AI Drug Discovery: How DeepMind, Insilico and Recursion Are Rewriting Pharma Economics
How AI Is Transforming the Future of Healthcare:
Repricing the Value of AI-Driven Drug Discovery in Pharma & Biotech
A structured deep dive into how AI is shortening clinical timelines, cutting costs, and reducing failure rates in drug development — and what that means for investors.
📘 Part 1. Why AI Is Transforming the Pharma & Biotech Industry
— A structural shift where shorter clinical timelines, lower costs, and reduced failure rates happen at the same time
Drug development is often described as “closer to a probabilistic game than a pure scientific process.”
The industry is defined by extreme uncertainty and very high failure costs.
According to the Tufts Center for the Study of Drug Development (2020), bringing a single new drug to market
takes on average 10–15 years and costs the equivalent of roughly 3 to 4 trillion KRW (several billion USD).
Yet, among the drug candidates that reach clinical trials,
fewer than 9% ultimately make it to regulatory approval.
(Source: Nature Reviews Drug Discovery, 2019)
In an industry where failure rates are this high, AI is not just a convenience technology.
It is increasingly viewed as a tool that fundamentally rewrites how drugs are discovered and developed.
🔹 1) What Global Research Institutions Are Saying About the Impact of AI
(Based on peer-reviewed papers and official research reports)
✔ ① Over 90% of clinical failures stem from “poor candidate prediction”
Nature Reviews Drug Discovery (2020) analyzed the reasons behind drug development failures
and concluded that the largest single factor is “inadequate target and candidate prediction in early-stage research.”
In other words, many programs are doomed from the start because the wrong candidates are selected,
and that only becomes apparent after companies spend hundreds of millions of dollars
on clinical trials that eventually fail.
This is exactly where AI’s contribution becomes explosive.
AI models can simultaneously learn from billions of molecular structures, protein–ligand interactions, and toxicity patterns,
and narrow down promising candidates with a level of precision that manual approaches simply cannot match.
A process that would take researchers years of iterative experiments
can be compressed into months or even weeks.
✔ ② Lead discovery timelines: From “4–7 years → under 1 year”
If we synthesize disclosures and case studies from leading AI drug discovery companies —
such as DeepMind (AlphaFold2), Insilico Medicine, and Exscientia — we get the following picture:
Traditional approach: 4–7 years for target identification + lead discovery
AI-driven approach: 6 months to 1 year
Insilico Medicine famously identified a fibrosis drug candidate
in just 45 days (published in 2021),
which has been cited as roughly 40–50 times faster than conventional discovery timelines.
This speed difference matters because in drug development,
“time is literally money.”
Shorter discovery cycles reshape not only project-level economics,
but also company-wide portfolio strategy and industry structure.
✔ ③ Overall development costs can be reduced by 20–30%
Boston Consulting Group (BCG) and Deloitte both estimate that
AI has the potential to cut overall drug development costs by 20–30%.
The cost reductions mainly come from three levers:
1. Lower failure rates in the earliest stages
2. Fewer repetitive in vivo and in vitro experiments
3. Faster optimization before entering clinical trials
Preclinical work alone can account for more than 40% of total development expenses,
and this is exactly where AI has demonstrated the clearest improvements so far.
🔹 2) Why AI Has Become “Mandatory,” Not Optional, in This Industry
AI has become a must-have technology in pharma and biotech
not just because it is faster or cheaper,
but because it represents a fundamentally different way of doing science.
✔ ① The legacy model was essentially “brute-force exploration”
Traditional drug discovery typically involves the following steps:
1. Selecting a target protein
2. Running large-scale lab-based high-throughput screens
3. Conducting toxicity studies
4. Performing animal studies
5. Moving from preclinical into clinical development
The problem is that all of these steps rely on massive trial and error.
Teams have to test tens of thousands or even hundreds of thousands of compounds in the lab,
leading to huge time and cost burdens that weigh on the entire industry.
✔ ② AI eliminates weak candidates in silico before they reach the lab
AI models can filter out molecules with low drug-likeness or high toxicity
before a single well plate is prepared.
For example, AI can be used for:
Protein–ligand binding prediction
Toxicity prediction and safety profiling
Molecular docking and binding mode modeling
Cell image–based phenotypic screening and pattern recognition
Because of this, wet labs only need to validate the “high probability” candidates,
creating a much more compressed discovery pipeline.
✔ ③ Reducing early failures dramatically lowers total cost of development
When a single preclinical candidate drops out,
the loss is not just the molecule itself.
What disappears with it are:
All the internal R&D labor costs
Lab equipment and consumables
Fees paid to contract research organizations (CROs)
Costs associated with revising clinical plans and timelines
All of these expenses evaporate at once when a candidate fails.
AI attacks this waste at its root.
In short,
becomes “the ability to save billions of dollars” across a portfolio.
🔹 3) A New Economics of Drug Development Powered by AI
Historically, drug development has been a game of
“how long and how much can you afford to burn?”
In the age of AI, that logic is being rewritten.
✔ ① An industry structure where speed is everything
In many therapeutic categories, the first company to reach approval
can effectively dominate the market for years.
Cancer therapies
Treatments for rare diseases
Immuno-oncology drugs
Antiviral therapies
These areas tend to show especially strong “first-mover advantage.”
AI-first companies that can move rapidly into the clinic
translate their R&D efficiency directly into enterprise value.
✔ ② Lower failure rates → explosive upside in company valuations
Deloitte estimates that even a 1 percentage point increase
in Phase I success probability can add billions of dollars
to a company’s long-term valuation.
When AI raises overall success probabilities from 9% toward the 15–20% range,
this is not just a technical improvement.
It fundamentally shifts how investors should think about risk, cash flows, and capital allocation in pharma.
✔ ③ “AI + Pharma” is now a survival strategy, not a side project
Global giants such as Samsung Biologics, SK Biopharm, Pfizer,
Novartis, Sanofi, and Roche
are either building in-house AI research teams
or signing large, multi-year deals with specialized AI startups.
As of 2024, major pharma companies are gradually increasing
the share of AI-related programs in their total R&D budgets,
with some estimates putting this at roughly 12–20% and rising.
📌 Key Takeaway — “AI is a technology that completely rewrites the economics of drug development”
AI is not just another automation tool.
It directly tackles the four structural pain points of drug development:
Long development timelines
High capital intensity
Low historical success rates
Repeated and expensive failures
Because of this, AI adoption is increasingly seen
as a core driver of enterprise value in global pharma.
From 2025 onward, many expect AI drug discovery companies
to be repriced as investors, industry, and academia
converge around this new reality.
📘 Part 2. Global Leaders in AI-Driven Drug Discovery
— Focusing only on “validated players” that actually move the industry
There are hundreds of companies that describe themselves as AI drug discovery players,
but only a small number have achieved
three concrete milestones:
1) Clinical-stage pipelines
2) Signed partnerships with global pharma companies
3) Demonstrated progress toward commercialization
The companies below meet all three criteria,
based on peer-reviewed publications, industry coverage, and official company reports.
(References: Nature Biotechnology, MIT Technology Review, company annual reports)
⭐ 1) DeepMind / Isomorphic Labs (Alphabet)
— The company that created a paradigm shift in protein structure prediction
DeepMind’s launch of AlphaFold2 in 2021
is widely considered a historic moment in drug discovery.
AlphaFold2 achieved over 92% accuracy in protein structure prediction
and was hailed by Nature (2021) as having “solved a 50-year grand challenge in biology.”
This was not just an academic milestone.
It fundamentally changed how pharma companies around the world
think about target identification and protein modeling.
Alphabet later spun out Isomorphic Labs,
which between 2022 and 2025 signed partnerships with
GSK, Novartis, and other major pharma companies,
turning the AlphaFold ecosystem into a commercial drug design platform.
Why this matters:
1. Protein structure prediction is the starting point of nearly every rational drug design effort →
one of the highest-impact use cases for AI in pharma
2. Backed by Alphabet’s balance sheet and access to world-class AI talent
3. Since 2024, these models have been progressively integrated
into real joint research programs at large pharma companies,
marking the transition from “research project” to “commercial infrastructure”
In short, DeepMind and Isomorphic Labs have effectively set
the baseline standard for foundational AI models in drug discovery.
⭐ 2) Insilico Medicine
— The first company to bring an “end-to-end AI-designed drug” into the clinic
Insilico Medicine is widely regarded as one of the most concrete success stories
in AI drug discovery so far.
Its flagship program, INS018_055,
is the first clinical candidate where the entire workflow —
from target discovery and molecular design
through preclinical optimization and candidate selection —
was driven by an AI platform.
The molecule successfully completed Phase I trials in 2023
and, as of 2024–2025, is in Phase II.
Technical significance:
Whereas many pharma companies still use AI as a “supportive tool”
for certain stages, Insilico rebuilt the whole development chain
around AI as its core engine.
Industrial significance:
Insilico has demonstrated two key breakthroughs:
1. Lead discovery timelines compressed from 4–5 years to 45 days
2. A concrete example of an AI-generated molecule
successfully progressing into the clinic
Backed by investors such as Warburg Pincus and Fosun Pharma,
Insilico is often cited as one of the fastest-growing and most closely watched
AI drug discovery companies in the world.
⭐ 3) Recursion Pharmaceuticals (NASDAQ: RXRX)
— The only AI drug discovery company with a direct strategic investment from NVIDIA
Recursion stands out from its peers by combining
high-throughput lab automation, image-based AI models, and supercomputing at scale.
In 2023, NVIDIA announced a 50 million USD strategic investment in Recursion,
and Recursion OS was integrated with the NVIDIA DGX platform,
creating one of the largest biological data processing environments in the industry.
Technical foundation:
Recursion has built a dataset of more than 210 million biological images,
training AI models to understand cell-level responses
via phenotypic screening.
This approach differs from traditional structure-based methods
focused on protein–ligand binding.
Instead, Recursion’s models learn from actual biological phenotypes,
making them especially powerful for drug repurposing
and uncovering unexpected mechanism-of-action relationships.
Industrial achievements:
Collaboration with Bayer valued at over 1 billion USD
Multi-year deals with Roche/Genentech
Several programs progressing into clinical stages
Following the NVIDIA partnership,
Recursion’s computing capacity and model training speed
have significantly improved,
further strengthening its position as a platform company in AI drug discovery.
⭐ 4) Exscientia (NASDAQ: EXAI)
— A global leader in “drug optimization algorithms”
Exscientia, based in Oxford in the UK,
is widely seen as one of the strongest players in
the critical but often underappreciated stage of drug optimization.
Finding an initial hit or lead is only the beginning.
The much harder challenge is optimization, which requires
balancing toxicity, absorption, metabolism, solubility,
and many other pharmacokinetic and pharmacodynamic properties at once.
Exscientia has shown that AI systems can perform this complex balancing act
with a level of precision comparable to seasoned medicinal chemists
with decades of experience.
Key milestones:
1. Three AI-designed drug candidates in clinical stages
2. A collaboration with Sanofi worth up to 5.2 billion USD in milestone value
3. Frequent recognition at the JP Morgan Healthcare Conference
as a benchmark company in AI-designed small molecules
Exscientia’s algorithms can rapidly navigate vast chemical spaces
to find optimal combinations,
positioning the company as a key determinant of drug quality
within the AI drug development ecosystem.
⭐ 5) Related Players in Korea (Reference Only)
— Companies expanding AI platforms through manufacturing and data infrastructure
There are still relatively few pure-play AI drug discovery companies in Korea.
However, several listed companies have publicly disclosed
meaningful AI-related initiatives in manufacturing and healthcare data.
Samsung Biologics
According to its annual reports,
Samsung Biologics has been expanding its use of AI and machine learning
across manufacturing automation and quality assurance (QA) systems.
While the focus is more on large-scale production and error reduction
than on upstream drug design,
it is considered one of the largest AI adopters in Korea’s biomanufacturing sector.
SK Biopharm
SK Biopharm has officially announced the introduction of AI platforms
for early-stage candidate discovery,
particularly in central nervous system (CNS) disorders.
CNS is an area where AI-based target identification
is expected to be especially impactful.
Kakao Healthcare
Kakao Healthcare is building a medical data platform,
aggregating data from hospitals, check-up centers, and other institutions,
and working on AI-based prescription support and medical information interfaces.
It is not a drug development company per se,
but it operates one of the most significant healthcare data infrastructures in Korea.
📌 Summary of Part 2
The AI drug discovery space is crowded,
but in practice, a small group of companies are leading the real-world adoption.
DeepMind / Isomorphic Labs: Standard-setters for foundational models
Insilico Medicine: First to bring an end-to-end AI-designed drug into the clinic
Recursion: Combining lab automation with NVIDIA-powered supercomputing
Exscientia: Best-in-class drug optimization algorithms
These companies have signed multi-billion-dollar, multi-year deals
with global pharma partners,
and are gradually turning AI drug discovery
from hype into real, recurring business.
📘 Part 3. How Investors Should Revalue AI Drug Discovery Companies
— Beyond “technology stories” to a new framework for valuing pharma and AI platforms
The recent rerating of AI drug discovery companies in capital markets
is not just because “they use AI.”
Instead, investors are responding to structural changes in
economic value creation, pipeline quality, and deal structures with big pharma.
From an investor’s standpoint, three shifts are particularly important.
🔹 1) Higher clinical success rates → structural uplift in enterprise value
The single biggest risk in drug development is low success probability.
Historically, fewer than 10% of drugs entering Phase I
have made it all the way to approval.
According to Deloitte’s 2023 report,
the average probability of transitioning from Phase I to Phase II
was around 52%.
For pipelines where AI was involved in target discovery and molecular design,
this figure rose to 65% or higher.
This is not a minor improvement.
It fundamentally changes how cash flows and risk-adjusted valuations
should be modeled.
Why does a small uplift in success rate matter so much?
The commercial value of a single successful drug
can range from several billion to tens of billions of dollars
over its lifecycle.
So even a one-percentage-point increase in success probability
can translate into billions of dollars of added expected value.
For companies that adopt AI across their pipelines, several things happen at once:
Faster candidate selection → more shots on goal
Lower preclinical failure rates → better cost structure
Higher clinical entry and progression rates → higher average pipeline quality
In this sense, AI is less about “cool technology”
and more about directly improving the future revenue potential
of a company’s entire portfolio.
🔹 2) Big pharma collaboration & M&A as direct catalysts for AI companies
From 2022 through 2025,
large pharmaceutical companies have signed
an unprecedented wave of partnerships
with AI-focused drug discovery firms.
Pfizer: Multiple AI-driven design platforms integrated into R&D
Roche: Strategic partnership with Recursion
Novartis: AI-based target discovery and clinical data platforms
Sanofi: Collaboration with Exscientia worth up to 5.2 billion USD
Why is big pharma so eager to work with AI companies?
1) R&D cost reduction
Global pharma collectively invests over 100 trillion KRW (tens of billions USD)
in R&D each year.
Even modest reductions in failure rates can generate massive savings over time.
2) The race to secure future pipelines
Competition to develop the next generation of blockbuster drugs is intense.
AI companies that can deliver differentiated candidates quickly
offer big pharma both speed and diversity in their pipelines.
3) Rising probability of M&A
As AI drug discovery platforms move from “promise” to “revenue,”
the strategic rationale for M&A grows stronger.
Acquiring a proven AI platform can give big pharma
a long-term edge in pipeline generation.
Evaluate Pharma and GlobalData both indicate
that the share of AI-related investments in pharma R&D budgets
has risen sharply in 2024–2025.
For AI companies, this shift is crucial because it creates
new business models that include:
Upfront and milestone-based revenue from collaborations
Long-term royalty streams from successful partnered drugs
Platform licensing and data-access fees
🔹 3) Objective criteria for identifying promising AI drug discovery companies
Because AI drug discovery is technically complex,
the gap between “marketing slideware” and real capability
can be extremely wide.
From an investor’s perspective, four objective filters are especially important.
✔ ① Accuracy and validation of the underlying models
Investors should look for companies whose models have been
rigorously evaluated by the scientific community:
Peer-reviewed publications in journals such as Nature, Science, Cell
Performance on public benchmarks like CASP for protein structure prediction
This is one reason DeepMind and Isomorphic Labs
are treated differently from less transparent players.
✔ ② Presence of clinical-stage programs
Many AI-first companies remain stuck in early-stage research.
Actual clinical programs are a crucial credibility check.
Insilico, Recursion, and Exscientia all have
multiple AI-influenced drug candidates in Phase I or Phase II trials,
which materially strengthens their investment cases.
✔ ③ Long-term, large-scale collaborations with big pharma
When a global pharma company commits to a long-dated partnership,
it signals that the technology has passed multiple layers of due diligence.
Key aspects to examine include:
Overall headline deal size
Whether milestones and royalties are included
Duration and strategic scope of the agreement
These factors are central to assessing both growth potential
and downside protection for AI platform companies.
✔ ④ Access to GPU and compute infrastructure (e.g., NVIDIA partnerships)
Training and deploying biological foundation models
requires significant compute capacity.
One reason Recursion received heightened attention
after NVIDIA’s investment
is that it gained access to large-scale GPU infrastructure,
which directly influences model iteration speed and scalability.
In AI drug discovery, compute is not a side consideration —
it is tightly linked to development speed and, ultimately,
to revenue generation potential.
📘 Part 4. Conclusion — “AI Is Rewriting the Economics of the Drug Industry”
The value of AI in drug development goes far beyond “making things faster.”
In an industry where it takes more than a decade and billions of dollars
to bring a single therapy to market,
AI has emerged as the only technology that can simultaneously
reduce cost and failure rates at scale.
The changes now underway across the pharma industry
are not just incremental digitization or basic automation.
We are witnessing a redefinition of the underlying economics of drug development.
🔹 Six Core Dimensions of Change Driven by AI
1. Shorter lab and discovery timelines
2. Higher precision in lead identification and validation
3. Improved clinical entry and progression rates
4. Strategic reallocation of big pharma R&D budgets toward AI
5. Rapid advancement of GPU-powered bioinformatics
6. Real-world clinical programs featuring AI-designed molecules
Because all six of these trends are moving at once,
the period from roughly 2025 to 2030
is likely to be a major repricing window
for AI-first drug discovery companies.
As more clinical data accumulates between 2026 and 2028,
we will increasingly see AI move from “research promise”
to “revenue-generating reality” in drug pipelines.
📚 Key References
Tufts Center for the Study of Drug Development, 2020
Nature Reviews Drug Discovery, 2019–2021
Nature Biotechnology, AI Drug Discovery Special Issue
MIT Technology Review, 2022–2024
Deloitte Pharmaceutical Innovation Report, 2023
Evaluate Pharma World Preview, 2024
GlobalData Pharma R&D Tracker, 2023–2024
Official annual reports and IR materials from the companies mentioned
Disclaimer: This article is based on publicly available research, industry reports, and company disclosures. It is intended for informational and educational purposes only and does not constitute investment advice or a recommendation to buy or sell any security.
댓글
댓글 쓰기