At VivaTech 2025, TetraScience CEO Patrick Grady joined NVIDIA’s Global Head of Healthcare, Rory Kelleher, and Sanofi Chief Digital Officer Emmanuel Frenehard for a panel on The Rise of Scientific AI. Their discussion highlighted both the transformative potential of AI in biopharma and the structural barriers that must be overcome to realize it. Here’s what stood out. (Full transcript below.)
The Data Dilemma
All three panelists agreed: Scientific AI’s greatest obstacle isn’t algorithms or compute power, it’s data readiness. As Grady noted, biopharma generates 20 exabytes of complex, siloed data annually, but “raw data trapped in vendor-specific formats has zero utility for AI.” Sanofi’s Frenehard emphasized the need for longitudinal, anonymized patient data to build accurate “digital twins” for drug testing, while Kelleher stressed that AI’s success in language models stemmed from open, internet-scale data, a model biology lacks.
Collaboration Over Contention
The panel underscored the necessity of cross-industry collaboration. NVIDIA’s Kelleher highlighted partnerships with TetraScience and Sanofi to industrialize AI-native datasets, calling it a “moral imperative” to fix biopharma’s collapsing economic model (Eroom’s Law). Frenehard advocated for global data-sharing frameworks, arguing that “patients don’t have borders,” while Grady positioned TetraScience’s Scientific Data and AI Cloud as the bridge between lab data and AI infrastructure: “We’re here to replatform, engineer, and activate data, not replace scientists.”
From Hype to Reality
While optimism about AI’s potential ran high throughout the entire conference, in this panel discussion the speakers grounded their expectations in reality. While Kelleher outlined NVIDIA’s vision for “science factories” combining AI-driven hypothesis generation with robotic labs, Frenehard cautioned that human biology’s complexity demands humility: “We’re still decoding the ‘source code’ of life.” Grady added that TetraScience customers already see tangible wins, like cutting out 80% of the time it takes to perform lead clone selection, proving AI’s value in optimizing existing workflows.
A Call for Sovereign Innovation
The discussion closed with a nuanced debate on data sovereignty. Kelleher championed regional AI strategies to safeguard critical health data, while Frenehard warned against protectionism stifling global science. Grady reconciled both views: “Sovereignty isn’t about walls—it’s about enabling nations to harness AI securely while federating insights for humanity’s benefit.”
Watch the Full Discussion
For more insights on AI’s role in reversing Eroom’s Law, the ethics of data sharing, and the future of automated drug discovery, view the full VivaTech 2025 panel:
Transcript
Moderator (Charlie Perreau, Les Echos): Hello, everyone. Thank you very much for being with us at VivaTech today. I’m really glad to introduce our panelists: Patrick Grady, CEO of TetraScience; Rory Kelleher, Global Head of Business Development at NVIDIA; and Emmanuel Frenehard, Chief Digital Officer at Sanofi.
Let’s start with you, Rory. NVIDIA is a bit like the superstar of VivaTech this year. Most people know NVIDIA from the gaming industry and, of course, for AI. Can you tell us more about what you’re doing in the healthcare space?
Rory Kelleher (NVIDIA): Absolutely. About 30 years ago, NVIDIA pioneered a type of computing called accelerated computing, fundamentally reinventing the computing architecture for the first time in 60 years. This platform has been central to the rise of modern AI. But accelerated computing isn’t just about GPUs; these GPUs power systems that fill data centers—these are now the modern infrastructure of computing.
At NVIDIA, we also build a lot of software on top of our hardware: accelerated libraries, inference microservices, models, frameworks, and blueprints that make it easier for developers to build and deploy applications across various industries. I spend much of my time in healthcare and life sciences, and we focus on three areas:
- Digital Instruments: We started in healthcare about 10 years ago, working with diagnostic imaging companies to transform raw sensor data into interpretable images for physicians. Today, we collaborate with nearly every medical device company to make their instruments software-defined and AI-enabled.
- Digital Health: We use AI agents to transform the patient experience, from booking appointments to making routine visits more human—so doctors can focus on patients, not just their computers.
- Digital Biology: This is a huge opportunity. It still takes about 10 years and $2 billion to bring a drug to market. There’s massive potential for efficiency across pharma’s value chain, especially in early discovery. Model-driven drug discovery is becoming industry standard; every pharma company is using this today.
Charlie Perreau: So, do you all work together?
Patrick Grady (TetraScience): We do. We try to work very closely—he [Emmanuel] has all the money! But yes, we work together.
Charlie Perreau: There’s a lot of hype around AI in healthcare: from discovering nutrients to patient twins, to early cancer detection. But what’s real today, Patrick?
Patrick Grady: The hype versus reality is interesting. TetraScience works across the biopharma landscape, with about 40 leading biopharma customers, including 12 of the top 25. We see "green shoots" of innovation—AI is being applied across the value chain, from in silico and digital twin technologies in drug discovery, through development, and into early-stage manufacturing innovation.
But it’s a mixed bag. The main obstacle to scientific AI isn’t a lack of data—companies like Sanofi produce enormous amounts. The challenge is data preparation: availability, organization, and readiness. McKinsey’s recent survey found 80% of top pharma respondents cited data readiness as the number one obstacle to achieving scientific AI. We’re excited by the possibilities, but until we solve these structural data problems, we’ll keep talking about hope rather than fulfillment.
Charlie Perreau: And when we think of data, we often think of structured data—names, numbers, etc. But in your industry, data could be scans, blood samples, biopsies. How do you store and exploit that?
Patrick Grady: That’s where technology can really help us. The industry needs to collaborate much more, treating this as a partnership for humanity and patients, not just a standard customer-vendor relationship. None of us can deliver the final product alone.
In preclinical R&D, there are about 10,000 biopharma companies globally, generating around 20 exabytes of data—a huge, complex, and rapidly growing dataset. Solving this requires the expertise of pharma, data infrastructure companies, and AI specialists working together.
Rory Kelleher: In natural language AI, progress has been rapid because data was free and open on the internet. In biology, there’s no "internet of biology"—data is siloed within companies and institutions. To realize scientific AI, we need industrial-scale generation of experimental data, and we need to bring people along so they actually use these new tools. Change management is hard—sometimes old ways need to give way to new ones.
Patrick Grady: The biopharma industry has long operated like an artist colony, giving brilliant scientists the freedom to pursue their work. That’s transformed humanity, but we’ve reached diminishing returns. Now, we need to augment those minds with industrialized scientific data and use cases. Change management in the scientific community is a major challenge.
Emmanuel Frenehard (Sanofi): Adoption is always a challenge, no matter the industry. Technology adoption is 70% of the difficulty; we’re proud of building the tech, but it can quickly become orphaned if users don’t embrace it.
On data: our intention is to develop drugs by simulating as much as possible—creating molecules, testing them virtually against simulated patient populations before ever going to human trials. To do that, we need more longitudinal, anonymized patient data—complete patient histories—to create digital twins. My plea to health agencies and individuals: share your data. It could help future generations.
Rory Kelleher: Most of our work in healthcare at NVIDIA has focused on early discovery, where
models generate molecules with desired characteristics. But it still takes 10 years to bring a drug to market. If we can accelerate clinical trials—where 70% of costs and time are spent—the impact would be massive.
Patrick Grady: In tech, Moore’s Law means performance doubles every 18 months. In biopharma, there’s "Eroom’s Law" (Moore’s Law spelled backwards): it takes longer and costs more each year to bring a drug to market—now 12+ years and $2.5–6 billion. The economic model is collapsing. Our goal is to reverse that trend through AI, optimizing every step and cutting years and billions from the process.
Charlie Perreau: What about sovereignty in your sector? Sanofi, I’m sure, is concerned about data sovereignty. What about NVIDIA?
Rory Kelleher: We’re a global platform, working with every cloud provider and OEM. We believe it’s important for nations to have a sovereign AI strategy, especially for public health and economic development. That’s why we’re supporting Europe in building sovereign AI healthcare services.
Patrick Grady: If AI is the most powerful tool ever wielded, then there’s a moral imperative to deliver scientific AI for all, save the biopharma industry, and ensure every nation can safeguard its most valuable data. Sovereignty isn’t about American companies selling things—it’s about enabling nations to control their data and use AI safely and securely.
Emmanuel Frenehard: I’d add that sovereignty shouldn’t be an excuse for protectionism and overregulation. We’re moving from a globalized world to a more fragmented one, but patients don’t have borders. Protectionism and sovereignty are not the same—let’s not conflate them.
Patrick Grady: I agree. Science shouldn’t be politicized. We need a balanced approach—science at its center.
Charlie Perreau: What borders remain for AI? What’s coming in the next 10–20 years?
Rory Kelleher: In the past decade, we’ve moved from perception AI to generative AI, and now to agentic AI—where agents can reason, interact, and use tools. The next frontier is connecting digital agents with physical robots, especially in science. Imagine digital agents generating hypotheses, designing experiments, and calling on automated robotic labs to run those experiments—creating a feedback loop that accelerates discovery. Science "factories" will emerge, generating experimental data at scale.
Patrick Grady: This is on the near-term horizon if we solve the data challenge. It sounds futuristic, but breakthroughs are close.
Emmanuel Frenehard: We live in an incredible time. AI is accelerating what was once impossible. But we must remain humble—human biology is complex, and personalized medicine is still a distant goal. The "virtual cell" is the next big project: mapping disease at the cellular level to predict which treatments will restore health. It’s an exciting vision, requiring industrial-scale digital biology and talented researchers.
Charlie Perreau: Thank you all. This was a fascinating discussion. I’m sure we’ll hear much more about AI in healthcare in the months to come.