AI Drug Discovery Has Drawn Heavy Investment, So Why Is the First Approval Still Missing?

As artificial intelligence is expected to reshape the pharmaceutical industry, its most enticing promise has always been speed: finding targets faster, designing molecules faster, and moving an idea toward patients faster. But what truly determines whether a drug can reach the market has never been early discovery alone. It is the long, expensive, and failure-prone process of human trials. That is also the sharp point raised in a recent Clinical Trial Vanguard article: the industry has invested about $7 billion in AI drug discovery, yet so far no AI-discovered or AI-designed drug has been approved for market.

The methodology behind that figure should still be treated with caution, because there are currently no external sources for the same event that can be cross-checked. But the structural issue the article points to is familiar. AI can screen possible molecules across vast chemical space, predict protein interactions, or help identify disease-related pathways. However, an elegant candidate molecule still has to prove that it can be absorbed by the human body, reach the right tissue, produce sufficient efficacy, and avoid unacceptable toxicity.

In recent years, some AI drug pipelines have indeed entered human trials, including candidate drugs for diseases such as idiopathic pulmonary fibrosis, which were once viewed as representative examples of generative AI being used to find targets and design small molecules. Such progress shows that AI is not merely a promotional slogan and can already move some early-stage R&D work into the clinical phase. But entering the clinic is not the same as reaching the finish line. From Phase 1 safety, to Phase 2 preliminary efficacy, to Phase 3 large-scale validation, every stage can stop because of insufficient efficacy, dosing issues, population selection, or safety signals.

The “wrong race” described in the article’s headline centers on the possibility that funding and attention may be overly concentrated on the discovery side. For pharmaceutical companies and investors, an algorithm’s identification of a new target or molecule can easily become a demonstrable result and is easier to package as a platform capability. By contrast, clinical trial design, patient recruitment, endpoint selection, real-world data quality, and regulatory communication sound less dazzling than generating molecules, yet they are often closer to where failure occurs.

Background Context

Drug development is not simply an engineering problem to begin with. Many diseases involve multilayered biological networks such as immunity, metabolism, inflammation, fibrosis, or neurodegeneration, and animal models and cell experiments may not accurately predict human responses. If AI learns only from incomplete, biased, or hard-to-integrate datasets, it may accelerate the production of candidate answers without necessarily improving the success rate of those answers in the clinic.

This does not mean AI has no place in new drug development. A more pragmatic use may be to apply AI to specific steps: improving trial inclusion criteria, identifying patient subgroups more likely to benefit, analyzing imaging and biomarkers, monitoring safety signals, or helping research centers identify suitable participants faster. These tasks may not produce the label of the “first AI drug,” but they could actually shorten trial timelines, reduce misallocation, and make clinical evidence clearer.

Therefore, the gap between $7 billion and zero approved drugs is less a final verdict on AI drug development than an industry recalibration. If AI only increases the number of candidate drugs without simultaneously improving validation, regulation, and clinical execution, what the pharmaceutical industry may get is a more crowded pipeline, not more genuinely usable treatment options.