While artificial intelligence (AI) is creating new opportunities across biopharmaceutical manufacturing, challenges related to how manufacturing data is managed and shared are hampering the industry’s ability to scale these technologies.

A new report from Axio BioPharma, AI in Biomanufacturing: The 2026–2029 Outlook, examines the gap between the industry’s growing interest in AI and its ability to integrate these technologies into biologics manufacturing. Though AI adoption across life sciences has accelerated, the report highlights fragmented data, limited interoperability between manufacturing systems and uneven digital maturity as barriers to broader implementation.

“We started out with the report by really wanting to look at what we thought was a pretty simple question: Why are we seeing so much excitement around AI, but here in the biomanufacturing space, we’re not really seeing it scale quite as quickly,” says Justin Byers, founder and CEO of Axio BioPharma.

The report identifies the “multi-party” nature of biologics manufacturing, particularly across sponsor and contract development and manufacturing organization (CDMO) relationships, as a factor contributing to that disconnect.

“When pharma companies work to get their drugs to market, they use quite a large network of manufacturers, testing sites, and that means their data is spread across that network and it becomes fragmented,” says Byers. “You’ve got different systems, different sites, different organizations, and across all those, the same information, the same process parameters, which can be defined or titled in so many different ways, just depending on what organization you’re with.”

A manufacturer’s level of digital maturity is an important consideration when implementing AI. Byers says facilities may still rely on combinations of paper and electronic records and different approaches to tagging and organizing data.

“Even if you do have the best AI tools in the world, if you don’t have your data cleaned up in a way where you can actually use it repeatedly, it really doesn’t matter how good your tool is,” he says.

A data foundation for AI

Byers emphasized that manufacturers should first evaluate the state of their existing data infrastructure prior to deciding on how they can embed AI into their processes.

That does not necessarily mean increasing the size of a database or determining “how many different sites you can roll under one single large data lake,” he says. Rather, the data needs to contain enough context to be consistently understood and compared.

As an example, he points to pH and conductivity measurements. Manufacturers need to retain key pieces of information including about where a measurement originated in the process, the equipment involved, and why and when it was taken. That context becomes increasingly important when comparing information across production scales or facilities.

“When you start thinking about things in terms of the actual manufacturing itself, such as small- scale purification of a drug substance versus a large commercial scale, that quickly gets a lot of complexities thrown in there,” Byers added. “That information needs to be very easily, very simply accessible to anybody so that when you’re kind of exporting this and wanting to compare apples to apples, you can do it the same way each and every time.”

The challenge is compounded by the fact that biologics manufacturing is “intrinsically multi-party” in nature, according to the Axio report. Sponsors, CDMOs, and testing partners involved in a program may each use different systems and terminology to manage process information.

Byers makes the case that those differences become particularly apparent during technology transfer, when manufacturing information must move from development into clinical and eventually commercial production.

“Every one of these organizations has its own data language,” says Byers. “None of us kind of have the same language, or at least not the same dialect when it comes to sharing information.”

As a result, a liaison may be necessary to manually “translate” information between sponsors and their manufacturing partners to ensure it is interpreted consistently, Byers added. Any errors in the exchange of data can have serious manufacturing consequences.

“That transition of information, right now, is what can cause engineering batches, clinical batches — where you’re wanting this to go into a phase one or a phase two — to completely fail,” according to Byers. “You have to go back and remanufacture because there was a misunderstanding, or there was a transcription error, or a translation error from exchanging that data back and forth between the languages.”

The report proposes a federated approach to data integration in which sensitive information remains under the control of individual organizations, while standardized data artifacts and models are exchanged under shared governance.

Matching AI ambitions to manufacturing readiness

The challenges associated with fragmented data become particularly relevant as manufacturers explore more advanced applications, such as digital twins.

The report found that current AI deployments remain concentrated in certain areas including process development analytics, monitoring and advisory applications, while digital twins and advanced process control remain largely in pilot stages.

Byers says digital twins focused on individual unit operations offer a more practical starting point, as opposed to being used to model an entire biologics manufacturing process. He expressed skepticism regarding the ability to develop comprehensive biomanufacturing digital twins before manufacturers establish the infrastructure needed to connect data across different operations, sites, and organizations.

“I think we have to first get these individualized — or smaller — digital twins working really, really well before we start stitching them together into a twin that mimics the entire manufacturing process,” Byers says.

The report anticipates that digital twins of key unit operations will increasingly be incorporated into the design and operation of new and upgraded facilities, especially among advanced CDMOs.

To help manufacturers determine whether their existing infrastructure can support more advanced AI applications, Axio BioPharma developed the Biomanufacturing AI Maturity Index (BAMI), a six-level framework for evaluating AI readiness.

The framework considers both the digital maturity of manufacturing operations and the ability to integrate data and AI capabilities across organizational boundaries. It is intended to help manufacturers identify gaps in their infrastructure and governance before pursuing more advanced AI applications.

Governing AI in GMP manufacturing

Beyond data fragmentation and limited interoperability between systems, the report identifies gaps in governance maturity — another potential barrier to scaling AI in biologics manufacturing.

According to Axio BioPharma, many organizations “implement AI before mature governance, risking non-compliance or rework when models influence regulated decisions.”

Byers initially expected regulators to resist the use of AI in GMP manufacturing. He instead discovered regulators being open to how these technologies can improve processes but emphasized implementing the technology, while also continuing to meet established expectations.

“There’s just expectation that we use it in a reasonable way and that we make sure that the same thing that has always been true — that we have validation, we have change control, data integrity, and at the end of the day the humans behind these tools are still accountable to apply them in the appropriate ways,” says Byers.

Looking ahead, he expects AI to become increasingly integrated into biologics manufacturing but says adoption will be gradual and depend on manufacturers’ readiness.

“I fully anticipate we’ll see AI and AI tools being applied to nearly every part of the process,” Byers concludes. “It’s going to be a gradual change. I don’t think this is going to happen overnight. And it’s going to be based on our level of maturity that we have and how solid that foundation of our data is as well. Not only our data within our own four walls but our data across all of our different partners and how strong those connections are.”