Bristol Myers Squibb Building Life Science Industry’s Most Advanced AI Factory on NVIDIA Vera Rubin

Erin Davis has a playful nickname for the latest technological leap at pharmaceutical titan Bristol Myers Squibb (BMS): the “SuperDuperPOD.” While the name is lighthearted, the implications for the future of medicine are profoundly serious. BMS, already a leader in utilizing massive AI clusters to accelerate drug discovery, is significantly scaling its capabilities. The company announced today that it is deploying its second NVIDIA DGX SuperPOD, a state-of-the-art infrastructure built upon eight NVIDIA DGX Vera Rubin NVL72 systems. This deployment marks the creation of what is arguably the most powerful and energy-efficient AI cluster currently operating within the life sciences sector.

Democratizing Access to Limitless Compute

For Davis, who serves as the vice president of research business insights and technology at BMS, the primary goal of this massive infrastructure investment is accessibility. Historically, supercomputing power was a bottleneck, restricted to a select few researchers with specialized technical knowledge. That era is coming to an end at BMS.

“Instead of equipping a small group of researchers with access to the supercomputer, we’re opening it up to literally every scientist. No one has to wait, and no one is told they have a limit.”

The new rack-scale systems, which integrate NVIDIA Vera CPUs and Rubin GPUs, offer a staggering 10x increase in performance per megawatt compared to the legacy infrastructure they replace. By providing a unified AI platform—complete with the NVIDIA BioNeMo Agent Toolkit—BMS is empowering its global research teams to run complex predictions, train foundational models, and deploy agentic workflows across the entire drug discovery lifecycle. The objective is to shift the focus of the company’s scientists away from the logistical hurdles of resource management and back toward the core of scientific innovation.

Translating AI Potential into Measurable Impact

Payal Sheth, who stepped into the role of senior vice president of therapeutic discovery sciences at BMS earlier this year, emphasizes that the company has moved past the phase of theoretical AI exploration. The mandate today is clear: move from abstract concepts to tangible, measurable clinical impact.

BMS has successfully utilized its first DGX SuperPOD for the past three years, yielding significant results. AI-driven target identification has already reclaimed weeks of manual labor for scientists, allowing them to prioritize high-value decision-making. Furthermore, the team has leveraged AI to broaden its library of CELMoD compounds—molecules designed to selectively degrade proteins linked to cancer. This capability has opened new avenues for treating blood cancers and various other diseases.

The “Predict First” Methodology

A cornerstone of this transformation is what Sheth describes as the “Predict First” approach. By utilizing computational models to simulate molecular behavior before moving to the laboratory, researchers can gate their experiments more effectively.

• Multi-parameter optimization: AI helps prioritize the synthesis of molecules that are most likely to succeed.
• Risk mitigation: By weeding out molecules that fail to meet specific property landscapes, the company ensures that precious laboratory resources are directed toward the most promising candidates.
• Anticipatory design: Scientists can now model how a molecule will behave in a clinical setting long before it reaches the trial phase.

Overcoming the Compute Bottleneck

The demand for this technology is reaching a fever pitch. “We’re saturated,” Davis admits. The organization is currently engaged in large-scale predictions involving complex molecules and the development of proprietary foundational models, both of which are incredibly GPU-intensive.

Davis, a computational chemist who spent 15 years in the vendor space at companies like Schrödinger and X-Chem, understands the frustration of technology failing to keep pace with scientific ambition. For her, this mission is also deeply personal. Having lost her father to Alzheimer’s five years ago, she is acutely aware of the urgency required in drug discovery.

“Even if he was still going to die,” Davis reflects, “if there was symptom remediation along the way, it would have saved suffering for everybody in the family. Dementia is especially cruel.”

A Unified Global Intelligence Framework

To maximize the impact of the new hardware, Davis’s team is merging the existing DGX SuperPOD with the new Vera Rubin-powered system into a single, cohesive data plane. This environment will be accessible to every BMS site worldwide, effectively dismantling the silos created by past acquisitions and technical barriers.

Key Features of the New AI Architecture:

• NVIDIA Mission Control: AI-native tooling that simplifies management and orchestration.
• Natural Language Interfaces: Researchers can initiate complex predictive tasks using plain English, removing the need for deep computational expertise.
• Cumulative Learning Loops: Data generated in Lawrenceville, New Jersey, can now inform models used by teams in San Diego, California, ensuring that every experiment compounds into institutional intelligence.

As Sheth notes, the industry previously treated every project as a discrete event. Today, BMS is building a discovery system where every clinical readout and partnership feeds into a higher-conviction scientific framework.

The Rise of Agentic Workflows

The integration of agentic workflows represents the next frontier for BMS. These AI agents, which can operate across various silos and programs, are viewed as a “game-changer” by Davis. They allow for a level of cross-pollination of data that was previously impossible.

“When you as a scientist can go to an army of well-vetted, fully trained virtual scientists that have BMS knowledge baked in, now you’re a whole team in and of yourself.”

While the technology is powerful, both Davis and Sheth stress that human intuition remains the driving force. The AI does not replace the scientist; it augments their capabilities, providing the quantitative insights necessary to make faster, more accurate decisions.

When asked by BMS Chief Digital and Technology Officer Greg Meyers if she was certain the team could fully utilize the massive capacity of the new SuperDuperPOD, Davis’s response was characteristically confident: “Just give us time.” With a roadmap already in place covering everything from small molecule design to the creation of digital twins, BMS is clearly positioning itself to lead the next generation of AI-driven pharmaceutical innovation.