Startups are using virtual simulations to forecast pharmaceutical clinical trial outcomes, aiming to lower the $140 billion spent annually on high-risk research. While these tools have successfully signaled failures—triggering significant stock reactions for companies like Novartis—the industry remains cautious as AI occasionally struggles with complex, new drug classes.
The global pharmaceutical industry, which spends approximately $140 billion annually on clinical trials, is facing a shift as AI startups begin to predict the success or failure of drug candidates before human testing even begins. Only about 12% of drug candidates typically pass the regulatory approval process, and companies like Copenhagen-based BioinvestGPT and Tel Aviv-based QuantHealth are using DNA and clinical data to build virtual simulations of how patients might respond to experimental compounds.
This technology is now influencing how the market reacts to high-stakes drug development. A notable example occurred in 2026, when BioinvestGPT’s model successfully flagged the potential failure of Novartis’s muscular dystrophy drug, del-desiran. When the drug failed its late-stage trial in September 2026, the market reaction was swift, with Novartis shares falling 11% and erasing $30 billion in market value.
However, these AI models are not entirely foolproof. Because they rely heavily on existing data, they can misinterpret biological mechanisms when dealing with novel drug classes that lack enough historical precedent. This was illustrated by the same model incorrectly predicting a positive outcome for Novartis’s cholesterol drug, pelacarsen, which ultimately failed its late-stage clinical trial in September 2026.
This inconsistency is why major pharmaceutical companies remain skeptical. Research leadership at firms like Takeda and Biogen has noted that while AI is useful for molecule design and internal vetting, the technology is not yet ready to make definitive clinical predictions that replace human-led studies. Pharmaceutical leaders argue that algorithms cannot fully replace the unpredictable nature of biological systems, especially for new types of treatments.
Despite this caution, the regulatory environment is beginning to shift. US health regulators have started discussing initiatives to create a formal framework for how AI-driven predictive modeling fits into the drug development pipeline. For investors, the important monitorable is how these simulations evolve in accuracy. If they become more reliable, they may eventually become a standard preliminary step before major trials, which could change how pharmaceutical companies report R&D expenses and manage the financial risk of drug failure.