AI experts say most diseases could be cured within 5 to 10 years—but Bernstein says wait. On August 26, Bernstein released a report on the U.S. biopharmaceutical industry, systematically reviewing the real progress of AI in drug development: approximately 70 AI-driven projects have entered clinical pipelines, the vast majority of which remain in early stages, with the true test lying in the next 3 to 5 years.
DeepMind CEO Demis Hassabis said, “Perhaps AI can cure all diseases within 10 years,” while Anthropic CEO Dario Amodei was even more ambitious: “Most human diseases could be cured in 5 to 10 years.” Bernstein believes these claims are more akin to defenses of AI’s societal value rather than grounded in the realities of drug development. From target identification to clinical trials, and through the healthcare system’s capacity to absorb new treatments, each stage presents a bottleneck. AI may accelerate certain steps, but biology ultimately must be validated in the human body—a process that cannot be compressed by algorithms.
Improving success rate is more valuable than increasing speed.
Bernstein calculates that the R&D return from a 20% increase in success rate far exceeds the value of saving an equivalent amount of time and cost. The costs of failure are concentrated in later stages, making it more valuable to avoid failure than to accelerate the process.
The impact of AI on success rates depends on the problems it addresses. If AI is primarily used for molecular optimization targeting known sites, success rates may increase, but the value of these “easy problems” is limited. If AI is applied to tackle difficult-to-treat diseases with insufficient biological understanding, even if overall success rates remain low, the value created can be significantly greater.
Bernstein points out a counterintuitive conclusion: the more AI is used, the more wet lab experiments may be required. The hardest problems are facing the greatest scarcity of training data, and generating new data demands substantial experimental investment. AI and wet lab experiments are complementary. The cost of conducting one additional wet lab experiment is far lower than the cost of a project failing in later stages.
Pharmaceutical companies' AI strategies: Eli Lilly has the broadest coverage, Amgen the deepest integration, and Roche the greatest computational power.
Bernstein analyzed 183 AI drug discovery partnerships among 15 major pharmaceutical companies, revealing significant differences in their strategies.
Eli Lilly has the broadest external collaboration network, with 25 transactions since 2019 and publicly disclosed upfront payments totaling approximately $353 million. Partners include Insilico Medicine, Profluent, Verge Genomics, and Genesis Therapeutics. Eli Lilly also operates an NVIDIA Co-Innovation AI Lab, achieving the most balanced integration of external reach and internal computational power.
Roche ranks second with 20 collaborations, clearly focused on oncology (8) and neuroscience (4). Genentech’s hybrid AI factory has been called by Bernstein “the largest locally deployed GPU setup announced in the industry.” Recursion’s $1.2 billion collaboration is the largest single deal.
Amgen has the fewest external partnerships (5), yet the most differentiated internal AI and computing infrastructure. The acquisition of deCODE Genetics in 2012 provided an industry-leading population genomics engine capable of identifying and validating disease-related genetic variants to prioritize targets. Amgen has trained five proprietary antibody language models on its local DGX SuperPOD, establishing a clear data moat.
AstraZeneca has 19 transactions, with economic activity heavily concentrated in three agreements with CSPC Pharmaceutical Group, accounting for the vast majority of its disclosed upfront payments.
Novo Nordisk's 10 collaborations are all concentrated after 2022, with seven focused on cardiovascular, metabolic, and kidney diseases, making it the largest pharmaceutical company with the most targeted therapeutic focus.
Pfizer has the most platform-based partnerships (nine across its entire pipeline); its AI foundation model relationships date back to IBM Watson in 2016, with no identified collaborations involving dedicated computing power.
Bernstein emphasized that these are all external perspectives; truly integrating quality and internal cultural transformation is difficult to quantify.
AI Biotechnology: Clinical assets are the true litmus test.
Approximately 70 AI-related projects are primarily focused on publicly traded companies, with definitions varying widely—from AI-driven drug repurposing to AI-powered target discovery combined with AI-based molecular design.
Insilico’s Rentosertib is a representative case of end-to-end AI-driven drug discovery. The AI platform PandaOmics identified TNIK as a target for idiopathic pulmonary fibrosis, and Chemistry42’s generative AI designed the molecule, reducing the timeline from target hypothesis to preclinical candidate to approximately 18 months. The drug has now entered Phase III trials, with primary endpoint readout expected in October 2029—making it one of the most significant data points in the AI drug discovery industry.
Recursion partners with Roche, Bayer, and Sanofi through cell imaging and multi-omics-driven target discovery.
Generate Biomedicines is designing protein therapeutics; GB-0895 (an anti-TSLP antibody) has entered Phase III to validate whether AI-designed proteins can be translated into differentiated dosing regimens.
Relay Therapeutics designs drugs through protein motion modeling; zoogalalisib (a PI3Kα inhibitor) has entered Phase III to test whether conformational selectivity can improve the therapeutic window.
Schrödinger adopted a physics-based computing priority strategy, and Zasocitinib (a TYK2 inhibitor) has been submitted for market approval. This target has already been clinically validated by BMS’s deucravacitinib, making it more of a success story in computational-assisted optimization rather than an AI-driven breakthrough in novel target discovery. The timeline from target selection to clinical testing remains 8 to 10 years, unchanged from the industry standard.
FDA approval data has not yet reflected the impact of AI.
Over the past five years, the FDA has approved an average of 48.5 new drugs annually, approximately 43% higher than the average of 33.8 per year from 1985 to 2025. Bernstein believes this increase reflects regulatory modernization, the rise of biologics, and the cumulative impact of past R&D investments, rather than a visible contribution from AI. Given that AI pipelines remain small and clinical validation still requires time, this situation is unlikely to change in the near term.
Bernstein's conclusion is pragmatic. AI has already played a role in molecular design, virtual screening, and target identification, with some early projects advancing to late-stage clinical trials. "Curing all diseases" requires overcoming multiple bottlenecks—including biological understanding, clinical trials, regulatory approval, and healthcare system capacity—that cannot be solved by algorithmic iteration alone. The real value of AI in drug discovery may lie in helping the industry make better decisions, not in accelerating processes. The true answer will take another three to five years to emerge.
Disclaimer
This article is a compilation and interpretation by Chaoxiang Research of a third-party brokerage research report (Bernstein, August 26, 2026), combined with publicly available market information. The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution, not the views of Chaoxiang Research, nor do they constitute any investment advice.
The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.
Written by: Rita