible future steps include expanding into additional AI modalities (e.g. biologicals, if Insilico or partners develop those capabilities), adjusting targets based on initial progress, or even deeper integration (perhaps joint labs or equity stakes). For the broader field, we expect regulatory bodies to possibly pilot “living” AI model approvals (^[64]), and for industry consortia (e.g. NIH initiatives) to set shared data standards for AI drug discovery.

In sum, the Lilly–Insilico alliance is more than a single deal: it is a case study in the evolving nexus of pharma and AI, shedding light on how the next generation of medicines might be created. Its full impact will only emerge over time, but already it signals a paradigm shift in how pharmaceutical R&D is conceptualized and executed in the AI era (^[10]) (^[8]).

References

• Lilly & Insilico 2026 Collaboration Announcement (^[21]) (^[2])
• Lilly NVIDIA AI Lab Press Release (Jan 12, 2026) (^[4]) (^[8])
• Insilico Medicine press materials and pipeline data (^[9]) (^[47])
• Alex Philippidis and Fay Lin, GEN Eng. News, “Lilly Grows AI Footprint with Up-to-$2.75B Insilico Collaboration” (Mar. 2026) (^[21]) (^[47])
• Gabrielle Masson, Fierce Biotech, “Lilly continues AI push, inking $100M-plus pact with Insilico” (Nov. 10, 2025) (^[19]) (^[71])
• Annabel Kartal Allen, PharmaTech (GlobalData), “Eli Lilly and Insilico Medicine forge $2.75bn AI-driven R&D pact” (Mar. 30, 2026) (^[2]) (^[28])
• Axios news (Davos 2026): Jensen Huang on AI in Pharma (^[11])
• Nature News Feature “AI deals show no sign of slowing” (Marshall, 2 June 2025) (^[12]) (^[6])
• McKinsey & Co., “Generative AI in the pharmaceutical industry: Moving from hype to reality” (Jan. 9, 2024) (^[10])
• International Bar Association, “AI in drug discovery: a regulatory tightrope walk” (Dec. 2025) (^[60]) (^[61])
• Axios BFD (Nov. 18, 2025) – industry summit highlights (^[72])
• Insilico Medicine blog (July 1, 2023) – “First Generative AI Drug Begins Phase II” (^[9]) (^[18])
• South China Morning Post, Lilly–Insilico deal coverage (Mar. 2026) (^[73]) (translated from Swedish/English sources)
• Biopharma deal trackers (PharmInt, GlobalData) (^[14]) (^[28])
• Simply Wall St News and LinkedIn thought pieces on Lilly/Insilico (Mar. 2026) (^[69]) (for industry perspective)
• Other industry news (CNBC, FiercePharma, etc.) and SEC/HKEX filings (cited via PharmaTech and SeekingAlpha summaries) (^[71]) (^[2]).

(All cited documents link to lines as indicated by the “【source†Lx-Ly” notation.)