Thesis
Despite decades of sustained healthcare investment, the cost of finding a new medicine keeps rising. In 2025, the average cost to develop a drug from discovery to launch reached $2.7 billion, up from $2.2 billion in 2024, and clinical development alone, from phase I through regulatory filing, took over 100 months in 2024. One reason behind this cost is that drug discovery is an inefficient system built on trial and error. Traditional drug discovery methods rely on high-throughput screening, which tests libraries of existing molecules against a drug target, a method that has a maximum hit rate of 2% and costs $500K per week to run. Molecules that show promise are then optimized, as chemists search for a molecule with the right balance of properties, which are often inversely correlated. At the end of this slow and lengthy process, a drug candidate can still fail because it is not effective, toxic in humans, or not properly absorbed by humans, leading to a 90% failure rate among drug candidates that enter clinical trials.
Source: Deloitte
That may be changing. In 2020, DeepMind's AlphaFold solved the protein folding problem using deep learning, demonstrating that AI can address a 50-year-old problem in molecular biology. The capabilities of AI have since extended into molecular design, binding affinity prediction, and the modeling of a molecule's absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties. Instead of experimentally searching chemical space, AI models can computationally search for new drug candidates. One June 2023 estimate put the savings AI could deliver in drug discovery, up to the preclinical stage, at 25-50% or more of the time and cost involved. What limits these models is data. PDBbind, a curated public database of experimentally measured drug-protein structures, held only 35K structures in 2025.
Genesis Molecular AI builds AI models for small molecule drug discovery and uses them to design drugs for its own pipeline and for pharmaceutical partners. Its platform, Genesis Exploration of Molecular Space (GEMS), predicts the 3D structure, potency, and drug-like properties of candidate molecules before they are made in the lab, and filters out those with poor activity or unfavorable safety profiles. Genesis's approach to the data constraint has two parts. It trains its structure prediction model, Pearl, on proprietary synthetic structures generated by physics simulations, and its Incyte partnership gives it proprietary experimental data from Incyte's programs, which Incyte agreed to share for model training in May 2026. Genesis sells this capability to pharmaceutical partners while using it to design its own drugs.
Founding Story
Genesis Molecular AI, originally called Genesis Therapeutics, was founded in 2019 by Evan Feinberg (CEO) and Ben Sklaroff (former CTO), who were introduced through a close mutual friend while Feinberg was looking to spin out his PhD research at Stanford University. Before cofounding Genesis, Sklaroff was Director of Software Engineering at the 3D printing company Markforged, where he had been the 12th employee. Sklaroff served as CTO of Genesis for over five years before leaving the company in June 2024.
Feinberg's biotech roots began in high school, where he had interned at a local biotech company. He went on to study Applied Physics at Yale University starting in 2009. After Yale, Feinberg was attracted to the lab of Vijay Pande at Stanford University because of Folding@home, a project run out of Stanford that uses volunteers' home computers to simulate protein folding. In September 2007, Folding@home set the Guinness World Record for the most powerful distributed computing network. Feinberg studied as a PhD student in Pande's lab from 2013 to 2018 with a Blue Waters Graduate Fellowship. His PhD research focused on conducting molecular dynamics simulations of proteins and improving deep learning algorithms for predicting drug-receptor binding affinities.
The company was founded on two key papers from Feinberg's research at Stanford University, which formed the basis of GEMS. In November 2018, Feinberg and Pande co-authored a paper on PotentialNet, a neural network architecture for optimizing small-molecule binding to proteins. In April 2020, Feinberg and Pande co-authored a paper with scientists from Merck focused on improving the prediction of ADMET properties, which drug developers analyze to select compounds to advance into animal and human studies. Both technologies were patented by Stanford University in March 2019 and were licensed to Genesis. The PotentialNet patent was granted in August 2023, and the ADMET patent was granted in September 2024.
Andreessen Horowitz funded the commercialization of Feinberg's research through Pande. Pande was the first professor-in-residence at the firm, known as a16z, and became its ninth general partner in 2015, when he launched the firm's first $200 million bio fund. a16z led Genesis's seed round and co-led its Series B.
Genesis launched as Genesis Therapeutics, with Leonard Bell as its founding chairman. Bell was the principal founder and longtime CEO of Alexion Pharmaceuticals, which AstraZeneca acquired for $39 billion in July 2021. In April 2024, Shifeng Pan, the former Head of Discovery Chemistry at the Genomics Institute of the Novartis Research Foundation, joined Genesis as Chief Scientific Officer.
In October 2025, the company changed its name from Genesis Therapeutics to Genesis Molecular AI to better emphasize its use of AI. In December 2025, Genesis appointed Sergey Edunov, who had led the Core Llama team at Meta Superintelligence as a Director of AI Research, as its Senior Vice President of Foundation Models. As of September 2026, Edunov served as Genesis's Chief Technology Officer.
Product
Genesis Molecular AI's product is GEMS, which the company describes as "the AI operating system for drug discovery." GEMS combines foundation models, AI agents, and software tools that chemists use to generate candidate molecules, predict their properties, examine those predictions, and decide which molecules to synthesize. Genesis deploys GEMS on its own drug programs, run with a wet lab in San Diego, and inside pharmaceutical partners, where Genesis chemists and engineers work alongside partner teams as forward-deployed engineers. Experimental results from both settings feed back into model training.
Design Workflow
Source: Genesis Molecular AI
GEMS organizes a chemist's work into four stages:
In the generate stage, GEMS proposes new, synthesizable molecules, either at large scale during hit identification, the search for initial active compounds, or in targeted rounds during lead optimization.
In the predict stage, GEMS predicts the 3D structure of each molecule bound to its target, then uses that structure to predict potency and selectivity, alongside absorption and metabolism properties from separate models.
In the interrogate stage, chemists visualize predicted structures, filter and sort candidates, and compare them across the properties that matter for a program.
In the decide stage, chemists weigh trade-offs between properties and choose which molecules to synthesize.
In July 2026, Feinberg and Edunov described Sapphire, an internal agentic drug discovery system built on Genesis's models that iterates like a chemist, examining predicted poses, forming hypotheses, and proposing candidates for the next round. The company says its agents handle the routine steps of the workflow so that chemists can focus on design decisions.
Pearl
Source: Genesis Molecular AI
Pearl is Genesis's foundation model for protein-ligand cofolding and the model GEMS uses to predict 3D structures. Cofolding models predict the structures of proteins interacting with small molecules, nucleic acids, or other proteins, which is important for understanding how an experimental drug will bind to a target. In October 2025, Genesis released a technical paper on Pearl co-authored with scientists from NVIDIA.
The authors reported that Pearl outperformed AlphaFold 3 by 15% on the Runs N' Poses benchmark and by up to 40% on the PoseBusters benchmark. When tested on crystal structures from the company's own drug programs, Genesis reported that Pearl was several-fold more successful than the closest model at the ultra-high accuracy threshold of under 1Å.
The technical paper highlights three innovations:
Synthetic Data for Model Training. The field of drug-protein structure prediction is limited by the amount of public protein-ligand structures available. Given that neural network performance has been empirically observed to scale with model size and dataset size, this limited dataset constrains the field's ability to train general models for drug-protein structure prediction. Genesis trains Pearl on a proprietary synthetic dataset of physics-generated structures, generated through molecular dynamics, to supplement public data.
Physics-Native Architecture. One of the challenges with structure prediction is models that produce physically invalid structures. Models such as AlphaFold have been criticized01006-3) for their limited "understanding" of the physics that define biomolecular interactions. Genesis addresses these challenges through architectural and training changes. First, the company introduced an SO(3)-equivariant diffusion module, which is used to preserve the rotation symmetry of 3D data. Second, lightweight triangle multiplication (trimul) modules were used to increase the computational efficiency of the model. Third, NVIDIA cuEquivariance kernels were used to speed up inference and training time.
Pearl can be used in unconditional cofolding and pocket-conditional cofolding modes. Unconditional cofolding tests a model's ability to generate a complex structure using a protein sequence and the 2D chemical structure of a molecule. Conditional cofolding emulates a scenario where a reference structure for a target protein is available and a specific binding pocket is hypothesized. This mode allows scientists to guide model predictions using known information to test specific hypotheses.
In June 2026, Genesis reported results on OpenBind, a public benchmark of 802 drug-protein complexes for a single viral protease. Given only the protein sequence, the drug's chemical structure, and an unbound protein structure, the Pearl system, which pairs the Pearl model with additional search at prediction time and physics- and AI-based ranking, succeeded on 78% of compounds, above every cofolding model OpenBind had evaluated. At the stricter sub-1Å threshold, it succeeded on 60% of compounds, against 27% for the best individual competing model. The same month, Genesis published DeCAF-Pearl, a distilled version of Pearl with a 5x inference speedup, which the company says makes large-scale virtual screening and synthetic data generation cheaper.
Molecular Generation and Property Prediction
Source: Pande Lab
Beyond structure prediction, GEMS proposes new molecules conditioned on what a chemist is trying to achieve, including property, structural, and program-specific constraints. For potency and selectivity, it combines structure-based deep learning with physical simulation, including molecular dynamics and quantum chemistry, which the company says lets it find drug candidates for targets that lack training data. Separate multitask models predict over 30 ADME properties, including solubility, permeability, and metabolic stability.
Genesis's property prediction work began with PotentialNet, which the company licensed from Stanford University in 2019. PotentialNet is a graph convolutional neural network technique developed by the Pande Lab for molecular property prediction. It represents drug molecules as graphs. A molecule is featurized into a square adjacency matrix by describing each atom with its element, hybridization, and formal charge. The neural network then takes these features and learns the organic chemistry relevant to molecular property prediction.
Molecular property prediction is a rate-limiting step in drug discovery. A medicinal chemist manages two tasks in the drug discovery process: generating molecular ideas and ranking which of those ideas should be tested in the lab. Predicting properties accurately would derisk molecules before they are synthesized and tested in animals, which would speed up drug design and reduce the cost of drugs that fail in late-stage trials.
Drug Pipeline
Genesis uses GEMS to develop its own wholly owned pipeline of therapeutic assets. The company lists oncology and immunology as its main focus areas. The internal pipeline gives Genesis assets it can develop or license, and it tests the platform on programs the company controls.
The company's lead disclosed program targets PI3Kα, the protein encoded by PIK3CA, which is one of the most mutated genes across cancers. Mutations in PI3Kα drive uncontrolled cell growth, making it a well-validated oncology target. As of September 2026, Genesis was approaching development candidate nomination, the point at which a lead molecule is chosen for the studies required before human trials, for pan-mutant allosteric inhibitors of PI3Kα in breast and colorectal cancer. The drug is designed to selectively target mutant forms of PI3Kα while avoiding the normal, or "wild-type," version. Inhibiting wild-type PI3Kα is linked to severe high blood sugar, so a mutant-selective inhibitor could improve safety and tolerability.
The company's other targets are not disclosed. As of September 2026, Genesis listed a first-in-class oncology program designed to promote cancer cell death by inhibiting a regulator of apoptosis, programs developing oral small molecules against immunology targets that biologic drugs have already validated, and a first-in-class program against proteins central to several inflammatory signaling pathways.
Market
Customer
Genesis Molecular AI sells to large pharmaceutical companies that pay to use GEMS on targets they select, and it is developing its own drugs for patients with cancer and autoimmune diseases.
Pharmaceutical Companies: Genesis's partners have come from the top 50 pharmaceutical companies by revenue. It has announced drug discovery partnerships with Genentech, Eli Lilly, Gilead Sciences, and Incyte, and as of September 2026, its partners page listed Incyte and Gilead as its active collaborations. The partnerships are milestone-based. The partner chooses the targets Genesis works on and, in the Incyte deal, holds exclusive rights to develop and sell the resulting drugs.
Market Size
Genesis Molecular AI focuses on small molecules, which accounted for $785 billion, or 58%, of pharmaceutical sales in 2023. The global small molecule drug discovery market was estimated at $40.2 billion in 2023 and projected to reach $58.4 billion by 2030, representing a 5.5% CAGR from 2023 to 2030, driven by growth in pipeline molecules and the expiry of patents on major biologic drugs.
The two disease segments the company focuses on are oncology and immunology. Global spending on cancer medicines reached $252 billion in 2024 and is expected to reach $441 billion by 2029, and cancer is a leading cause of premature death globally. Immunology was estimated to be a $112.3 billion market in 2025. For Genesis, the size of its opportunity will depend on the specific drug targets it chooses to pursue.
As of September 2026, the only disclosed drug target was PI3Kα, which is encoded by the PIK3CA gene. Mutations in this gene are implicated in a wide variety of cancers, including an estimated 40% of patients whose breast cancer tumors are hormone receptor-positive but don't express the protein HER2. In its 2020 IPO filing, Relay Therapeutics estimated that cancers with the PI3Kα H1047X mutation affected approximately 10K late-line patients annually in the US, and that mutations at E542 and E545 together affected approximately 15K late-line and 60K total patients annually. These figures cover a single target. Genesis can use GEMS to design drugs for other targets and indications, widening the patient population it addresses.
Competition
Mutant-Selective PI3Kα Competitors
Genesis Molecular AI describes its lead program as a "best-in-class" pan-mutant PIK3CA inhibitor, meaning it aims to beat competing compounds on properties such as mutation coverage, tolerability, or potency. As of September 2026, the program was preclinical. The first approved PI3Kα inhibitor, alpelisib, which acts on both mutant and wild-type PI3Kα, was approved in May 2019. Mutant-selective PI3Kα inhibitors have shown better safety and tolerability compared to inhibition of the wild-type form of PI3Kα, leading to efforts to develop mutant-selective versions. Multiple mutant-selective PI3Kα inhibitors are in clinical trials, with Relay Therapeutics furthest ahead in Phase 3.
Relay Therapeutics: Relay Therapeutics was founded in 2016 and launched with a $57 million Series A from Third Rock Ventures. The company develops therapeutics by modeling protein motion. Its lead asset is zovegalisib (RLY-2608), an allosteric, pan-mutant-selective PI3Kα inhibitor that was in Phase 3 studies for breast cancer as of September 2026. Like Genesis's program, zovegalisib is designed to bind to a separate "allosteric" pocket on PI3Kα. In May 2026, Relay raised $316 million in a public stock offering. As of September 2026, Relay had a market cap of $4.2 billion. Relay is the closest comparison for Genesis: a computation-led company pursuing the same mechanism, with a candidate already in Phase 3.
Scorpion Therapeutics: Scorpion Therapeutics was launched in 2020 by Gary Glick, Keith Flaherty, Gaddy Getz, and Liron Bar-Peled with $108 million in financing. In January 2025, Eli Lilly agreed to acquire Scorpion's Phase 1/2 allosteric mutant-selective PI3Kα inhibitor program (STX-478) for up to $2.5 billion. Prior to its acquisition, Scorpion had raised a total of $420 million. In June 2025, Antares Therapeutics was spun out of Scorpion with $177 million in financing to advance its other programs. The deal left Eli Lilly, which partnered with Genesis in 2022, owning a clinical-stage drug with the same mechanism as Genesis's lead program.
Synnovation Therapeutics: Synnovation Therapeutics was founded in 2021 by Wenqing Yao (CEO) and Reid Huber. The founders worked together for nearly 20 years at Incyte, where Yao was Head of Discovery Chemistry, and Huber was Chief Scientific Officer. The company launched in January 2024 with a $102 million Series A at an undisclosed valuation, led by Third Rock Ventures. In March 2026, Novartis agreed to acquire SNV4818, a Phase 1/2 pan-mutant-selective PI3Kα inhibitor from Synnovation, for $2 billion upfront and up to $1 billion in milestones. The deal set a price of $2 billion upfront on an early clinical program with the same mechanism as Genesis's lead program.
Drug Discovery Platform Competitors
Isomorphic Labs: Isomorphic Labs was founded in 2021 as a DeepMind spinout led by Demis Hassabis. Hassabis co-founded DeepMind in 2010, where AlphaFold was developed. Isomorphic Labs builds a drug design engine, IsoDDE, that it describes as progressing beyond AlphaFold 3. In January 2024, Isomorphic Labs announced partnerships with Eli Lilly and Novartis, and in January 2026, it announced a partnership with Johnson & Johnson. In March 2025, it raised its first external round of $600 million, led by Thrive Capital, and in May 2026, it raised a $2.1 billion Series B, also led by Thrive Capital. As of September 2026, Isomorphic Labs had raised a total of $2.7 billion, and its valuation was not publicly disclosed. Isomorphic Labs had raised about nine times Genesis's total funding, and it works with Eli Lilly, which partnered with Genesis in 2022.
Chai Discovery: Founded in 2024, Chai Discovery builds AI models that predict and design interactions between molecules. Its Chai-2 model, released in 2025, designs antibodies from scratch. In July 2026, Chai Discovery raised a $400 million Series C led by Index Ventures at a $3.8 billion valuation, and as of September 2026, it had raised a total of $631 million. Its partners include Eli Lilly and Pfizer. Chai Discovery's lead product designs antibodies, a class of large-molecule drugs, while Genesis focuses on small molecules; the two overlap in cofolding, where Genesis benchmarks Pearl against Chai-1.
Iambic Therapeutics: Iambic Therapeutics was founded in 2019 by Tom Miller (CEO), Fred Manby (CTO), and Matthew Welborn (VP Machine Learning). The company's core technology includes OrbNet, which predicts the molecular, electronic, and energetic properties of chemicals. In November 2025, Iambic Therapeutics raised over $100 million at an undisclosed valuation to advance its platform and clinical pipeline, following clinical data for its lead drug IAM1363. As of September 2026, Iambic Therapeutics had raised a total of $396.7 million, and in September 2026, it filed for an IPO. Unlike Genesis, Iambic already has an AI-designed drug in clinical trials.
Terray Therapeutics: Terray Therapeutics was founded in 2018 by Jacob Berlin (CEO) and Eli Berlin (COO/CFO). In February 2022, Terray Therapeutics announced a $60 million Series A to advance its tNova platform, following a previously unannounced $20 million seed round co-led by Digitalis Ventures and Two Sigma Ventures. In October 2024, it announced a $120 million Series B. In February 2024, the company published a paper on COATI, a chemistry foundation model, and in August 2024, it published a paper on using latent diffusion machine learning for small molecule design. As of September 2026, Terray Therapeutics had raised a total of $225 million. In secondary markets, its valuation was estimated at $600 million as of March 2026. Terray's platform is built around generating its own experimental chemistry data at scale, whereas Genesis supplements public data with physics-generated synthetic data.
Business Model
Genesis Molecular AI earns revenue from partnerships in which pharmaceutical companies pay to use GEMS, and it is building its own drug pipeline, which it can develop or license to larger companies. There are primarily two ways the business generates revenue: direct drug sales and drug discovery partnerships.
Direct Drug Sales
Genesis can generate revenue from sales of molecules it owns or has royalties on. As of September 2026, its most advanced program, a PI3Kα inhibitor, was preclinical. For comparison, in a Canadian reimbursement review, the approved PI3Kα inhibitor alpelisib had a submitted price of C$95.2 for a 150 mg tablet, with a 28-day cycle cost of around C$5K.
Genesis can also out-license the development of parts of its drug pipeline to other companies. These deals typically involve a combination of an upfront payment, milestones, and royalties, depending on the stage of development of the asset.
Drug Discovery Partnerships
Genesis generates revenue from drug discovery partnerships with other biopharmaceutical companies. These partnerships are long-term, multi-target agreements in which the company uses GEMS to discover, design, and optimize therapeutic candidates for partners. As of September 2026, the company had publicly disclosed four drug discovery partnerships, with Genentech in 2020, Eli Lilly in 2022, Gilead Sciences in 2024, and Incyte in 2025.
Three of the original partnerships included an upfront payment, with the company receiving $20 million from Eli Lilly, $35 million from Gilead Sciences, and $30 million from Incyte. The Lilly deal was worth up to $670 million, including upfront, target nomination, and milestone payments. The original Incyte deal made Genesis eligible for "biobucks," contingent milestone payments, of up to $885 million if Incyte nominated a third target.
In May 2026, Incyte expanded its collaboration, paying $120 million in upfront consideration, made up of $80 million in cash and a $40 million equity investment, plus recurring research funding for model training and computing costs. Under the expansion, Genesis is eligible for milestone payments of up to $232 million per program, or over $1 billion across the initial five new targets, along with royalties on any approved products.
Traction
Pharmaceutical Partnerships
Between 2020 and 2025, Genesis Molecular AI signed partnerships with four of the top 50 pharmaceutical companies, and as of September 2026, it had received $150 million in upfront consideration from Incyte, $35 million from Gilead Sciences, and $20 million from Eli Lilly. As of September 2026, Genesis listed Incyte and Gilead as its active collaborations.
In October 2020, Genesis partnered with Genentech, a subsidiary of Roche, to use its platform to discover drug candidates. In May 2022, Genesis announced a partnership with Eli Lilly worth up to $670 million, with both companies agreeing on three initial targets. In September 2024, Genesis announced a partnership with Gilead Sciences covering three initial targets. In February 2025, Incyte announced a deal to use GEMS on two initial targets.
In May 2026, Incyte expanded that deal to at least five new targets. Incyte's head of research and development attributed the expansion to results on the initial two programs. On one, the companies created first-in-class chemical matter for a "very hard-to-drug, novel target," and on the other, they made progress on a target that other companies had tried and failed to make druggable.
Academic Credibility
The two foundational papers for Genesis are PotentialNet, published in November 2018 in the journal ACS Central Science, and a paper on using graph convolutions to predict ADMET properties, published in April 2020 in the Journal of Medicinal Chemistry. As of September 2026, the PotentialNet paper had been cited 521 times, and the ADMET paper had been cited 255 times. Genesis has also collaborated with scientists at MIT in the development of Boltz-1, an open-source alternative to AlphaFold 3. Two members of staff at Genesis, Ken Leidal and Wojtek Swiderski, were named as equal contributors on the paper. As of September 2026, the Boltz-1 paper had been cited 717 times.
Valuation
In May 2026, Incyte made a $40 million equity investment in Genesis Molecular AI as part of their expanded collaboration, at an undisclosed valuation. Genesis's last priced venture round was a $200 million Series B in August 2023, co-led by a16z and an unnamed life sciences investor, which did not disclose a valuation. Before that, Genesis raised a $52 million Series A in December 2020, led by Rock Springs Capital, and a $4.1 million seed round in November 2019, led by a16z, neither at a disclosed valuation.
As of September 2026, Crunchbase put Genesis's total funding at $296.1 million, the sum of those four rounds. The company said its Series B brought total capital raised to over $280 million, more than the disclosed rounds add up to, and NVIDIA made an equity investment of undisclosed size in November 2024, so the true total is likely higher. Other investors in the company include Rock Springs Capital, Menlo Ventures, T. Rowe Price, Fidelity, and Radical Ventures.
Since its Series B, Genesis's new capital has come from NVIDIA's equity investment and from pharmaceutical partners rather than from venture rounds. As of September 2026, Incyte's two deals had brought $150 million in upfront consideration, including the $40 million equity purchase. Among public companies, the closest comparable is Relay Therapeutics, which pairs a computational platform with a mutant-selective PI3Kα program. With its drug in Phase 3, Relay had a market cap of $4.2 billion as of September 2026, which reflects a later-stage pipeline than Genesis's preclinical one.
Key Opportunities
Breaking Eroom's Law with AI-Driven Discovery
The blue sky scenario for using AI in drug discovery is breaking the trend of Eroom's Law, which describes how the cost of developing a new drug roughly doubles every nine years. The name is a deliberate inversion of Moore's Law, which describes the opposite trend in semiconductors. This trend of declining productivity indicates that the tools being used by the biopharmaceutical industry are not working.
Genesis aims to address these challenges with GEMS. The company is establishing itself as a drug discovery partner to biopharmaceutical companies with large R&D budgets. The top 20 pharmaceutical companies spent $180 billion on R&D in 2024, with an increasing share of that budget directed to external innovation. The Incyte expansion shows the form this can take. In May 2026, Incyte expanded its deal to at least five new targets, with options to nominate more over time, and began sharing its own experimental data to train GEMS, in what the companies described as "among the first major pharma-AI collaborations to power large-scale foundation model training with a partner's proprietary experimental data." Each partner that follows Incyte in sharing data would add to the training data Genesis's models depend on, and expansions of this kind add targets from an existing partner without the cost of winning a new one.
Capitalizing on the Biopharma Patent Cliff
The biopharma industry is facing a revenue cliff as its best-selling drugs begin to come off patent. For example, Keytruda is Merck's best-selling drug, generating $31.7 billion in sales in 2025. However, the core patents for this drug are scheduled to expire in 2028, meaning Merck needs to replace the revenue lost when exclusivity ends. New high-impact medicines need to be developed to fill this gap, yet most biopharma companies have reduced their internal drug discovery and design capabilities, instead depending on M&A to source innovation.
Genesis can meet some of this demand through discovery partnerships, as it has with Incyte and Gilead. The company is also developing its own drug candidates, which could be sold to biopharma companies willing to pay high prices for sought-after programs. In March 2026, Novartis agreed to pay $2 billion upfront for Synnovation's early clinical PI3Kα program, the same mechanism as Genesis's lead program. In November 2025, Pfizer and Novo Nordisk entered a bidding war over Metsera, a maker of obesity drugs, which raised the purchase price from Pfizer's $7.3 billion offer to $10 billion.
Onshoring Biotech Intellectual Property
Historically, China's role in the biopharma industry has been confined to contract research, manufacturing, and early clinical development. Between 2019 and 2025, however, the industry's dependence on China for drug discovery and high-value licensing grew. In 2025, China-sourced assets accounted for 48% of biopharma licensing deal value, up from 5% between 2019 and 2021.
Source: BCG
Some in the industry are outspoken about this dependency. Noubar Afeyan, the founder of Flagship Pioneering, wrote in his 2026 annual letter:
China is well on its way to overtaking the United States' historical lead in biotechnology. China's share of global biotech patents jumped from 1% in 2000 to 28% in 2019, ahead of the U.S.' share. The number of novel medicines under development in China has skyrocketed by 8x over the past nine years and is now almost equal to total U.S. output. Ironically, U.S. investors, who might have preferred to fund life-science innovators here at home, are now placing their bets on China. U.S. and other Western investment in China has reached $150 billion over the past five years. In fact, roughly a third of all global biopharmaceutical licensing deals are now for Chinese-developed assets.
US policy has since moved toward reshoring biotech capabilities. In November 2025, President Donald Trump signed an executive order to launch the Genesis Mission, an initiative unrelated to the company that aims to boost AI innovation in the US, with biotechnology as one focus area. In December 2025, the BIOSECURE Act, which introduced restrictions tied to entities designated as "biotechnology companies of concern," became law. In March 2026, the Department of Energy announced $293 million in funding for teams using AI models to address national challenges.
Genesis could benefit both from increased federal investment in American companies applying AI to biotechnology and from restrictions on companies that source intellectual property from China. In September 2025, the Trump administration considered requiring a review by the Committee on Foreign Investment in the United States of deals in which American biopharmaceutical companies buy rights to intellectual property from Chinese biotech companies. If more "America First" policies are enacted, biopharmaceutical companies may have to turn to American biotech companies for new drug IP.
Key Risks
Increasing Technology Commoditization
The AI drug discovery market has seen numerous entrants. In the 12 months to November 2025, venture investors put $3.2 billion into 135 AI drug development startups. A 2024 study found that among 165 drugs whose developers reported using AI, 76% used it for drug molecule discovery, the stage of development GEMS serves.
Many groups are building technology that competes with Genesis's products. In protein-ligand structure prediction, two notable examples are NeuralPLexer3 from Iambic Therapeutics and IsoDDE from Isomorphic Labs, and both companies reported models that outperformed AlphaFold 3. In November 2024, Iambic Therapeutics reported that "NP3 improves protein-ligand binding structure prediction with greater accuracy than AF3." In February 2026, Isomorphic Labs reported that its IsoDDE system more than doubled the accuracy of AlphaFold 3 on a protein-ligand structure prediction benchmark.
These competitors are well funded. In 2026, Isomorphic Labs raised a $2.1 billion Series B and Chai Discovery raised a $400 million Series C, while Genesis had raised a total of $296.1 million as of September 2026. This leads to a Red Queen effect, where the company and its competitors must constantly out-innovate each other to stay ahead and raise financing. Genesis not only needs to prove the value of GEMS by delivering new drugs into the clinic, but must also differentiate itself among its competitors.
Therapeutic Crowding
Biotechnology companies like Genesis must make a trade-off between technology risk and target risk. In this context, technology risk involves validating the GEMS platform by bringing new drugs forward, while target risk involves pursuing unproven drug targets, known as first-in-class drugs.
Many companies minimize target risk and lean into known drug mechanisms to validate the technology platform they have developed. This is known as "therapeutic crowding," where companies herd into well-established mechanisms of action. As of May 2025, about 2% of active R&D targets were each associated with 50 or more drugs, and those 37 targets accounted for roughly one quarter of the preclinical and clinical R&D pipeline.
Source: L.E.K. Consulting
Genesis's lead program is a pan-mutant allosteric inhibitor of PI3Kα. PI3K inhibition is a drug mechanism with multiple approved drugs and 52 active clinical studies as of September 2026. Although Genesis is differentiating its program as a pan-mutant allosteric inhibitor, Relay Therapeutics, Eli Lilly (with the program it acquired from Scorpion Therapeutics), and Novartis (with the program it agreed to acquire from Synnovation Therapeutics) are developing similar drugs that are in later clinical stages.
Genesis needs to differentiate its drug clinically from approved PI3K inhibitors and from newer mutant-selective ones, and it will not know whether its drug is differentiated enough until the asset reaches mid- to late-stage clinical trials. Competitors in a crowded class also follow quickly. The average time for a drug class to have three FDA approvals fell from about 15 years in 1990 to 2003 to about two years in 2013 to 2021.
Lack of Data for Model Training
The models Genesis trains depend on data availability and quality. Small, incomplete, biased, or low-quality data leads to poor results. Structure prediction models like Pearl depend on the Protein Data Bank, which had accumulated over 260K experimentally determined structures between 1976 and September 2026.
To train better models, Genesis needs proprietary data from internal experiments or external partnerships. In a February 2026 interview, Marissa Baker, a director at Genesis, described how the company validates model predictions and generates training data through biophysical and cellular assays. The challenge with this approach is that data scales with the number of scientists running assays in the laboratory, which is a major operational expense. The May 2026 Incyte expansion partly addresses this, since Incyte agreed to share proprietary experimental data for training GEMS, but the arrangement makes Genesis's data supply dependent on a single partner's programs.
Genesis also supplements its training data through synthetic generation using molecular dynamics. However, the quality of this data is unproven. Molecular dynamics methods rely on parameterized "force fields," which are approximations. The company claims to be "the first to demonstrate LLM-like scaling laws" with synthetic data. Genesis's synthetic data comes from physics simulations rather than from its own models, which avoids the best-known failure mode of synthetic data: model collapse from training on generated outputs. But a model trained on simulated structures inherits the errors of the force fields that produced them, and those errors could cap how accurate the model can become without more experimental data.
Summary
Genesis Molecular AI is a Stanford University spinout that builds AI models, led by its Pearl structure prediction model, and uses them to design small molecule drugs for its own pipeline and for pharmaceutical partners. The company designed its approach to identify and optimize drug candidates faster and at lower cost than traditional high-throughput screening, and between 2020 and 2025, it signed partnerships with four of the top 50 biopharmaceutical companies, including an Incyte collaboration that was expanded in May 2026. However, the company faces meaningful risk from an increasingly crowded and better-funded AI drug discovery market, from companies pursuing the same therapeutic target with drugs already in the clinic, and from the limited supply of high-quality data to differentiate its models.
