Is the AI in Drug Discovery Market growing now?

What do we actually mean by the AI in drug discovery market today?

The AI in drug discovery market today covers companies and pharma programs using AI to find therapeutic targets, design or optimize molecules, and choose drug candidates before human trials.

We keep the definition relatively tight because AI now appears almost everywhere in pharma. Clinical-trial recruitment software, medical chatbots, manufacturing optimization and commercial analytics may all use AI, but including them would make the drug discovery market impossibly broad.

The companies we care about are closer to the molecule itself. Insilico Medicine uses AI for target discovery and molecular design. Chai Discovery builds models for designing biological molecules. Recursion combines large biological datasets, machine learning and automated experiments. Schrödinger sells computational tools used to predict molecular behavior. XtalPi combines AI, physics and robotics to help pharma companies discover and optimize drug candidates.

These companies also make money in very different ways. Some sell software, some sign research partnerships with pharma, and some use their own AI platforms to create drugs that they hope to license or eventually commercialize. So there is no single revenue figure that perfectly captures this market. We need to look at several forms of real activity together.

Is AI drug discovery really growing now, or does it just look louder?

AI drug discovery is genuinely growing now because pharma adoption, investment, paid usage and clinical activity are all moving at the same time.

Several fresh developments make the current picture especially hard to dismiss. Novo Nordisk recently chose AWS as its preferred cloud and strategic AI partner and created a joint hub aimed partly at shortening the path from target discovery to first human dose. Bristol Myers Squibb has lately joined Lilly, Pfizer and Novartis in working with Chai Discovery. Genentech has exercised its first validated-target option from its neuroscience collaboration with Recursion. Schrödinger's latest quarter showed 27% growth in annual contract value.

Those developments measure different things, which is useful here. One tells us that a huge pharmaceutical company is building AI into its own research infrastructure. Another shows a young AI model company collecting repeat Big Pharma customers. Another shows a pharma partner advancing an AI-generated target. Schrödinger gives us a direct commercial measure.

The harder question now is how much of this growth represents better drug discovery rather than simply more spending on the possibility of better drug discovery. We already have a clear answer to the first question. The second is still open.

If you want more recent data on this point, please see our latest AI in drug discovery market report.

Are pharma companies actually spending serious money on AI drug discovery?

Big pharmaceutical companies are spending serious money on AI drug discovery now, and some commitments have become large enough to affect how they run research.

Lilly and Nvidia are jointly committing up to $1 billion over five years to an AI drug discovery laboratory combining researchers, compute, robotics and automated experimentation. Lilly already has substantial internal AI infrastructure, so this is a deeper commitment than buying access to another software product.

Takeda has taken a different route. Its collaboration with Iambic gives the pharma company access to AI models such as NeuralPLexer and to Iambic's automated wet-lab capabilities. Iambic can receive payments exceeding $1.7 billion if the programs succeed.

The cleaner commercial example may be Incyte's expanded relationship with Genesis Molecular AI. Incyte agreed to provide proprietary experimental data for training Genesis' GEMS platform, while Genesis is set to receive $120 million before potential success-based payments and royalties.

We therefore see three spending models already operating at scale: building AI infrastructure internally, paying AI companies to work on specific drug programs, and paying for models trained directly on proprietary pharmaceutical data. Pharma companies are experimenting with all three because they still do not know which model will produce the best economics.

Are AI drug discovery partnerships actually accelerating?

AI drug discovery partnerships are still expanding, although pharma companies have lately become much tighter about how much cash they pay upfront.

DealForma counted 84 AI/ML drug discovery and licensing partnerships in 2024 and 114 in 2025, a 36% increase. Potential deal value jumped from $11.8 billion to $43.4 billion over the same period.

Activity stayed high into 2026. DealForma recorded 70 broader AI and machine-learning biopharma partnerships in the first half of the year, carrying $45.9 billion in potential value and $1.9 billion in upfront commitments.

The second quarter was noticeably more cautious. Thirty-two partnerships were signed, versus 38 in the first quarter, while median upfront payments fell from $78 million to $15 million. Headline potential value stayed high, but buyers pushed more of the economics toward future milestones.

Insilico's deal with SK Biopharmaceuticals shows how extreme that difference can become. The agreement carries potential value of about $2.6 billion, yet only $5 million was due upfront. We would badly overstate current spending if we treated every $2 billion press release as $2 billion of market revenue.

Partnership growth is still real. Pharma companies are simply getting more disciplined about when the biggest payments are released.

AI/ML partnership activity	Earlier period	More recent period	Change
Drug discovery/licensing deals	84 in 2024	114 in 2025	+36%
Potential value	$11.8B	$43.4B	About 3.7x
H1 2026 broader AI/ML partnerships	—	70	$45.9B potential value
Median upfront, quarterly	$78M in Q1 2026	$15M in Q2 2026	Sharp decline

Is AI drug discovery funding really growing, or is one giant round fooling us?

AI drug discovery funding is growing, but the typical startup is seeing nothing close to the explosion suggested by the headline dollar total.

Our pure-play funding tracker found about $1.54 billion of disclosed equity funding in 2024 and $1.84 billion in 2025. By early July 2026, the total had already reached roughly $2.42 billion.

Deal count also moved up. We identified 16 disclosed financings in 2024 and 28 in 2025. By early July 2026, 19 companies had already raised, compared with 12 during the equivalent period a year earlier. So more companies are getting funded, rather than one company creating the entire increase.

The distribution of the money is much more extreme. Isomorphic Labs' $2.1 billion Series B represented around 87% of the early-July 2026 total. The median round was only about $10.5 million while the average was roughly $127 million. An average more than twelve times the median tells us that the supposed "average AI drug discovery financing" barely exists.

Chai Discovery then added another $400 million Series C at a $3.8 billion valuation shortly after that funding cutoff. Chai had been valued at $1.3 billion in its previous round only months earlier, so investors are willing to reprice perceived winners very aggressively.

There is also now a public-market route for selected companies. Generate Biomedicines raised $400 million in its IPO and is funding multiple clinical programs from the proceeds. That gives later-stage AI biotechs another source of capital beyond venture investors and pharma partnerships.

Period	Disclosed pure-play funding	Deals	What the headline hides
2024	~$1.54B	16	One $1B round dominated the year
2025	~$1.84B	28	More companies raised capital
Early 2026	~$2.42B	19	Isomorphic represented ~87%
Latest large private round	$400M	Chai Discovery	$3.8B valuation

If you want more recent data on this point, please see our latest AI in drug discovery market report.

Are AI drug discovery companies making real money yet?

AI drug discovery companies are generating real revenue now, although the strongest commercial proof still comes from a small group of platforms.

XtalPi provides the clearest growth example. Its drug discovery solutions revenue rose from RMB103.7 million in 2024 to RMB537.9 million in 2025. That works out to roughly 419% growth in one year. Total company revenue reached RMB802.6 million.

Schrödinger is growing at a much steadier pace. Its latest quarter produced $29.6 million of annual contract value, up 27% year over year, and trailing-four-quarter ACV reached $208 million. Quarterly software revenue actually fell 10% as the company shifted more customers toward hosted licensing, while total revenue still increased 8%.

Recursion gives us the opposite case. Its latest quarterly revenue fell from $19.2 million a year earlier to $7.7 million because of the timing of collaboration revenue. The company remains heavily dependent on research payments and milestones, so its quarterly numbers can move sharply even when the underlying partnerships continue.

That spread tells us more than a single market-growth percentage would. AI drug discovery already supports nine-figure commercial businesses, but software subscriptions, research payments and drug milestones behave very differently. Revenue growth across the category will remain lumpy for some time.

Company	Latest useful commercial measure	What we learn
XtalPi	RMB537.9M drug discovery revenue in 2025, +419%	Rapid paid adoption
Schrödinger	$208M trailing-four-quarter ACV	Established software demand
Recursion	$7.7M latest quarterly revenue vs. $19.2M a year earlier	Partnership revenue remains lumpy

Is the AI drug discovery clinical pipeline actually getting bigger?

The AI drug discovery clinical pipeline is now large enough that we can study a real cohort rather than a handful of famous molecules.

An industry-wide analysis presented through the American Society of Clinical Oncology identified 117 AI-enabled therapeutic assets from 63 companies that had entered Phase I, II or III trials. Sixty had completed Phase I and eight had completed Phase II by the end of the study's observation period.

The composition is revealing. Oncology accounted for 69 of the 117 assets, or 59%, while 96 were small molecules. Around 36% targeted novel biology. AI drug discovery has therefore reached clinical scale, but the pipeline is still heavily concentrated in areas where datasets, biomarkers and molecular targets are relatively abundant.

A previous analysis published in Drug Discovery Today had catalogued 75 molecules entering clinical testing through the end of 2023. The methodologies are different enough that we should avoid calculating a precise growth rate between the two studies, but the direction is obvious: the clinical pool available for evaluating AI-discovered drugs has become much larger.

We finally have enough programs to see failures as well as successes, which makes the market more measurable than it was a few years ago.

Are AI-designed drugs actually doing better in Phase I?

AI-discovered drugs have performed unusually well in Phase I so far, and this remains one of the strongest pieces of evidence that AI can improve molecular design.

The Drug Discovery Today study found that 21 of 24 AI-discovered molecules that had completed Phase I were successful. That is an observed success rate of 87.5%, which the authors summarized as roughly 80% to 90%.

Historical Phase I success rates for conventional drugs are usually much lower, often around 40% to 65% depending on the dataset and therapeutic area. The gap is large enough to take seriously even with the small AI sample.

There is a plausible explanation. Phase I asks whether a molecule has acceptable safety, pharmacokinetics and basic drug-like behavior. Those properties can increasingly be modeled before researchers commit to expensive synthesis and testing. AI can optimize potency, selectivity, metabolism and off-target behavior across huge chemical spaces faster than a human team exploring compounds one by one.

We should still resist turning 24 completed trials into a universal law. The early result is impressive, but the sample remains small and probably contains selection effects. For now, though, the Phase I record gives AI drug discovery a genuine clinical achievement rather than another promise.

If you want more recent data on this point, please see our latest AI in drug discovery market report.

Does AI drug discovery still hit the same wall in Phase II?

AI drug discovery loses most of its apparent advantage once trials have to prove that the medicine actually works in patients.

In the same Drug Discovery Today analysis, only four of the ten AI-discovered drugs that had completed Phase II were successful. That 40% rate sits close to historical industry estimates around the high-30% range.

The difference between Phase I and Phase II gets to the heart of the problem. AI can help scientists design a cleaner molecule against a chosen target. Human biology then decides whether that target was worth attacking in the first place.

BenevolentAI learned this painfully with BEN-2293. The drug met its safety and tolerability objectives in a Phase IIa atopic dermatitis study but failed to produce the efficacy the company wanted. BenevolentAI stopped investing in the program and subsequently restructured the business.

Other AI-enabled programs have run into similar problems. A Nature Biotechnology review published in late 2025 listed multiple assets that had failed or been discontinued after Phase II, including programs from BioAge, Recursion and Valo Health.

This is currently the biggest reason we cannot say that AI has already transformed pharmaceutical productivity. Molecule design looks better. Reliable prediction of human disease biology still does not.

Has AI drug discovery finally reached serious Phase III trials?

AI drug discovery has now reached Phase III, moving the market into a much more demanding clinical test.

Insilico Medicine's rentosertib is the clearest example. Insilico used its platform to identify TNIK as a target and then generated the molecule computationally. The drug has moved into a Phase III program for idiopathic pulmonary fibrosis following Phase IIa results published in Nature Medicine.

The earlier trial included 71 patients. At the highest once-daily dose, average forced vital capacity improved by 98.4 mL after 12 weeks, while the placebo group declined by 20.3 mL. The trial was small, so the Phase III program now has to show whether that result holds in a much larger population.

Generate Biomedicines is also running two global Phase III studies of GB-0895 in severe asthma. Generate built its platform around machine learning and large-scale experimental protein design, although GB-0895 has a different AI provenance from rentosertib's target-to-molecule story.

Rentosertib is the cleaner test for the full AI discovery thesis because the target and molecule both came through Insilico's platform. A successful pivotal trial would give the industry something it has never had before: strong late-stage evidence behind a drug whose origin can be traced directly through an AI-native discovery process.

For now, no AI-originated drug has created the kind of regulatory and commercial track record that would settle the argument. Phase III is where we may finally get it.

If you want more recent data on this point, please see our latest AI in drug discovery market report.

Is AI actually making drug discovery faster?

AI is already making parts of drug discovery much faster, especially the path from an initial biological idea to a nominated drug candidate.

Insilico reported that rentosertib went from project launch to preclinical candidate nomination in about 18 months. The program then completed early human testing within roughly 30 months from the beginning of target discovery. For a novel target paired with a newly designed molecule, that is unusually fast.

Recursion reports a similar change in medicinal chemistry. Its platform has produced advanced candidates after synthesizing roughly 330 compounds per program over about 17 months. The company compares that with an industry average above 2,500 synthesized compounds and 42 months.

Those Recursion benchmarks come from Recursion itself, so we should treat the exact comparison as company-reported rather than independent proof. Even so, reducing physical synthesis from thousands of compounds to hundreds would change the economics of early discovery if the result holds across many programs.

Where we still lack proof is the full journey to an approved drug. Faster candidate design does little to shorten a three-year Phase III trial or remove years of clinical follow-up. AI can already compress the front end of the process. We still do not know how many years it can ultimately remove from the whole drug-development cycle.

Are big pharma companies starting to build AI drug discovery themselves?

Big pharma companies are increasingly building their own AI drug discovery capabilities, which raises the bar for independent AI startups.

Lilly's collaboration with Nvidia shows how far internalization can go. The companies are creating a laboratory where AI models, automated experiments, proprietary Lilly data and scientists operate in one continuous loop. Lilly wants the system to generate experimental data that improves its models, which then choose what the lab should test next.

Novo Nordisk is now pursuing a similar direction with AWS. Its new AI partnership includes a co-innovation hub where Novo scientists work alongside AWS engineers and AI specialists, with the explicit goal of shortening the path from therapeutic target to first human dose.

This changes what an AI drug discovery startup needs to sell. A pharmaceutical company with its own foundation models, compute and internal AI team has less reason to pay a vendor for a generic molecular-prediction model.

Specialized outside companies can still win when they bring something the pharma company lacks: proprietary experimental data, better molecular-design models, an automated lab, unusual chemistry, or drugs that already exist as licensable assets.

Big Pharma bringing AI inside its walls therefore makes the technology more important while making the standalone software opportunity tougher.

Are failures and consolidation shrinking the AI drug discovery market?

Failures and consolidation are making AI drug discovery harsher for individual companies, while the overall market keeps getting bigger.

Recursion's combination with Exscientia showed the pressure clearly. The merger created a larger company with integrated biology, chemistry, data and clinical capabilities, but Recursion later cut roughly 20% of its workforce and narrowed its pipeline. Several clinical programs were discontinued or pushed toward partnering after the company reviewed where it wanted to spend its cash.

BenevolentAI went through an even sharper retrenchment after its clinical setback. The company cut staff, reduced laboratory operations and eventually dropped plans to build part of its business around standalone AI software products.

At the same time, valuable AI talent and technology are being absorbed by much larger companies. Anthropic acquired Coefficient Bio in a transaction reported at roughly $400 million, bringing a small team of computational drug discovery specialists into its life-sciences operation.

These developments suggest that simply owning an AI model is becoming less valuable. The companies attracting the most capital now tend to combine models with proprietary biological data, automated experimentation, chemistry or owned therapeutic programs.

We are watching the market concentrate around companies that can connect prediction to physical experiments and eventually to drugs. The weaker platforms may disappear even while spending on AI drug discovery continues to rise.

If you want more recent data on this point, please see our latest AI in drug discovery market report.

What still stops AI drug discovery from working reliably?

Human biology is still the main bottleneck for AI drug discovery, and better models alone have not removed it.

The Phase II results make that clear. AI has shown that it can produce molecules with strong early safety and pharmacology, but the advantage largely disappears when trials ask whether changing a biological target improves a real disease.

Companies are responding by investing heavily in proprietary experimental data. Recursion has built tens of petabytes of biological and molecular data and uses automated experiments to validate model predictions. Its latest collaboration progress with Genentech came after predicted neuroscience targets went through pathway, functional and disease validation before Genentech selected one for further discovery work.

The industry is moving toward the same basic architecture: predict something, test it physically, feed the result back into the model, then repeat. Chai is pairing foundation models with experimental validation. Iambic operates automated wet labs. Lilly and Nvidia are explicitly designing their joint facility around a continuous learning loop.

That direction makes sense because public biological data eventually become available to everyone. A model trained on the same literature and databases as every competitor is hard to defend. A company that continuously generates its own experimental evidence can build a dataset competitors cannot simply download.

So the current race is becoming less about who has the flashiest model and more about who can close the loop between computation and biology.

So, is the AI in drug discovery market growing now?

Yes. The AI in drug discovery market is clearly growing now, although its commercial growth is running ahead of the clinical proof needed to know how transformative the technology will ultimately be.

We see growth in several independent datasets. DealForma recorded AI/ML drug discovery partnership count rising from 84 to 114 in one year. Our narrower funding tracker found disclosed pure-play funding rising from about $1.54 billion in 2024 to $1.84 billion in 2025, with 2026 already above both before the year was half over. The ASCO industry analysis identified 117 AI-enabled clinical assets across 63 companies. Paid commercial activity is also becoming visible through companies such as XtalPi and Schrödinger.

That combination gives us much more confidence than any projected CAGR. More companies are raising money, pharma companies are spending more, AI tools are generating real revenue, and a growing number of AI-enabled drugs are reaching patients in clinical trials.

The uncomfortable part is clinical efficacy. The published Phase I record looks excellent, while Phase II success remains close to the conventional industry baseline. AI appears much better proven at designing viable molecules than at figuring out which biological ideas will become successful medicines.

For investors and entrepreneurs, that distinction is important. We are already looking at a real and expanding market. We are not yet looking at a technology that has conclusively rewritten the probability of drug-development success.

The market-growth question can now be answered with a clear yes. Whether AI drug discovery can turn that growth into meaningfully more approved drugs is the question that still has to be settled.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.