Biotech and pharma are rapidly adopting AI. NVIDIA’s State of AI in Healthcare and Life Sciences report found that 74% of pharma and biotech organizations are actively using AI. About half (47%) report ROI from AI for drug discovery and development, and 48% are using AI agents for drug discovery and biomarker identification.

The Benchling 2026 Biotech AI Report also reports AI adoption. The report found that 46% of biotech and pharma organizations are using AI apps and scientific models in R&D regularly, and a majority of them (81%) use it for scientific use cases.

AI Biotech: AI-Powered Drug Discovery and Development

Benchling’s report says that about half of those adopting AI report faster time-to-target, 56% expect meaningful cost reductions within two years, and 42% see an uplift in accuracy and hit rates.

“90% of clinical trial failures happen in Phase 2 or 3 clinical trials, not in discovery,” Iker Huerga, CEO at Pathos , says.

“That is where AI has to work if it is going to change medicine, and it is where almost no one is actually building.”

Pathos is an AI and technology company that develops drugs. Its core is Pathos Foundry, a home-grown platform running thousands of AI agents that reason continuously over more than 200 petabytes of multimodal patient data, Pathos’ real-time clinical trial systems, Pathos’ wet-lab experiments, and published literature.

“It does three things a traditional biotech does not,” says Huerga. “It sources drug candidates from data other companies never look at, designs and steers clinical trials in real time, and makes disciplined stop decisions early, before hundreds of millions of dollars are burned on assets that were never going to work,” says Huerga.

In the last 12 months, Foundry has made four platform-informed decisions across four different drugs, each one changed the trajectory of the program. “That is what happens when AI stops being a slide in a deck and starts making the decisions that matter,” says Huerga.

Those decisions are visible today in a clinical pipeline of four oncology assets, Pocenbrodib in prostate cancer and multiple myeloma; DO-2 in non-small-cell lung cancer; P-100 in ER+/HER2- breast cancer; and JSKN-016, a first-in-class TROP2×HER3 bispecific antibody-drug conjugate for triple-negative breast and lung cancer.

AI Biotech Virtual Cells, Models for Human Biology

When drugs fail in clinical trials, the most common reasons include lack of efficacy and unexpected toxicity, Dr. Le Song co-founder and CTO of GenBio says. “Both problems largely come from the same gap: no existing tool can predict how a molecule will behave inside the full complexity of a living cell,” says Dr. Song.

At GenBio, an in-house team is building the first world model of the human cell with AIDO Cell. Rather than modeling single layers like protein structures in isolation, the model is a stateful system that simulates how a living human cell responds across all biological scales to a disease signal or therapeutic intervention.

“I believe world models of the human cell are one of the most impactful technologies in biotech today,” says Dr. Song. “Rather than answering one narrow question, a world model learns how the whole system works, so it can simulate how everything responds when one thing changes, whether that’s a disease taking hold or a drug being introduced.”

“Ultimately, this could mean getting safer, more effective treatments to patients sooner and making it viable to take on diseases that have been too costly or too complex to tackle until now,” said Dr. Song.

AI Biotech Precision Medicine, Antibodies, Gene Therapies, Vaccines, RNA-Based Therapies

A recent Deloitte survey found that executives expect large molecules (64%), cell, gene, and RNA-based therapies (62%), and antibody-drug conjugates (54%) to power revenue growth over the next two to three years.

“The technology I’d point to sits underneath the AI headlines we see today: the ability to write DNA on demand,” Eric Esser CEO of Telesis Bio , says.

Every biologic, including antibodies, mRNA vaccines, and gene therapies, starts with a DNA sequence. While AI has made hypothesis generation dramatically faster, turning digital designs into physical molecules for validation involves outside, time-consuming DNA service providers, Esser explains.

Telesis Bio’s Gibson SOLA platform solves this problem by enabling labs to make gene-length DNA in-house, fully on-demand, in just a few hours using standard lab automation equipment.

“Gibson SOLA software orchestrates the entire process from sequence design to hardware interface and integrates with researchers’ AI stack, unlocking the potential for fully automated and even autonomous discovery,” Esser explains.

“Antibodies are already one of the most important classes of medicine we have,” says Dov Gertz, CEO and co-founder of Converge Bio. “What’s holding the field back is how slowly we can make good ones.”

ConvergeAB, Converge Bio’s antibody design platform, allows scientists to start from a parent antibody and get back candidates that are more human, more stable and easier to manufacture, and that still bind the target.

“Patients benefit in two ways,” says Gertz. The first is speed: fewer design and test loops means candidates reach the clinic sooner. The second is better versions of drugs that already work. “We call these biobetters,” says Gertz. “An antibody that is less likely to trigger an immune response, or is more stable in the vial, can be safer for patients and easier to manufacture at scale,” Gertz explains.

AI Biotech and Pharma Supply Chain Opportunities

While many biotech and pharma companies are modernizing in-house, others are merging, acquiring, or partnering with AI firms. EY’s strategic U.S. mergers and acquisitions (M&A) August report ranks life science second, with deal values increasing by 123% and volume by 109%.

As biotech companies embrace AI , challenges and gaps ranging from legacy tech, data demands, budgets and security emerge. These challenges are opportunities in the supply chain.

Huerga from Pathos says that the real supply chain constraint in this era is not molecules. “It is patients matched to trials, decisions made inside those trials, and the wet-lab and manufacturing infrastructure that turns a computational hypothesis into a physical drug fast enough to matter,” says Huerga.

“That work has not meaningfully accelerated in twenty years, and it is where the industry needs new partners,” Huerga says.

Pathos sees three specific gaps. First, molecular diagnostics at real scale. “This is the ability to enrich a trial population by actual biology rather than by broad clinical criteria,” says Huerga.

Second, real-time clinical data capture that AI agents can reason over the moment it enters an electronic data capture system, not months later, said Huerga. Third, chemistry, manufacturing and controls (CMC), GMP batches, salt-form screening, controlled-release formulation, etc., that can turn around in weeks and match the speed of an AI-guided program instead of gating it.

Pathos runs Foundry directly against these workflows. Foundry’s agents talk to our clinical data systems, our wet-lab, and our manufacturing supply base in real time.

“That is how you compress a program timeline without cutting corners, and it is how you get more of the capital raised for a drug to actually reach patients,” says Huerga. “The partners who can operate at that speed are the ones who will define the next decade of oncology drug development,” Huerga adds.

AI Biotech Supply Chain Demands for Automated Labs, CROs

The biggest need is fast, standardized wet lab validation, said Gertz from Converge Bio. Design teams can now send out hundreds of candidates at a time and now need partners who can express, purify, and characterize them quickly with consistent assays for binding, stability and aggregation. “Automated labs and CROs built for high volume and short turnaround will be in demand for years,” says Gertz.

The second gap is data. “Our models are only as good as the data behind them, and much of the most useful data is never published, especially failed experiments,” says Gertz.

Labs and service providers that generate clean, consistent experimental data, and structure it so models can learn from it, are becoming part of the drug discovery supply chain.

“Third, manufacturing input has to come earlier,” Gertz adds. CDMOs that work with design teams from the start can help make sure a candidate is manufacturable from the first design, instead of finding out it isn’t a year later,” says Gertz.

“Companies that make the lab step as fast as the design step will set how quickly this technology reaches patients”, Gertz explains.

AI Biotech: Randomized controlled trials (RCTs) and FDA Regulations

Another challenge in biotech and pharma is FDA regulations. In the paper Integrated and holistic evidence generation 2.0, Raghav Dave of Sonata Software offers a concrete, technology-grounded framework to bridge regulatory gaps with new technologies. The framework is grounded in existing FDA programs, published research, and implementable technology standards.

The 21st Century Cures Act of 2016 directed the FDA to develop frameworks for integrating real-world evidence (RWE) into regulatory decision-making. Randomized controlled trials (RCTs), which regulatory frameworks are built around, have validated countless therapies and protected patients from ineffective or harmful treatments, the paper says, but gaps between what science knows and what the evidence record reflects exist.

“Technology is now uniquely positioned to close those gaps,” says Dave.

AI Biotech Manufacturing, Demand and Supply, and Distribution

“Lower discovery costs or scientific breakthroughs alone won’t bring more medicines to patients,” says Leonard Mazur, co-founder, CEO, and Chairman of the Board of Directors at Citius Pharmaceuticals . Much of a drug’s cost and risk comes later, in clinical development, manufacturing, and getting it to the patients who need it, Mazur explains.

“If AI floods the pipeline with promising candidates while manufacturing capacity, trial sites, and distribution stay the same, we’ve only moved the bottleneck,” says Mazur.

“The real opportunity is using AI across the whole chain.”

That means optimizing manufacturing processes, predicting supply shortages before they happen, matching patients to trials faster, and helping specialty products reach the right treatment centers, says Mazur, highlighting that supply chain partners can make the biggest difference for companies like Citius Oncology.

“We learned at Citius that getting a therapy to patients depends as much on manufacturing and distribution as on the science,” says Mazur. “The companies that will succeed in this era will treat AI as a tool for the entire journey from lab to patient, not just the first step.”

Citius Oncology’s LYMPHIR, which targets the cells that suppress the body’s response to cancer, is just one approved therapy being explored to do just that. “Although developed before the availability of AI, the potential of AI-aided drug discovery to augment existing therapies remains largely untapped,” says Mazur.

AI-aided discovery matures and will accelerate and be more precise, but the therapies with the greatest global impact will be the ones that can be reliably manufactured, paid for, and accessed by patients, says Mazur.