药物研发以高失败率著称：进入人体试验的候选药中，每十个最终只有一个能上市；从发现到成为药物通常需10–15年，因此每个成功药物的开发成本约为28亿美元。由于药物最终会失去专利保护，对下一个重磅药物的追逐持续不断。

生成式AI正在快速被制药业采用：AI设计分子在早期安全性试验中的成功率为80–90%，而历史平均仅为40–65%。即使后期试验成功率不变，一项模型仍显示，仅早期改进就可将全流程成功率从5–10%提高到9–18%；麦肯锡估计，若充分利用AI，每年可带来600亿–1100亿美元的增量价值。

这一变化也在重塑产业结构：美国和中国的AI原生生物技术初创公司涌现，制药公司与AI生物技术公司及亚马逊、谷歌、微软、英伟达等科技巨头加速结盟；英伟达在10月与礼来合作建设业内最强超级计算机。监管机构（如美国FDA与欧洲EMA）也开始用AI筛查海量数据；若创新成本与风险显著下降，通常提供10–15年市场独占的专利期限可能需要缩短，同时更快审评与隐私保护的数据共享将成为关键瓶颈。

Drug development is notoriously failure-prone: only one in every ten candidates that enter human trials reaches the market; it typically takes 10–15 years from discovery to medicine, making the cost per successful drug roughly $2.8bn. Because medicines ultimately come off-patent, the push for the next blockbuster is relentless.

Generative AI is being adopted rapidly in pharma: AI-designed molecules show an 80–90% success rate in early-stage safety trials versus a historical 40–65% average. Even if later-stage success does not improve, one model suggests early gains alone could raise end-to-end pipeline success from 5–10% to 9–18%; McKinsey estimates fully utilized AI could add $60bn–110bn in annual value.

The shift is also reshaping industry structure: AI-native biotech startups are emerging in America and China, and pharma firms are forming alliances with AI biotechs and tech giants such as Amazon, Google, Microsoft, and Nvidia; in October Nvidia teamed up with Eli Lilly to build the industry’s most powerful supercomputer. Regulators (including the FDA and EMA) are starting to use AI to screen mountains of data; if innovation becomes far cheaper and less risky, patent terms that typically grant 10–15 years of market exclusivity may need shortening, while faster reviews and privacy-preserving data sharing become key bottlenecks.

Source: AI is transforming the pharma industry for the better

Subtitle: It is changing the way drugs are discovered and tested

Dateline: 1月 08, 2026 07:16 上午