The AI revolution is distinct from the PC and cloud eras because it makes programming accessible to everyone, not just a few million specialists.
The shift to data-center-scale computers required high-performance networking to connect thousands of GPUs, making the Mellanox acquisition essential for building AI factories.
An AI factory is a new type of data center that produces intelligence from data and energy, with networking cables being its most visible and critical component.
Agentic AI shifts from human-speed interactions to rapid machine-to-machine communication within the data center, causing an exponential increase in compute requirements.
With Moore's Law ending, scale-up (NVLink) and scale-out (networking) architectures are combined to meet the exponential demand for AI compute power.
To overcome power and land constraints, new networking technology enables multiple data centers hundreds of miles apart to function as a single, cohesive AI factory.
The CUDA platform provides a stable API that abstracts hardware complexity, allowing developers to innovate on top of NVIDIA's architecture without needing deep hardware knowledge.
Energy is a key constraint in the five-layer cake of AI, driving innovations like locating training factories near cheap power sources and running inference elsewhere.
Physical AI in robots will require on-device intelligence for immediate response, while periodically syncing with a central AI factory to update models and share learnings.
Technological advancement must be paired with mindful regulation and security guardrails to ensure AI is developed and used for the overall benefit of humanity.
A strong foundation in basics like math and physics, combined with a deep curiosity for how things work, are essential for future innovators in fields like digital biology.
The future ambition for AI is to create a world simulation that can model the long-term consequences of our actions, turning history into an experimental science.
Michael Kagan, NVIDIA’s CTO, explains how AI factories are scaling toward million-GPU data centers, what drove NVIDIA’s evolution from a chip company into an AI infrastructure company, and why networking has become the “nervous system” of AI.
Jobs that call for the skills explored in this talk.