Manufacturing Intelligence: Sizing Up Pharma's Biggest AI Deals, Part One

In the first half of a two-part deep dive on major pharma-AI partnerships, Richard Jaenisch, senior director of education, outreach, and digital experience at Open Biopharma, breaks down why Roche's raw GPU infrastructure buildout and Sanofi's regional talent hub represent fundamentally different bets on where AI adoption actually stalls in pharmaceutical manufacturing. (continued below)

Video Chapters

1. 1:18 - The biggest differences between the at least six major AI pharma moves in the last month
2. 12:30 - Which of these six is most likely to be quietly walked back or scaled down in 18 months
3. 17:50 - Whether we’re looking at in-house ML training and dedicated datacenter buildout or closed-system subscription expansion with a workforce layer bolted on, and if it matters
4. 24:14 - If the latter, what a genuine in-house build would look like in contrast
5. 30:22 - As far as the end result, be that better drugs or faster timelines, whether it matters if it’s in-house infrastructure versus a closed-system subscription
6. 36:51 - Whether Roche going big on raw GPU infrastructure and Sanofi on a regional talent hub are different best on the same bottleneck in pharma AI
7. 42:47 - If a pharma CEO on a frontier AI lab’s board is governance, oversight, or pharma keeping an eye on what’s coming out of these labs
8. 46:48 - If the embedding of pharma execs into AI labs is something to expect more of

Speaking with Chris Cole, associate editorial director at PharmTech, Jaenisch argues that compute investment and workforce investment solve different bottlenecks entirely. Roche's approach, he explains, functions as a hedge: excess GPU capacity gets monetized as a service to offset cost, and if external demand falls short, the infrastructure simply gets absorbed into internal operations. Sanofi's talent-first strategy, by contrast, targets what Jaenisch sees as the deeper obstacle to AI value capture: not hardware, but people who neither trust nor know how to use the tools they've been handed. Jaenisch notes that when it comes to AI training, "it is rare where I have talked with someone and they've said, 'Yeah, I was completely satisfied with the training that I got from my work on AI.'"

Jaenisch is skeptical of enterprise-wide rollout claims, describing most current deployments as top-down implementations that generate low-quality use cases because employees are forced into tools that don't address their actual pain points. Bottom-up adoption, he argues, produces far better outcomes because it starts from real friction points rather than mandates. He also flags a distinction pharma leadership often glosses over: whether these deals represent genuine in-house ML infrastructure or subscription-based tools with a workforce layer bolted on — a difference he says matters more to medium-sized companies, who lack the scale to benefit from proprietary systems the way large enterprises do.

On the trend of pharma executives joining frontier AI lab boards, Jaenisch is measured. He says the practice is "more than optics" but adds that "I don't think there's going to be a huge influence on how the model gets built." He predicts the real value will come not from executive-level relationships but from embedded, cross-functional project teams working directly with AI labs on specific manufacturing use cases.