What are the main business models in AI drug discovery?
In our AI in drug discovery market deck, you will find everything you need to understand the market
SUMMARY
The main business models in AI drug discovery are software, discovery services, pharma partnerships, drug licensing, proprietary pipelines, data and automated labs, co-development, and venture creation. In practice, the strongest model today is usually a hybrid built around pharma partnerships rather than a pure software or pure biotech strategy.
The same “AI drug discovery” label can hide completely different economics. Schrödinger looks partly like a software company, Recursion looks much closer to a partner-funded biotech, and Insilico mixes software, licensing, co-development and its own pipeline.
Software is already a real business, but it is not yet the sector’s default. Schrödinger has shown that recurring computational tools can reach meaningful scale and strong gross margins, while Insilico’s much smaller software revenue shows how difficult that model is to reproduce.
Discovery services bring in money earlier, but they become much more interesting when project work opens the door to milestones, royalties or rights in the resulting drug. A fixed research fee alone leaves most of the eventual value with the customer.
Pharma partnerships currently offer the cleanest trade-off between funding and upside. They let AI companies get paid during discovery, push much of the clinical and commercial spending onto larger partners, and still retain milestone and royalty exposure if a program succeeds.
The giant partnership numbers are easy to misread. Across the nine large transactions reviewed here, more than $36 billion of announced potential value corresponded to only about $777 million of disclosed upfront or near-term consideration, so most of the headline value remains contingent.
Drug licensing becomes more valuable as a program is derisked. A molecule with preclinical or human data can command a much larger upfront payment than access to an algorithm, which makes the timing of a license one of the most important economic choices these companies make.
Keeping drugs internally preserves the biggest possible payoff, but it quickly turns an AI platform into a capital-intensive biotech. AI may make discovery faster or cheaper; it does not remove the cost of clinical trials, regulatory work, manufacturing or commercialization.
Proprietary data, automated laboratories and venture creation can all become businesses in their own right, but their bigger role is often to strengthen the core discovery engine. They create useful extra monetization paths, though none currently looks like the center of the sector.
The direction of travel is clear: as AI-originated drugs produce stronger human data, the best platforms should be able to capture more value from the drugs themselves. Until that happens at scale, the most convincing model remains a selective hybrid that collects software or research revenue where possible, uses pharma capital for much of the risk, licenses some assets after they gain value, and keeps only a limited number of high-conviction drugs.
This market map, featured in our AI in drug discovery market deck, highlights top companies and startups in the AI in drug discovery market
Why are AI drug discovery business models so hard to compare today?
AI drug discovery companies are much easier to understand once we stop treating them as one industry: some mainly sell software, some get paid by pharma to run research, and others are effectively biotech companies building drugs of their own.
The difference has become much clearer lately because several companies now have enough commercial and clinical history to show what their business actually looks like. Schrödinger generated $199.5 million of software revenue in 2025, around 78% of its total revenue. Recursion generated $74.7 million that year primarily from collaborations. Insilico Medicine made only $4.9 million from software while most of its revenue came from drug discovery and pipeline development.
Their clinical strategies are just as different. Insilico has taken rentosertib, developed against an AI-identified target, into Phase III. Recursion is running its own clinical programs while working on large portfolios with Roche/Genentech and Sanofi. Isomorphic Labs has signed discovery partnerships with Lilly, Novartis and, more recently, Johnson & Johnson while also building an internal pipeline.
So when people ask how AI drug discovery companies make money, “AI” tells us surprisingly little. The useful question is who pays for the research, what exactly gets sold, and how much ownership the AI company keeps if a drug eventually succeeds.
What does an AI drug discovery company actually sell?
An AI drug discovery company can sell anything from a software seat to the rights to an actual drug, and those products have completely different economics.
Schrödinger can license computational tools directly to pharmaceutical researchers. A company such as Recursion can enter a multi-year partnership in which a pharma company pays for access to its discovery system and the programs produced by it. Insilico can license Pharma.AI, perform research for a partner, out-license a molecule, co-develop another molecule and keep other drugs inside its own pipeline.
XtalPi stretches the definition further because customers can also pay for automated experimentation and robotic laboratory capacity. Schrödinger has created another route through companies such as Nimbus, Morphic and Ajax, where its technology and discovery work produce equity ownership and downstream drug economics.
We therefore found seven business models that matter in practice. Several companies use four or five of them at the same time.
| Business model | What is being sold | How money comes in |
|---|---|---|
| Software | Access to computational drug-discovery tools | Subscriptions and licenses |
| Discovery services | Research performed for a customer | Project fees and research funding |
| Pharma partnerships | Access to a discovery platform across several targets | Upfront payments, research funding, milestones, royalties |
| Drug licensing | Rights to a specific molecule or program | Upfront payments, milestones, royalties |
| Proprietary pipeline | Drugs kept inside the company | Partnering income later or eventual drug sales |
| Data and automated labs | Proprietary datasets, experiments and infrastructure | Access fees, services and licensing |
| Venture creation | Equity in companies built around discovered drugs | Equity gains, milestones and royalties |
As this chart shows, and as featured in our AI in drug discovery market deck, search interest in AI drug discovery has grown rapidly
Can AI drug discovery really work as a software business?
Yes, AI drug discovery can support a real software business, but Schrödinger currently looks much more like the exception than the template.
Schrödinger is the cleanest example we have. Its 2025 software revenue reached $199.5 million, software gross margin was 74%, and the top 20 pharmaceutical companies accounted for $80.8 million of annual contract value. Its latest quarterly results also showed trailing four-quarter ACV reaching $208 million.
The company is currently moving customers toward hosted software contracts, which temporarily makes reported revenue look weaker than the underlying contract base. In its latest quarter, software revenue fell 10% year over year while ACV grew 27%. Hosted products had reached 47% of quarterly software revenue. Schrödinger has also begun deploying Bunsen, its agentic AI co-scientist, with Bristol Myers Squibb, showing that the software model is still expanding even as pharma builds more AI internally.
Insilico shows how hard this is to replicate. Pharma.AI had 181 subscription customers in 2025 and software revenue grew by almost 24%, yet software still generated only $4.9 million, or 8.7% of company revenue. Bigger checks came from drug programs and pharmaceutical partnerships.
Software works particularly well when a customer can use the product repeatedly without asking the vendor to run every experiment. Once the AI company also has to provide scientists, chemistry, wet-lab validation and program management, the business starts moving toward research services and biotech economics.
If you want more recent data on this point, please see our latest AI in drug discovery market report.
When does an AI drug discovery company become more like a CRO?
AI drug discovery starts looking like a CRO when customers mainly pay the company to do scientific work for them.
A pharmaceutical company might arrive with a target and pay for hit identification, molecule generation, lead optimization or experimental validation. The AI platform helps perform that work, but revenue still depends on delivering projects. More projects usually mean more laboratory activity and more scientific staff.
Absci provides a fairly clean example. The company has said that essentially all of its historical revenue came from partnered drug-creation activities. In the first half of 2025, its revenue consisted of technology-development fees for performing drug-creation work rather than downstream milestone or royalty income.
XtalPi also has a large project-delivery component. Its drug-discovery-solutions revenue reached RMB537.9 million in 2025, up sharply from the previous year, with revenue coming from a mix of larger collaborations, antibody work and discovery projects.
There is nothing inherently weak about this model. Customers start paying long before clinical success, and the AI company avoids financing every molecule itself. The ceiling becomes more obvious when revenue grows mainly by adding people and laboratory work.
The more attractive version is usually a project that opens the door to milestones, royalties or rights in the resulting drugs. Then research services become the first layer of a much larger economic relationship.
This chart, featured in our AI in drug discovery market deck, shows annual venture capital investment in AI drug discovery startups
Why do pharma partnerships dominate AI drug discovery right now?
Pharma partnerships dominate AI drug discovery because they let AI companies keep some drug upside while handing much of the expensive downstream development to companies that already know how to run it.
Recursion's Sanofi partnership is a good example. Sanofi paid $100 million upfront for a relationship covering as many as 15 programs. Recursion can earn discovery, development, regulatory and commercial milestones plus royalties, while Sanofi takes responsibility for much of the later development and commercialization.
The Roche and Genentech relationship goes further. Recursion received $150 million upfront, and the agreement can cover as many as 40 programs. The potential milestones exceed $300 million per program. The collaboration has also moved beyond broad platform access: Genentech has now exercised an option on the first neuroscience target generated through the work and moved it into a joint early-discovery program.
Isomorphic Labs follows the same logic from a different starting point. Lilly and Novartis signed multi-target collaborations worth nearly $3 billion in potential payments, and Johnson & Johnson has since added another multi-target partnership covering several drug modalities. Isomorphic does much of the computational design while the pharmaceutical partners bring experimental, development and commercialization capabilities.
For an AI company, this structure solves a very practical problem. Drug discovery may be where its technology is strongest, while Phase III trials, manufacturing plants, regulatory organizations and global sales forces belong to a completely different scale of business. Partnerships let each side pay for the part it knows best.
If you want more recent data on this point, please see our latest AI in drug discovery market report.
Are the billion-dollar AI drug discovery deals really worth billions?
Usually no: the giant numbers attached to AI drug discovery partnerships are maximum milestone values, while the amount of money guaranteed at signing is much smaller.
We checked nine large disclosed transactions involving Recursion, Schrödinger, Isomorphic Labs, Generate:Biomedicines, XtalPi and Insilico. Together, they represent more than $36 billion of advertised potential value. The disclosed upfront or near-term consideration adds up to roughly $777 million.
That is about 2.2 cents of relatively certain upfront or near-term money for every dollar in the headline deal values.
The gap is especially large in multi-program platform partnerships. Recursion's Roche/Genentech collaboration can generate more than $12 billion in program milestones across its full scope, while the upfront payment was $150 million. XtalPi's DoveTree agreement carried nearly $6 billion of possible milestone value after a $51 million initial payment. Generate received $50 million upfront from Amgen for a collaboration initially advertised at $1.9 billion.
These headline figures still tell us something useful: pharma is willing to commit large future payments if the programs work. But adding all the “up to” values together and calling them revenue would give a wildly misleading picture of the industry.
| Selected partnership | Upfront or near-term payment | Maximum announced value |
|---|---|---|
| Recursion – Roche/Genentech | $150M | >$12B in program milestones |
| Recursion – Sanofi | $100M | >$5B in potential milestones |
| Schrödinger – Novartis | $150M | ~$2.4B including upfront and milestones |
| Isomorphic – Lilly + Novartis | $82.5M combined | Nearly $3B |
| Generate – Amgen | $50M | $1.9B |
| XtalPi – DoveTree | $51M initial | ~$5.9B in milestones |
| Insilico – Lilly | $115M | ~$2.75B |
| Insilico – SK Biopharmaceuticals | ~$18M upfront / near term | >$2.5B |
| Insilico – Takeda | ~$60M initiation / near term | ~$600M |
This chart, featured in our AI in drug discovery market deck, shows how Shrödinger is positioned in AI drug discovery
How do AI drug discovery milestones and royalties actually pay out?
AI drug discovery partnerships usually pay in stages, which means the biggest part of their theoretical value sits years in the future and disappears if the drugs fail.
A company might first receive an upfront payment. Research funding can then cover work performed during discovery. Additional payments arrive when a drug candidate is selected, enters IND-enabling studies, starts human trials, reaches later clinical phases, gets approved or passes sales targets. Royalties only begin if a product is eventually sold.
Schrödinger's Novartis agreement makes the structure easy to see. Schrödinger received $150 million upfront. It can later receive as much as $892 million tied to discovery, development and regulatory progress, another $1.38 billion from commercial milestones, and royalties ranging from mid-single digits to low double digits.
At the end of 2025, Schrödinger had not yet recognized any of those potential milestone payments. The agreement could eventually become extremely valuable, but the $2 billion-plus of future payments were still contingent.
Collaboration revenue can therefore jump one year and fall the next without telling us much about whether the underlying platform got better or worse. Insilico offers a recent example: its preliminary first-half 2026 figures pointed to more than $100 million of revenue and a swing into profit, largely after a burst of licensing, co-development and research deals. That revenue profile is much closer to biotech deal-making than to predictable SaaS.
Why license an AI-discovered drug instead of just selling the AI platform?
Licensing an actual AI-discovered drug can bring in much more money because the buyer is paying for a partially derisked medicine rather than access to a tool.
Insilico's deal with Exelixis shows the jump in value. Exelixis paid $80 million upfront for global rights to XL309, an AI-enabled USP1 inhibitor that had already reached clinical development. Menarini paid $20 million upfront for another Insilico oncology program, with total potential payments above $550 million.
The company's much larger Lilly transaction pushes the idea further. Lilly agreed to pay $115 million upfront for a collaboration that includes exclusive worldwide rights to several preclinical oral therapeutic programs and additional discovery work. Total milestones can bring the deal to roughly $2.75 billion, plus royalties.
These checks are larger because the buyer can inspect far more than an algorithm. There is a molecule, experimental data, pharmacology, toxicology work and sometimes human evidence.
Co-development sits between licensing the whole asset and keeping it alone. Insilico, for example, has agreed to share global rights 50/50 with Hygtia Therapeutics on an NLRP3 program. The company keeps more potential upside but must also keep funding and managing part of the development.
The timing of a license therefore matters enormously. Selling a program after target discovery produces cash earlier. Waiting until preclinical or clinical validation can increase its price dramatically, provided the drug survives long enough to get there.
This chart, featured in our AI in drug discovery market deck, shows annual funding in AI drug discovery startups
Why do AI drug discovery companies keep some drugs for themselves?
AI drug discovery companies keep proprietary drugs because selling every program early would leave most of the value created by a successful medicine with somebody else.
A software license might be worth millions of dollars. A research partnership can be worth tens or hundreds of millions upfront. A successful medicine can generate billions in lifetime sales. If an AI platform genuinely creates better drug candidates, keeping at least a few of them is the obvious way to preserve exposure to that outcome.
There is another reason: proprietary drugs give customers and investors a much harder test of the platform. Recursion's REC-4881 produced preliminary Phase II evidence in familial adenomatous polyposis, including a 43% median reduction in polyp burden after 12 weeks among evaluable patients at the reported dose. That clinical result tells us more about Recursion's discovery engine than another presentation about model accuracy.
Insilico has made the same bet with rentosertib by taking the program much deeper into development instead of licensing it at the discovery stage.
But keeping drugs changes the company. Clinical trials consume cash, timelines stretch over years, and biological failures can wipe out large investments regardless of how impressive the AI looked during discovery. That is why even companies with big internal pipelines keep signing external partnerships at the same time.
Can proprietary data and automated labs become businesses of their own?
Yes, proprietary data and automated labs can generate revenue directly, although today their bigger value often comes from making the rest of the AI drug discovery business better.
Recursion has built enormous proprietary biological datasets through automated experimentation. Its Roche/Genentech work included a whole-genome CRISPR knockout map built from experiments involving a subset of more than one trillion internally manufactured iPSC-derived neuronal cells. Pharma partners can pay for maps, targets and programs generated from that system.
What customers want here goes well beyond a static dataset. They are paying for an experimental engine that can generate new biological information and connect those measurements to models that suggest what to do next.
XtalPi has taken the laboratory side further. The company combines AI with automated synthesis, experimentation and robotics, and its intelligent-solutions business generated RMB264.7 million in 2025, up 62.6%. That made the laboratory and automation business meaningful alongside its much larger drug-discovery segment.
These businesses also create a useful feedback loop. More experiments produce proprietary data. The data can improve the models. Better models can produce better candidates, which make the partnership and licensing businesses more valuable.
The catch is cost. Robots, labs and scientists are harder to scale than software. A company monetizing physical experimentation should not be given SaaS economics simply because AI controls part of the workflow.
This chart, featured in our AI in drug discovery market deck, compares the main business model options for AI drug discovery biotech companies
Can AI drug discovery companies make money by spinning out new biotechs?
Yes, spinning discoveries into separate biotech companies can produce large returns without forcing the AI platform to finance every clinical program on its own balance sheet.
Schrödinger has used this model particularly well. It has helped create companies including Nimbus, Morphic and Ajax Therapeutics, contributing technology and drug-discovery work while receiving equity and, in some cases, royalties or other downstream rights.
The exits show how different these economics can become from normal software revenue. Lilly acquired Morphic for approximately $3.2 billion in 2024. Schrödinger received about $48 million for its shares while keeping low-single-digit royalties on selected programs.
Lilly has since acquired Ajax in a transaction worth as much as $2.3 billion to Ajax shareholders. Schrödinger received roughly $47 million from its equity stake and another $10 million collaboration milestone at closing, while preserving the possibility of additional contingent payments.
This structure lets outside investors finance the biotech risk. The AI platform can keep building software and discovery technology while still owning a piece of the drugs it helped create.
Returns will obviously be irregular. Most spinouts will never produce multibillion-dollar exits. But one strong exit can be worth several years of ordinary software revenue, which makes venture creation a meaningful part of the model rather than a side experiment.
Which AI drug discovery business model uses the least capital?
Software uses the least capital once the platform exists, while partner-funded drug discovery gives companies the best route to meaningful drug upside without paying for every clinical trial themselves.
Schrödinger shows why software is attractive. Once computational products are developed, the company can sell them repeatedly across pharmaceutical organizations with gross margins around 70% or higher. A new software customer does not require a new Phase II trial.
Partner-funded discovery is more expensive to operate, but pharma helps absorb those costs. Recursion's collaborations are structured so partners fund much of the direct discovery work while Recursion keeps access to milestones and royalties.
Wholly owned clinical development sits at the opposite end. Recursion spent $475 million on R&D in 2025 and used $372 million of operating cash. Its latest quarterly update showed the company still holding $556.8 million of cash and restricted cash while targeting less than $375 million of cash operating expenses for 2026. That is the financial profile of a heavily funded clinical biotech, even though the company also owns an AI platform.
The reason is simple: AI can make parts of discovery cheaper without making a large human trial cheap. Manufacturing, regulatory work and commercialization also remain expensive.
For most AI drug discovery companies, funding every promising molecule internally would erase much of the capital advantage they claim AI creates in the first place.
| Business model | Capital needed | Revenue visibility | Potential drug upside |
|---|---|---|---|
| Software | Low after platform development | High | Low to moderate |
| Discovery services | Moderate | Moderate | Low |
| Pharma partnerships | Moderate | Moderate but lumpy | Moderate to high |
| Drug licensing | High before the deal | Low and event-driven | High |
| Co-development | High | Low | Very high |
| Wholly owned pipeline | Very high | Very low before launch | Maximum |
If you want more recent data on this point, please see our latest AI in drug discovery market report.
This chart, featured in our AI in drug discovery market deck, shows revenue breakdown by customer segment in the AI in drug discovery market
Which AI drug discovery model keeps the most upside?
Keeping a drug through commercialization gives an AI company the biggest possible payoff, but it is unlikely to be the best decision for every molecule.
If an internally discovered drug eventually generates $2 billion a year in sales, owning 100% obviously beats collecting a 5% or 10% royalty. The problem is everything that has to happen before those sales appear.
A company holding ten programs deep into clinical development has to finance ten separate bets. Some will fail because of efficacy, some because of safety, some because competitors arrive first, and some because the commercial market ends up smaller than expected.
Licensing turns part of that uncertain future value into cash now. It also lets another company pay for the expensive stages while the original developer keeps milestones and royalties.
Insilico's strategy is a good example of this portfolio logic. It has licensed programs to companies including Exelixis, Menarini and Lilly, entered research partnerships with several other pharma groups, co-developed selected assets and kept some programs internally.
The sensible goal is to keep the drugs where internal conviction and potential value justify the spending. Trying to maximize ownership across the entire pipeline can easily leave a company with maximum upside on paper and a financing problem in reality.
Are hybrid business models becoming the default in AI drug discovery?
Yes, hybrid business models are becoming the default because one monetization model rarely makes sense across every stage of drug discovery.
Schrödinger currently mixes software, discovery partnerships, proprietary therapeutics, royalties and stakes in spinout companies. Recursion combines pharma collaborations with internal clinical programs and proprietary biological data. Insilico mixes Pharma.AI subscriptions, research partnerships, licensing, co-development and internally developed drugs. XtalPi adds automated laboratories and robotics to the mix.
This convergence follows the economics of the pipeline. Standardized tools can be sold repeatedly as software. Partner programs bring in outside funding. Valuable molecules can be licensed once they have been derisked. A few high-conviction assets can stay inside the company for longer.
Pure software gives away most of the value if the software creates a blockbuster medicine. Pure biotech forces the platform owner to finance years of clinical development. A service-heavy model brings in revenue sooner but can leave the company doing a lot of work for relatively limited downstream economics.
The hybrid approach lets management choose differently for each program.
There is a cost, though. Running enterprise software, automated labs, pharma alliances and clinical trials inside one organization is messy. These businesses hire different people, use capital differently and move at different speeds. Hybrid economics look attractive on paper; building a company that can execute all of them is harder.
If you want more recent data on this point, please see our latest AI in drug discovery market report.
This chart, featured in our AI in drug discovery market deck, shows how AI drug discovery platform technology has evolved over time
Does clinical success change the AI drug discovery business model?
Yes, clinical success should gradually push the best AI drug discovery companies toward making more money from drugs and less from simply selling access to AI.
At the beginning, customers are buying a claim: the platform might find better targets or molecules faster. Software licenses and research partnerships make sense because neither side yet knows how much medical value the technology will create.
Human data changes the conversation. Once an AI-originated molecule has shown safety, pharmacology and efficacy in patients, pharma can value it as a drug rather than mainly as evidence that an algorithm works.
We are getting closer to that point, although the industry still has something important left to prove. Research on early AI-discovered clinical programs found Phase I success rates around 80% to 90%, well above historical drug-industry averages, but Phase II success of roughly 40% looked much closer to conventional development. The samples were also small.
As we saw above, rentosertib has now reached Phase III, giving the field its most advanced test of whether an AI-originated program can survive the full development process. Recursion has produced separate clinical proof-of-concept from its internal pipeline, and other AI-enabled programs are moving deeper into human trials.
No fully AI-discovered and AI-designed drug has received FDA marketing approval so far. That keeps the industry's biggest claim open. If several such medicines eventually reach the market, successful platforms will have far more bargaining power to retain programs longer and demand better economics when they do partner them.
Which AI drug discovery business models are actually working today?
The business model working best in AI drug discovery today is a hybrid built around pharma partnerships: companies get paid early for discovery work, keep milestone and royalty exposure, license selected drugs after they gain value, and retain only a limited number of programs for themselves.
Software is already proven as a business, especially at Schrödinger, but we have little evidence that pure SaaS will become the dominant model across AI drug discovery. Much of the most valuable knowledge stays tied to proprietary experiments, scientists and molecules.
Discovery services also clearly generate revenue. Their weakness is upside. If an AI company gets paid a fixed research fee and the customer later sells a blockbuster drug, most of the value leaves with the customer.
Strategic pharma partnerships currently offer the strongest compromise. Upfront payments and research funding reduce financing pressure, while milestones and royalties preserve exposure to successful drugs. The amount advertised in these deals needs to be treated carefully, since our sample showed that only about 2% of headline value was represented by disclosed upfront or near-term payments.
Drug licensing becomes especially powerful after a program has gained enough evidence to command a serious price. Co-development lets a company keep even more economics when it has enough capital and conviction.
Keeping drugs internally offers the largest theoretical payoff and the largest financial risk. AI has accelerated parts of discovery, but clinical development still consumes enormous amounts of cash and remains biologically uncertain.
So the main business models in AI drug discovery are already visible. The more interesting question now is how companies combine them. The strongest businesses are learning where to collect predictable software or research revenue, where to let pharma finance the risk, and where a drug looks valuable enough to keep.
| AI drug discovery model | Examples | What it does best | Our current view |
|---|---|---|---|
| Software | Schrödinger, Insilico | Generates recurring revenue from standardized tools | Proven, but unlikely to dominate the whole sector |
| Discovery services | Absci, XtalPi, parts of Insilico | Monetizes scientific work early | Real business, limited upside alone |
| Pharma partnerships | Recursion, Isomorphic, Generate, Schrödinger, Insilico | Uses pharma capital while preserving milestones and royalties | The core model today |
| Drug licensing | Insilico and partnered assets across the sector | Converts scientific progress into large upfront payments | Becoming more important as pipelines mature |
| Proprietary pipeline | Recursion, Insilico, Generate | Keeps the largest share of a successful drug | Highest upside and highest capital burden |
| Co-development | Selected Insilico programs and similar structures | Keeps more ownership while sharing costs | Attractive when used selectively |
| Data, robotics and venture creation | Recursion, XtalPi, Schrödinger | Creates additional ways to monetize the discovery engine | Useful secondary models rather than the center of the market |
If you want more recent data on this point, please see our latest AI in drug discovery market report.
In our AI in drug discovery market deck, we identify pain points entrepreneurs should prioritize
OUR METHODOLOGY
This analysis asks which AI drug discovery business models are actually working today. We compare companies by what they sell, who pays them, when the money arrives, how predictable that revenue is, how much capital the model requires, how much downstream drug value the company retains, and how far the underlying technology has been validated commercially or clinically.
We prioritized the freshest decision-useful evidence available, especially 2025 financial results and 2026 updates. The core inputs were reported revenue mix, software contract metrics, partnership structures, upfront and milestone payments, licensing transactions, R&D spending, cash use, clinical progression and actual human data.
We kept recurring software revenue, research funding, upfront payments, milestones, royalties and potential future drug sales separate rather than treating them as equivalent. A multi-billion-dollar partnership ceiling is not the same thing as cash paid at signing, and a wholly owned clinical asset creates far more potential upside than a software contract while also demanding far more capital.
We also treated clinical progress as part of the business-model evidence. Human data can change what a drug program is worth, how long a company can justify keeping it internally, and how much bargaining power it has in a licensing or co-development deal. Early success rates were treated cautiously because the available samples remain small.
No single company or transaction determined the conclusion. We compared software-led companies, service-heavy businesses, partner-funded platforms, licensing strategies, proprietary pipelines, automated-lab models and venture-creation structures, then looked for patterns that held across several companies rather than relying on one company's narrative.
Key sources include Schrödinger's FY2025 results, Schrödinger's Q2 2026 results, Recursion's FY2025 results, Recursion's 2025 Form 10-K, Recursion's Q2 2026 update, Insilico Medicine's 2025 annual results, Insilico's rentosertib Phase III announcement, XtalPi's 2025 annual report, Isomorphic Labs' Lilly and Novartis partnership announcement, Generate:Biomedicines' Amgen collaboration announcement, and Drug Discovery Today for the early Phase I and Phase II success-rate comparison, and Nature Biotechnology for broader industry context on the still-unproven clinical performance of AI-generated drugs.
This chart, featured in our AI in drug discovery market deck, shows revenue breakdown by region across Europe, Asia, North America, Africa, and South America in the AI in drug discovery market
Related blog posts
- AI in Drug Discovery: what are the top startups?
- How large is the AI in drug discovery market in reality?
- What does the AI drug discovery startup landscape look like today?
