AI deals show no sign of slowing

Artificial intelligence (AI) deals are starting to command nine‑figure cash payments, equity investments, and multibillion‑dollar milestone and royalty structures. They span from out‑licensing discovery platforms, to data‑licensing pacts and joint development efforts; but, as yet, few clinical-ready assets have been licensed from end‑to‑end AI‑enabled pipelines. Collectively, four notable trends are emerging; an increase in upfront commitments; specialization in new therapeutic modalities (such as biologics rather than small molecules); the rise of niche dataset providers; and increasing participation from mid‑ to large-cap biotechs that are following the early adopters. According to Aris Persidis, president and CEO of Biovista, “Machine learning remains in vogue. Continued deal activity and increased upfront payments simply reflect that reality.” Industry is embracing the concept that integrating machine learning (ML) into discovery can accelerate target identification (ID) and lead optimization. But while industry is evolving a more nuanced approach to AI and ML, using them to fine-tune molecules developed using traditional methods, AI-derived assets have yet to prove themselves in the clinic.

Multinational pharma seeks greater AI integration

In recent years, multinational pharma has increasingly relied on search and development paradigms or biotech acquisitions to feed pipelines, most recently from bargain-basement assets through a flurry of deals in China. Companies have gradually built internal infrastructure in the AI space, but the appetite remains for AI ‘solutions’, particularly for providers offering solutions claiming to integrate multimodal datasets (for example genomics, transcriptomics, proteomics, and patient clinical data). The technology is offered up as a potentiator of drug target ID efforts and a tool to facilitate high potency small-molecule binders, saving effort, resources and time, as well as generally accelerating the pace of discovery and preclinical development.

An example highlighted early last year was a deal by Gilgamesh Pharmaceuticals with AbbVie (Table 1) This was the first psychedelic-drug‑inspired AI collaboration and leveraged Gilgamesh’s ‘neuroplastogen’ platform—an advanced molecular-modeling technology to design derivatives of ibogaine-like molecules. The deal offered AbbVie a turnkey entry into the surging central nervous system (CNS) space with a de‑risked modality. Beyond the $65 million cash payment, AbbVie secured rights to clinical candidates in depression, post-traumatic stress disorder (PTSD), and anxiety, with cumulative milestones up to $1.95 billion and escalating royalties tied to global sales.

Joining frontrunners BigHat Biosciences, Generate:Biomedicines, Absci and Nabla Bio, Biolojic Design’s June 2024 pact with Merck KGaA focuses on multispecific antibodies and antibody–drug conjugates (ADCs) rather than small molecules (Table 1). Biolojic’s AI‑driven antibody engineering engine uses deep learning to predict complex binding interfaces and linker design simultaneously, reducing candidate triage timelines by 50%. Merck KGaA’s low‑double‑digit‑million‑euro upfront—augmented by €346 million in target‑specific milestones—underscores the strategic importance of rapid biologics prototyping in oncology and inflammation.

Unlike pure discovery deals, Ochre Bio’s agreement last June 2024 with GSK centered on proprietary human datasets (Table 1). Ochre’s human liver single‑cell transcriptomic and organ perfusion data provides critical training material for hepatology AI models. Valued at $37.5 million over three years, the data‑licensing deal grants GSK access to raw and processed data for AI model development, with GSK providing expert annotation support. This arrangement exemplifies a growing willingness among pharma to pay for high‑quality real‑world datasets that cannot be easily replicated in academic or increasingly picked over public repositories.

Generate:Biomedicines’ September 2024 collaboration with Novartis centered around the AI biotech’s generative protein design platform, powered by a combination of transformer‑based language models and graph neural networks (Table 1). The Flagship Pioneering company claims to generate de novo small molecule drugs with predetermined structural and functional constraints. Novartis’s $65 million upfront (including $15 million of equity) and potential milestone payouts exceeding $1 billion demonstrate a willingness to bet on the AI startup’s ability to generate first‑in‑class biologics. The deal also includes provisions for joint steering committee governance, ensuring both parties shape project selection and candidate progression decisions.

Finally, Creyon Bio’s April 2025 deal with Eli Lilly underscores the strategic value of AI‑designed oligonucleotides (built on an oligo dataset accrued over decades of development by Ionis Pharmaceuticals). Creyon’s platform integrates ML‑optimized chemistry with predictive pharmacokinetic modeling to design oligos with improved stability and tissue‑targeting characteristics. Eli Lilly’s $13 million upfront (split between cash and equity) and a $1 billion+ milestone pool allow the company to pursue multiple indications without conventional risk associated with outlicensing.

Each of these megadeals reflects the shift to external sourcing of pharma research and development (R&D): outsourcing early discovery projects to specialized AI firms while continuing to leverage in‑house expertise in clinical development, regulatory affairs, and commercialization. As Insilico Medicine CEO Alex Zhavoronkov puts it, “Many of the big players have made their investments in an NVIDIA cluster, they have partnered with Microsoft or Amazon, they have made their internal hires. But if they still have a funding overflow, they are seeking out external parties to get AI solution deals.”

An increasing emphasis on modalities

While the vast majority of activity in the AI space has been on small-molecule drugs, we are now witnessing a diversification of AI platforms into engines focusing on other therapeutic modalities. “Part of that may reflect the dynamics of the Inflation Reduction Act, with biologics receiving a longer grace period than small molecules,” said Zhavoronkov.

Table 2 includes several deals highlighting partnerships around new therapeutic modalities, such as antisense, peptidomimetics, and antibodies. By concentrating tailored solutions on a single modality, this subset of AI providers seeks to offer niche expertise that is attractive to biopharma partners working with them.

In each instance, the clarity of focus appears to be associated with substantial upfronts from the pharma or biotech partner, again making a strategic bet that an AI provider’s claimed expertise will offer reduced technological risk and an accelerated pace of discovery.

Biotechs get in on the action

Mid‑sized biotechs—often historically lacking in‑house AI teams—have also begun embracing external AI platforms to accelerate target discovery and small‑molecule optimization. Unlike the blockbuster‑scale up‑fronts seen with multinationals, these agreements feature more moderate upfronts ($20–30 million), paired with generous milestone and royalty terms.

Cartography Bioscience’s deal with Gilead Sciences exemplifies such a partnership (Table 1). Cartography’s ATLAS/SUMMIT platforms combine multiomic patient datasets with AI inference to map tumor‑selective antigen landscapes. Gilead’s $20 million upfront funds a multi‑target research plan in triple‑negative breast cancer and non‑small‑cell lung cancer, with clinical and commercial milestones estimated at $500 million per program. Royalties on net sales ensure Cartography benefits from successful assets.

In a more recent deal this February 2025, Genesis Therapeutics struck an agreement with Incyte by selecting up to three oncology targets to identify small-molecule candidates from the startup’s machine learning ML models. Incyte’s $30 million upfront is dwarfed by potential $295 million milestones per target, reflecting confidence that these models will yield high‑quality hits with optimized absorption, distribution, metabolism, and excretion (ADME) properties. With a built‑in steering committee and data‑sharing agreement, Incyte gains transparency into the ML workflows while preserving option rights on additional targets.

Notably, Zhavoronkov’s Insilico Medicine has continued its success outlicensing candidates (it sold its first drug lead for ubiquitin carboxyl-terminal hydrolase 1 (USP1) to Exelixis in 2023). “That is when industry really took notice,” said Zhavoronkov. This year Menarini’s Stemline Therapeutics paid Insilico a $20 million upfront with a $550+ million milestone package, granting rights to multiple assets, with tiered royalties upon commercialization. The partnership highlights a two‑way value exchange: Menarini obtains cutting‑edge assets from AI discovery, while Insilico gains further industry validation of its platform together with joint intellectual property (IP) ownership (Table 2).

In each case, the biotech partner retains downstream rights to develop, register, and commercialize candidates, with in many cases AI firms primarily being used as assembly‑line innovators claiming to be capable of delivering ‘clinic‑ready’ compounds within 12–18 months of deal signing.

Collaborations with startups and consortia

With so much hype and money changing hands, there has been a Cambrian explosion of small firms offering AI/ ML solutions, with many AI startups seeking to find biopharma partners, academic consortia or disease-focused patient foundations to validate their platforms through research collaborations. In these non‑dilutive collaborations, AI providers offer platform access in exchange for data or joint IP options, accelerating platform validation across new therapeutic areas and attract larger players down the line.

In April 2025, Elix and PRISM BioLab announced they were using Elix’s AI‑driven target engagement prediction models with PRISM’s technology for generating cyclic peptide mimetics (Table 2). Under a collaborative research and cost‑share agreement, Elix is performing computational design of peptide scaffolds, while PRISM’s high‑throughput combinatorial peptide-mimetic synthesis and screening platform is used to validate the top-ranked in silico candidates. A pilot identified three novel chemotypes inhibiting difficult protein–protein interactions, enabling an option to license the resulting IP for preclinical development.

In the CNS field, Biostate AI’s April 2025 alliance with the Accelerated Cure Project (ACP) focuses on multiple sclerosis. Biostate’s transformer‑based models are being used to analyze ACP’s longitudinal biospecimen repository, identifying early proteomic and transcriptomic biomarker signatures predictive of relapse. The joint developmen^t agreement includes milestone payments tied to model performance metrics and option rights for co‑development of digital diagnostic tools.

Similarly, in February 2025, Receptor.AI made a pact with Moexa Pharmaceuticals to optimize Moexa’s SMAD3 inhibitors—developed for oncological and fibrotic indications—with Receptor.AI’s integrated small‑molecule and peptide modeling workflows. Structuring the deal as fee‑for‑service with a predefined licensing option, Moexa secured rapid series optimization, while Receptor.AI obtained option rights on future leads, driving both platform validation and potential equity upside.

These consortia and startup agreements illustrate how AI biotech firms are extending their footprint beyond pharma, tapping biopharma startups, academic institutions and disease‑foundation ecosystems to diversify validation use cases and create alternative revenue streams.

The data brokers

High‑quality, disease‑specific datasets are foundational to building robust causal and generative ML models. As public repositories plateau and become increasingly picked over, drug hunters are turning to curated private datasets.

In April 2025, AstraZeneca and Pathos AI crystallized this trend into a $200 million multi‑year pact with Tempus. Pathos AI will develop oncology foundation models trained on Tempus’s de‑identified clinical, genomic, and imaging data from over 150,000 patients. Shared IP terms allow AstraZeneca and Pathos AI to apply models in their respective pipelines, while joint publications ensure scientific transparency. By licensing real‑world data at scale—and sharing annotation costs—AstraZeneca gains custom foundation models for causal discovery and biomarker prediction, while Tempus monetizes its data assets without direct drug development risk.

Similarly, Ochre Bio’s contract with GSK highlights the rising value of human single‑cell datasets across organ systems. As AI models demand higher granularity, organ‑specific and perfusion‑based datasets fetch premium license fees, shifting data from being a cost center to being a revenue generator for AI firms.

A maturing sector

AI tools have gone from a ‘nice-to-have’ to a ‘must-have’ for drug discovery organizations. “I don’t think it’s controversial anymore to say that ML is now an integral part of the target ID and preclinical discovery endeavor,” said Persidis. Average upfronts have jumped from $5–15 million before 2024 to $20–65 million in the past year. Milestone structures often exceed $1 billion, compared with sub‑$500 million pools in previous years, also reflecting growing confidence in these tools. The participation of companies beyond big pharma, such as Gilead, Incyte and Menarini, is also a sign of maturity.

Although small in number, a few cash-rich AI tech-bio companies are in-licensing assets in the other direction—a trend started by Benevolent AI in 2019 when it took assets from Janssen Pharmaceutica. This is now also extending to advanced medicinal products; in March 2025, Hologen AI formed Hologen Neuro AI, a joint venture with gene therapy developer MeiraGTx to take forward a clinically tested adeno-associated viral gene therapy asset. In the deal, Hologen AI provides $200 million upfront and an annual funding commitment in return for a minority stake in MeiraGTx’s manufacturing subsidiary, which it will optimize using its models. Several other deals have applied AI solutions to developability and biotech manufacturing. And with the current icy financing environment shortening cash runways for biotechs, cash-rich tech-bio companies may continue to find opportunities for in-licensing and joint venture activity.

The real currency, however, is data. Deals like those between Tempus and AstraZeneca or Ochre and GSK (Table 1) demonstrate how high‑quality, disease‑specific datasets are becoming as valuable as discovery IP, underpinning next‑generation causal and foundational models. Looking ahead, as AI platforms continue to mature and demonstrate reproducible lead generation and preclinical validation, deal volumes and values are poised to increase further. Strategic alliances will increasingly blend generative, causal and foundational AI approaches with diverse types of biological, chemical, toxicological and clinical data. And industry-wide adoption of AI-focused drug discovery is firmly underway.