Artificial intelligence has genuinely revolutionized the early stage of drug discovery, helping researchers design promising molecules in months instead of years. But a critical gap is emerging: AI's success at finding molecules hasn't translated into success at proving those molecules actually work in patients. As the first meaningful batch of AI-designed drugs enters large-scale human trials in 2026, the industry is confronting an uncomfortable truth: the billions invested in AI drug discovery may have targeted the wrong bottleneck entirely.

Why Is AI So Good at Designing Molecules but Not at Proving They Work?

The answer lies in the fundamental difference between two types of problems. Designing a molecule is essentially a chemistry and physics puzzle with firm rules and enormous libraries of historical data. Machine learning was built for exactly this kind of problem, the same way it learned to recognize faces or predict the next word in a sentence. Companies like Absci can now design a working antibody from scratch by testing fewer than a hundred candidate designs, compared to the older approach that screened millions of physical compounds.

Insilico Medicine reported going from selecting a biological target to having a finished drug candidate ready for its first safety tests in about eighteen months for roughly $2.6 million, against the four to six years the same work traditionally took. By 2022, some 150 firms were applying AI to small-molecule design alone, and the money behind them has made AI drug discovery a multibillion-dollar industry. A single 2024 entrant, Xaira, launched with $1 billion in funding, and Alphabet's Isomorphic Labs raised $600 million in 2025.

But here's the problem: designing the molecule was never the expensive part. The roughly $2.6 billion it takes to bring a single drug to market is dominated not by discovery but by clinical trials and the failures within them. And there, AI has so far made almost no difference.

What Do the Early Clinical Results Actually Show?

A Boston Consulting Group analysis of about two dozen AI-discovered molecules that had reached clinical trials reveals a striking pattern. In Phase 1, the first time a drug is given to a small group of people to check that it is safe, AI-designed molecules succeed 80 to 90 percent of the time. This is far above the historical norm of about 50 percent.

But in Phase 2, the first real test in patients of whether the drug actually does anything, their success falls back to the industry's ordinary 40 percent or so. Phase 3, the large confirmatory trial that comes before approval, is where the first AI-designed drugs are only reaching now, so there is no track record there yet.

The net effect is positive but modest: the end-to-end odds that an AI-designed drug reaches the market roughly double, from something like 5 to 10 percent to 9 to 18 percent. And every point of that gain is banked in the cheap early stage. A Phase 1 study averages about $4 million, a Phase 2 study about $13 million, and a Phase 3 trial $20 million and often far more. AI is saving money at the front of a process whose costs pile up at the back.

Where Is the Real Bottleneck in Drug Development?

The reason for this divide reveals the most important and least appreciated fact about AI in medicine. Choosing the right biological target is a fundamentally different kind of problem than designing a molecule. To know whether blocking a particular protein will actually change the course of a disease requires understanding the tangled causal machinery of human biology, machinery that has been mapped only in fragments, on data that are sparse, noisy, and often missing.

A machine learning model can learn the rules of chemistry because those rules are well understood and documented. It cannot learn the rules of a disease we do not yet understand. Drugs fail in two ways: the molecule is wrong, or the biological bet behind it is wrong. AI has largely conquered the first while barely touching the second.

Michael A. Santoro, Professor of Management and Entrepreneurship at the Leavey School of Business at Santa Clara University, explained the stakes of this divide:

"The AI money has crowded into the tractable problem, molecule generation, where returns are competed down precisely because so many firms can now do it, and where success was never the binding constraint. The larger prize lies at the other end of the pipeline: using AI to attack the Phase 2 and Phase 3 problem directly."

Michael A. Santoro, Professor of Management and Entrepreneurship, Leavey School of Business, Santa Clara University

How Should Investors and Companies Rethink AI Drug Discovery?

The early clinical returns bear out this analysis. The brightest result so far is Insilico Medicine's rentosertib, an AI-discovered drug for idiopathic pulmonary fibrosis that posted positive mid-stage results in Nature Medicine in 2025. Against it sit the setbacks: Recursion, one of the field's most prominent companies, discontinued its lead AI-discovered program in 2025 after the efficacy signal failed to hold. The once-frothy field of standalone AI-drug firms has been thinned by mergers and shutdowns.

For investors, this should reframe the entire opportunity. A company that made even a dent in the efficacy failure rate would capture far more value than another molecule-design platform, because that failure rate is what the $2.6 billion is mostly paying for. The key areas where AI could make a real difference include:

• Target Validation: Using AI to determine which biological targets actually drive disease, rather than relying on incomplete biological understanding.
• Biomarker Discovery: Finding biomarkers that predict which patients will respond to a drug, reducing failure rates in clinical trials.
• Trial Design: Designing smarter and cheaper clinical trials that can more efficiently test whether a drug works.
• Efficacy Prediction: Developing AI systems that can predict whether a drug will be effective before expensive human trials begin.

Whether AI can reach into this harder problem is the open question on which the entire bet now turns. It may be that better data and new methods let AI begin to predict efficacy, in which case the value there dwarfs anything in discovery. It may be that the biology is a limit of knowledge no amount of computation can shortcut, in which case much of the discovery boom will have been misallocated capital chasing the wrong problem.

The year 2026 will provide the first real evidence. As AI-designed drugs move through Phase 3 trials, the industry will finally learn whether the billions invested in AI drug discovery have actually produced medicines that help patients, or whether the technology has simply made the easy part easier while leaving the hard part untouched.