The Future of Drug Discovery Is More Shots on Goal

"Discovering new products outright and bringing them to market using AI is not yet possible."Johnson & Johnson says it has cut the time to generate new drug leads in half using artificial intelligence according to a Reuters report.

The New Jersey pharmaceutical company is using the technology to screen what CIO Jim Swanson calls the "potential universe" of chemical compounds and biologics, accelerating the optimization phase of early research.

Jim Swanson, J&J CIOBut J&J is not shortening drug discovery. It is increasing the number of attempts.

Drug development remains a decade-long process with staggering attrition. Only about 12% of drugs entering clinical trials are ultimately approved by the FDA. From preclinical research through FDA approval, the median cost per drug has reached $985 million, with capitalized costs often exceeding $1.8 billion. The clinical phase alone consumes 95 months on average and accounts for 69% of overall R&D spending.

These numbers define the problem. A company cannot accelerate biological reality or speed the requirement to test compounds in human subjects or satisfy regulatory scrutiny. What it can do is run more experiments in parallel, test more candidate molecules, and move more attempts through the fixed constraints of the development pipeline faster.

This mathematical reality explains why throughput matters. If a company's odds of approval are roughly 12%, the probability of success improves not through breakthrough discoveries but through volume. More candidates screened means more opportunities for one molecule to survive the gauntlet of Phase I, II, and III trials.

This is what J&J's AI strategy represents: a shift from searching for the perfect compound to maximizing the number of compounds tested.

What AI Actually Does

The company is reducing the time required to screen candidates and optimize their properties. Lead optimization, the stage where researchers refine promising chemical compounds before moving to animal testing, had taken months. Halving that timeline creates room for more parallel experiments.

Swanson emphasized the point explicitly: "Discovering new products outright and bringing them to market using AI is not yet possible." What J&J is doing is optimization. The distinction matters. The company is not replacing the biological unpredictability of drug discovery. It is removing friction from the processes that surround it.

The industry is moving in the same direction. Eli Lilly announced a joint AI innovation lab with NVIDIA in January 2026, with investment up to $1 billion aimed at transforming drug development into an automated system. The partnership aims to combine Lilly's biological expertise with NVIDIA's computational power to build what the companies call a "drug development factory."

Roche recently expanded its partnership with NVIDIA to deploy over 3,500 Blackwell GPUs across on-premises and cloud infrastructure, positioning the pharmaceutical giant to process research at scale.

Where the Time Gets Saved

Source: Johnson and JohnsonThe efficiency gains are not evenly distributed across the drug pipeline. They cluster in the operational layers surrounding discovery and clinical work: the administrative bottlenecks that slow compound movement through development.

Pfizer's predictive analytics team outlined a strategy centered on continuous visibility, monitoring enrollment data and site performance dynamically to optimize trial activation. Rather than waiting for quarterly reports, companies can now adjust strategies in real time.

Documentation automation shows the practical payoff. A major pharmaceutical company using generative AI for clinical trial documentation reportedly reduced its first-draft report timeline from three weeks to three days, cutting manual touch time from 200 hours to 100 hours. Clinical study report drafting can be accelerated by 30% to 70% depending on trial complexity.

AI-powered patient recruitment identifies eligible candidates in minutes rather than hours. This matters because 80% of clinical trials miss enrollment timelines. Clinical trial enrollment has been accelerated by up to 85% in some implementations, compressing timelines that would otherwise stretch for months.

These gains compress the pipeline without altering the underlying science. They reveal what AI is actually solving, that is, workflow friction. The operational efficiency around that science has improved.

What Remains Unchanged

The bottlenecks persist where biology matters most. Phase II clinical trials have a 28% success rate; Phase III has 55%. These numbers have not shifted because they reflect failures of science, not process. Compounds that looked promising fail to deliver efficacy or safety in larger populations.

The average drug development timeline is still over a decade, constrained by the regulatory requirement to test safety across large patient populations. No amount of computational efficiency can compress the time required to observe whether a compound is safe for human use. No AI system can substitute for the methodological rigor of a Phase III trial.

But within those immovable constraints, the operational machinery has accelerated. Fewer molecules are lost to administrative delays. More candidates advance because the pipeline moves faster. Some will still fail in clinical trials. But more will get the chance to try. For J&J and the rest of the industry, that is where AI is making its mark.