GTC Paris 2025 Healthcare Special Address

Category: Healthcare Special Address · Year: 2025 · ▶ Watch

Switch language → 中文

Segments (25)

  • 00:00 · Introduction
    • Video montage showcasing NVIDIA’s impact on healthcare and life sciences over the past two decades.
  • 02:09 · Speaker Introduction
    • David Dellamonica introduces Rory Kelleher to the stage at GTC Paris.
  • 03:46 · Europe’s Legacy in Healthcare Breakthroughs
    • Rory highlights Europe’s historical leadership in life sciences, from DNA structure to mRNA vaccines.
  • 05:55 · Europe’s Core Healthcare Challenges
    • Discussion of rising costs, aging populations, and impending healthcare professional shortages.
  • 07:10 · The AI Revolution in Healthcare
    • The evolution from perception AI to generative, agentic, and physical AI.
  • 08:15 · Understanding Biology in the Age of AI
    • How wet lab technologies and AI are creating the ‘internet of biology’.
  • 09:55 · Scaling Laws for Biology
    • Applying pre-training, post-training, and test-time scaling laws to biological models.
  • 12:40 · Accelerated Computing for Life Sciences
    • Overview of NVIDIA’s domain-specific libraries and frameworks like BioNeMo.
  • 14:55 · RAPIDS-singlecell Acceleration
    • Demonstrating massive speedups in single-cell data batch integration using GPUs.
  • 16:25 · cuEquivariance for Biomolecular Design
    • Accelerating geometric-aware neural networks for drug discovery.
  • 17:45 · Announcing Boltz-2 NIM
    • Introduction of an accelerated microservice for the multi-modal Boltz-2 model.
  • 20:18 · AI Factories and DGX Cloud Lepton
    • How AI factories transform data into intelligence and the role of DGX Cloud Lepton.
  • 22:35 · Basecamp Research Customer Story
    • Basecamp Research uses DGX Cloud Lepton to train massive foundation models on proprietary biological data.
  • 24:30 · BioPharma as a Technology Industry
    • AI is transforming the entire pharmaceutical value chain, from discovery to manufacturing.
  • 26:15 · Novo Nordisk & DCAI Partnership
    • Novo Nordisk leverages sovereign AI infrastructure in Denmark to build proprietary models.
  • 28:30 · Robotics in Healthcare
    • The need for both physical robots and digital AI agents in healthcare settings.
  • 29:25 · Data Flywheel and NeMo Microservices
    • Building continuous improvement loops for AI agents using NVIDIA NeMo.
  • 31:15 · Biomedical AI-Q Research Agent Blueprint
    • A reference architecture for building AI agents that conduct biomedical research.
  • 33:05 · IQVIA Agents for Life Sciences
    • IQVIA builds healthcare-grade AI agents to assist clinical and commercial professionals.
  • 36:34 · Physical Robotics Scale in Healthcare
    • The massive addressable market for physical AI across hospitals, labs, and operating rooms.
  • 37:45 · Building Physical AI
    • Examples of partners like Philips, ImFusion, and Barco building AI-enabled medical devices.
  • 39:05 · NVIDIA’s Three Computers for Physical AI
    • The workflow of training, simulating, and deploying physical AI using DGX, OVX, and IGX.
  • 40:05 · Moon Surgical Customer Story
    • Moon Surgical discusses building the Maestro robotic surgical assistant.
  • 42:24 · Healthcare Robotics Startups in Europe
    • Highlighting various European startups building physical AI solutions on NVIDIA platforms.
  • 43:10 · Europe’s Opportunity to Lead
    • Concluding remarks on Europe’s potential to build sovereign AI healthcare services.

Product Announcements (2)

  • [17:45] Boltz-2 NIM
    • An accelerated inference microservice for the Boltz-2 multi-modal co-folding model.
    • specs: Provides up to 2x acceleration for structure and binding affinity predictions.
    • availability: Available today on build.nvidia.com.
  • [31:15] Biomedical AI-Q Research Agent Blueprint
    • A reference architecture and sample application for building biomedical research agents.
    • specs: Connects agents to tools, data, and reasoning models to automate literature review and research tasks.
    • availability: Available as a blueprint for developers to customize and deploy.

Specific Numbers (8)

Timestamp Metric Value Context
05:10 Nobel Prize Winners ~220 Number of Nobel prize winners in chemistry and medicine hailing from European countries.
06:00 Healthcare Expenditure 2 trillion euros Annual healthcare spending in Europe.
06:25 Drug Development Cost $2 billion Estimated cost to bring a new drug to market.
06:40 Labor Shortage 4 million Projected shortage of healthcare professionals in Europe by 2030.
16:00 Processing Time <4 seconds Time taken to run Harmony batch integration on a GPU using RAPIDS-singlecell, compared to ~20 minutes on a CPU.
17:05 Acceleration Factor 5x Speedup achieved for geometric-aware neural networks using cuEquivariance.
23:00 Dataset Size 9.2 trillion Number of nucleotide tokens in Basecamp Research’s BaseData database.
36:45 Operating Rooms 400,000 Number of operating rooms globally, representing a market for physical AI.

Benchmark Claims (3)

  • [16:00] Harmony Batch Integration: <4 seconds
    • vs: ~20 minutes on CPU
    • gain: Over 300x faster using RAPIDS-singlecell on GPU.
  • [17:05] cuEquivariance Acceleration: 5x faster
    • vs: Previous unoptimized implementations
    • gain: 5x faster for bio operations, 10x faster for MACE architectures.
  • [19:50] Boltz-2 NIM Inference: 2x acceleration
    • vs: Standard open-source implementation
    • gain: Twice the throughput for structure and affinity predictions.

Customer Stories (4)

  • [22:35] Basecamp Research
    • Built a massive proprietary database of biological sequences and used DGX Cloud Lepton to train foundation models.
    • outcome: Successfully training a 300 million sequence protein language model on 256 H200 GPUs.
  • [26:15] Novo Nordisk
    • Partnered with DCAI to utilize sovereign AI infrastructure in Denmark for training proprietary models across their value chain.
    • outcome: Developing models for computer-aided drug design, biomedical LLMs, and clinical development agents.
  • [33:05] IQVIA
    • Built healthcare-grade AI agents using the NVIDIA NeMo platform to assist clinical, medical, and commercial teams.
    • outcome: Created a Field Companion Agent that analyzes data to recommend and prepare sales representatives for provider meetings.
  • [40:05] Moon Surgical
    • Developed the Maestro robotic surgical assistant to provide a transversal, adaptable solution for laparoscopy.
    • outcome: Leveraging NVIDIA technologies to accelerate development and enable future AI features like simulation and real-time adaptation.

Key Technologies (6)

  • BioNeMo Microservices: A collection of optimized, containerized models for drug discovery exposed via APIs.
  • RAPIDS-singlecell: An accelerated analysis package for processing large-scale single-cell biological data on GPUs.
  • cuEquivariance: A library providing accelerated kernels for geometric-aware neural networks used in biomolecular design.
  • DGX Cloud Lepton: A unified AI platform and marketplace connecting developers to compute resources globally.
  • NeMo Microservices: An enterprise-grade platform providing tools to build, customize, and deploy AI agents and data flywheels.
  • Isaac for Healthcare: A platform utilizing three computers (DGX, OVX, IGX) for training, simulating, and deploying physical AI robots.

Demos Shown (2)

  • [13:40] A video demonstration of IQVIA’s Field Companion Agent analyzing data and preparing a sales rep for a meeting.
    • True
  • [22:05] A live-recorded demo of the Biomedical AI-Q Research Agent Blueprint conducting a literature review on cystic fibrosis.
    • True

Predictions / Commitments (3)

  • [06:40, 2030] Europe will be short around 4 million healthcare professionals by 2030.
  • [08:00, Near future] Agentic AI will reinvent scientific discovery as we know it.
  • [43:20, Next several decades] The decisions made today will define the next several decades of European healthcare AI.

Companies Mentioned (7)

Siemens · BioNTech · DeepMind · Genentech · AlphaFold, Boltz, Chai, DiffDock · MIT and Recursion · Philips, ImFusion, Barco

Notable Quotes (4)

The decisions that we make today as technology leaders, as company leaders, as country leaders, may well define the next several decades of European healthcare and life sciences. — Rory Kelleher @ 03:33

These technologies paired with robotic labs are scaling up biological data, creating what is ultimately going to be known as the internet of biology. — Rory Kelleher @ 09:00

AI factories are the essential instrument to turn data into intelligence. — Rory Kelleher @ 20:20

It is not AI who is going to replace you, it is somebody who knows how to manage a team of agents that is going to take your job. — Rory Kelleher @ 28:55

Key Topics

Sovereign AI · Generative AI · Agentic AI · Physical AI · Drug Discovery · Genomics · Single-cell Analysis · AI Agents · Healthcare Robotics · AI Factories · Accelerated Computing · Biomolecular Design

Takeaways

  • Europe has a rich history of life science breakthroughs and is well-positioned to lead in Sovereign AI for healthcare.
  • The AI revolution is progressing from generative models to agentic systems that reason and physical AI that interacts with the real world.
  • NVIDIA provides a comprehensive stack of accelerated computing libraries, such as BioNeMo, RAPIDS-singlecell, and cuEquivariance, to speed up biological research.
  • AI agents will transform healthcare workflows, acting as digital teammates for clinical, medical, and commercial professionals.
  • Building effective AI agents requires a ‘Data Flywheel’ for continuous improvement, supported by platforms like NeMo Microservices.
  • Physical AI and robotics represent a massive opportunity to scale healthcare delivery across hospitals, labs, and operating rooms.
  • Developing physical AI requires a three-computer approach: training on DGX, simulating on OVX, and deploying on IGX edge systems.