GTC DC '25 Pregame - Chapter 4: AI for Science
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
This special GTC edition of the NVIDIA AI podcast explores how AI is revolutionizing scientific discovery across multiple fields. The episode features industry leaders discussing quantum computing, drug discovery, genomics, and chip design, emphasizing how "science used to move at the speed of experiments, now it moves at the speed of compute." (00:51) The conversation highlights the intersection of classical and quantum computing, the acceleration of drug discovery through AI, and the three-phase evolution of AI adoption across industries.
- Main theme: The transformation of scientific research through AI acceleration, with particular focus on quantum computing integration, molecular design, and the shift from artisanal to engineering-based drug discovery processes.
Speakers
George Church
Chief Scientist at Leila Sciences, a pioneering researcher in genomics and synthetic biology. Church has made significant contributions to gene therapy and molecular engineering, with recent work including proteins targeting the nervous system with 100x better precision while reducing liver toxicity.
Matt Kinzella
CEO at Inflection, bringing 19 years of investment experience to quantum technology commercialization. Before joining Inflection, he was an investor at Maverick Capital and has positioned the company to achieve 12 logical qubits as of recent announcements.
Mark Tessier-Levine
Co-Founder, Chairman and CEO of Zyra Therapeutics, focused on applying AI to drug discovery and molecular design. His company is working to transform drug discovery from an artisanal process into an engineering discipline using foundation models of biology.
Anirudh Devgan
President and CEO of Cadence, leading a company that creates software products for chip and electronic system design. Devgan has over 20 years of experience working with NVIDIA and oversees Cadence's molecular science division that applies similar mathematics to both chip design and drug discovery.
Key Takeaways
Science Now Moves at the Speed of Compute
The fundamental paradigm of scientific discovery has shifted from being limited by experimental timelines to being accelerated by computational power. (00:51) This transformation enables researchers to model everything "from the atom to the atmosphere" and process data as fast as it can be collected. The intersection of exponential improvements in both computation and biology, including a 20 million-fold reduction in sequencing costs, creates unprecedented opportunities for breakthrough discoveries. This shift allows scientists to move from reactive analysis to predictive modeling, fundamentally changing how research is conducted across disciplines.
Hybrid Computing Systems Will Define the Future
The future of scientific computing isn't about choosing between classical, GPU, or quantum systems—it's about creating hybrid architectures that leverage each technology's strengths. (26:07) As Devgan explains, "CPUs are still important... FPGAs, custom silicon. So I think it's not either or thing. Quantum will have certain big applications. It will give dramatic speed up, but all the other hardware platforms will work together." This approach maximizes computational efficiency by matching specific problems to the most suitable computing paradigm, whether that's CPUs for general processing, GPUs for parallel tasks, or quantum processors for specialized algorithms.
AI is Transforming Drug Discovery from Art to Engineering
Traditional drug discovery is highly inefficient, taking 13 years on average with a 90% failure rate and costs of $2-4 billion per approved drug. (12:03) AI promises to transform this "artisanal endeavor into an engineering discipline" by enabling molecular design rather than just screening. Recent breakthroughs like AlphaFold and RF diffusion, recognized with Nobel Prizes, now make it possible to design drugs computationally—similar to asking AI to create specific content. The key is generating massive amounts of biological data to train AIs that can understand biological patterns humans cannot detect, moving more of the process from wet labs to computer simulations.
Logical Qubits Are the Key to Quantum Advantage
Quantum computing's real potential lies in achieving logical qubits—error-corrected quantum bits that can perform reliable computations. (23:12) Until 2023, logical qubits didn't exist, but now companies like Inflection have achieved 12 logical qubits. The scaling timeline suggests that around 100 logical qubits will enable quantum advantage in material science, while 1,000 logical qubits could revolutionize drug discovery. This represents orders of magnitude improvements—not just 50% or 100%, but potentially 10,000x to 1,000,000x performance gains in specific applications where quantum mechanics provides inherent advantages.
Three-Phase AI Evolution Spans Decades of Growth
AI development follows three distinct horizons with massive market potential. (27:08) Phase 1 focuses on infrastructure buildout, LLMs, and cloud applications—where we currently are with years of growth remaining. Phase 2 involves "physical AI" expanding into cars, planes, drones, and robots, representing trillions in monetization potential. Phase 3 encompasses AI for science, including drug discovery and material science, also worth trillions. Importantly, these phases reinforce each other—physical AI robots still need data center training, and scientific AI applications require robust computational infrastructure, creating compounding demand across all phases.
Statistics & Facts
- Drug discovery currently takes an average of 13 years from molecular target identification to FDA approval, with 9 out of 10 drugs failing in clinical trials and costing $2-4 billion per approved drug. (12:15) Mark Tessier-Levine provided this statistic to illustrate the inefficiency of current drug discovery processes that AI aims to transform.
- There has been a 20 million-fold reduction in the cost of genetic sequencing, as mentioned by George Church. (15:17) This dramatic cost reduction has made genetic sequencing not just a population study tool but a routine analytical and experimental guidance tool in laboratories.
- In chip design, 99% of the work is done on computers with "first time right silicon," while in drug discovery, only a few percent of work is currently done computationally. (17:36) Anirudh Devgan used this comparison to show the massive potential for computational acceleration in drug discovery as it follows the same path as chip design evolution.
