Companies have claimed improvements to yield, batch consistency and output while acknowledging the risks and challenges created by the technology.
Pharma’s rush to adopt artificial intelligence is changing manufacturing. Seeking to improve success rates, yields and many other variables, drugmakers are exploring AI across their manufacturing chains, with upstream, downstream and fill-finish uses all under consideration as targets. The technology comes with new risks and challenges, but companies are forging ahead in pursuit of potential efficiency gains.
The term AI is used to refer to a wide range of technologies, from machine learning systems that analyze images to large language models that help companies manage and interpret complex datasets. While older computational technologies excel at logging data and retrospective analyses, AI tools can spot patterns in large datasets and enable predictive analytics and maintenance.
Versions of some of the technologies entered pharma manufacturing plants years ago, with executives at Pfizer and Roche discussing the use of machine learning as early as 2018 and 2019, respectively. More and more companies have outlined their use of the technology since those early examples.
Joe Margarones, head of digital INT at Moderna, told BioSpace that near-term opportunities include the ability to “chat with our data,” enabling employees to “ask questions, explore trends and generate canvases using natural language.” Moderna also anticipates user configurations of prebuilt and custom agents to suit operational and quality needs as a near-term use.
Biogen signaled its intent to invest further in advanced automation and AI when it committed $2 billion to a manufacturing expansion in July. Nicole Murphy, head of pharmaceutical operations and technology at the company, told BioSpace the biggest near-term opportunities include predictive maintenance of equipment and automating investigations of the root causes of deviations (i.e., departures from approved instructions or standards).
Sanofi has identified similar near-term uses for AI. A spokesperson told BioSpace that the company is “reducing downtime, extending equipment life and improving overall operational efficiency” by adopting AI-driven predictive maintenance. Like Biogen, Sanofi is applying AI to deviations management, with the spokesperson saying the technology has “reduced closure time for minor deviations drastically.”
At Biogen, Murphy’s other near-term uses include the application throughout the manufacturing process of digital twins and soft sensors, data-driven models that digitally represent physical processes to provide insights beyond direct measurement. Using the technologies, Murphy said drugmakers can “enable adaptive process control to improve yields and batch-on-batch consistency.”
In 2023, AstraZeneca said digital twins for raw material planning drove a 90% reduction in dispensing planning time. Roche has hailed the impact of digital twins on its ability to predict cell age and growth for production cell lines. The company told investors last year that the technology increased production yields by 10% and quality by 40%.
One of Sanofi’s near-term use cases for AI is yield optimization. The spokesperson said Sanofi has seen substantial benefits from its AI-powered yield analytics platform. The tool enables manufacturing teams “to spend less time on data analysis and more time acting on insights, resulting in consistently higher yields and optimized use of raw materials,” the spokesperson said.
Pam Cheng, EVP of global operations, IT and chief sustainability officer at AstraZeneca, has credited the use of “over 30 digital tools and AI solutions for selected processes and products” with increasing output at a facility in China by 55%. Cheng, who told investors about the improvements in May, added that there was a 44% fall in lead time and a 54% boost to productivity.
Despite the improvements, overall manufacturing costs at at least some companies have continued to rise. Roche reported a 14% increase in manufacturing cost of goods sold and period costs in 2024. Costs rose by a further 7% in the first half of 2025. At AstraZeneca, which bundles manufacturing costs with royalties and other costs of sales, the cost of sales increased 20% over last year and, at the midpoint of the year, was on course to rise again in 2025.
Risks of Action, Risks of Inaction
AI is creating new challenges. Murphy said regulatory compliance and upskilling the workforce are the biggest challenges to deploying AI in good practice environments. Margarones hit on similar points, saying that as AI advances it requires new skills and governance and that regulators expect interpretable outputs for critical decisions. It is unclear what happens within some AI tools to produce outputs, making it hard to interpret and validate their findings.
Margarones added that there are AI data integrity and traceability challenges because models must be version-controlled, audit-trailed and compliant with FDA 21 CFR Part 11 regulations governing electronic records and signatures. An additional challenge, the Moderna digital leader said, is the need for a “continuous validation” framework that marries machine learning operations with standard operating practices for quality assurance.
The cybersecurity implications of AI tools create further challenges. Margarones noted that such tools increase a company’s attack surface. Connected sensors, cloud-based platforms and other technologies that collect and manage data for AI models create potential new entry points for attackers. Drugmakers now routinely list AI in disclosures to investors about the risks they face, including in some cases in relation to the impact of the tools on manufacturing and supply.
Moderna, noting its level of equipment automation and integration of AI systems, told investors in its 2024 annual report that facility digitization exposes it to the risk of process equipment malfunctions. The risks include system failures or shutdowns because of design issues, system compatibility problems or cybersecurity incidents.
While acknowledging the risks, drugmakers have concluded that choosing not to adopt AI would pose a bigger threat to their businesses. AbbVie summed up the risk of inaction in its annual report, telling investors that the “failure to effectively implement these technologies could hinder our ability to compete, as competitors’ advancements in AI may lead to more efficient operations.”
