Nvidia GTC: CEO Jensen Huang delivers keynote address

The Dawn of a New Computing Era [00:00]

• The narrative begins by framing America as the "land of innovation," where key inventions like the transistor at Bell Labs and Hedy Lamarr's work on wireless communication laid the groundwork for technological revolutions.
• Major milestones such as IBM's System 360, Intel's microprocessor, and Cray supercomputers are cited as driving forces in computing's advancement.
• Apple's personal computing and Microsoft's software contributions are acknowledged, alongside the foundational work of ARPANET for the internet.
• A revolutionary new computing model, driven by AI, is introduced as the catalyst for the next era, with predictions of it being a significant contribution to the computer industry.

"This is likely going to be the most important contribution we've made to the computer industry. It will likely be recognized as a revolution."

Accelerated Computing: The Core of Innovation [06:01]

• NVIDIA is credited with inventing a new computing model for the first time in 60 years, addressing problems general-purpose computers cannot.
• The end of Dennard scaling around 2006 is identified as a critical point where transistor performance and power efficiency slowed, despite increasing transistor counts.
• NVIDIA has spent 30 years developing "accelerated computing" through the invention of the GPU and the CUDA programming model.
• Accelerated computing combines parallel processing GPUs with sequential CPUs to extend computing capabilities, marking an inflection point that has now arrived.
• A fundamental shift in programming models is required for accelerated computing, necessitating new algorithms, libraries, and rewritten applications, a process that has taken nearly 30 years.

"We made this observation a long time ago and for 30 years we've been advancing this form of computing we call accelerated computing."

The Power of CUDA X Libraries [09:00]

• CUDA X is presented as the "treasure" of the company, comprising hundreds of libraries designed to enable accelerated computing across various domains.
• Examples of CUDA X libraries include those for computational lithography (KU Litho), sparse solvers for CAE, numerical optimization, simulations, data frame acceleration (QDF), and large language model training (CouDNN, Megatron Core).
• Monai, a leading medical imaging AI framework, is highlighted, with further details promised in a healthcare keynote.
• CUDA X also supports genomics processing and quantum computing (CUDA Quantum), representing a diverse set of applications.
• These libraries are crucial for allowing ecosystem partners to leverage accelerated computing and for opening new markets.

"And each one of these libraries redesigned the algorithm necessary for accelerated computing."

The AI Revolution and Infrastructure Needs [17:36]

• AI is identified as the "new industrial revolution," requiring specialized infrastructure, particularly NVIDIA GPUs.
• Like electricity and the internet, AI is deemed essential infrastructure that every company and nation will utilize.
• The competition in AI is framed as a test of national capacities, comparable to the space age.
• AI factories are emerging to support scientists, engineers, and innovators across various sectors.
• The presentation pivots to the telecommunications industry, noting that wireless technology, once led by American innovation, is now largely deployed on foreign technologies.

"Artificial intelligence, the new industrial revolution."

NVIDIA Arc for 6G and AI in Telecommunications [17:43]

• NVIDIA announces a new product line, NVIDIA Arc, an aerial radio network computer designed for 6G and AI integration in telecommunications.
• Arc is built on three core technologies: the Grace CPU, Blackwell GPU, and ConnectX networking.
• The Aerial CUDA X library enables Arc to function as a software-defined, programmable computer capable of wireless communication and AI processing simultaneously.
• A partnership with Nokia, a major telecommunications equipment manufacturer, is announced to integrate NVIDIA Arc into their base stations, aiming to bring telecommunication technology back to America.
• AI for RAN (Radio Access Network) will enhance spectral efficiency and enable AI on RAN, creating a new edge industrial robotics cloud on top of the wireless network.

"This is completely revolutionary. We call it Nvidia Arc."

Quantum Computing: A New Frontier [21:49]

• The concept of quantum computing, envisioned by Richard Feynman in 1981 to simulate nature, has seen a fundamental breakthrough 40 years later.
• The ability to create one logical, coherent, stable, and error-corrected qubit has been achieved, though it requires tens to hundreds of physical qubits.
• Quantum computers are susceptible to decoherence, necessitating well-controlled environments and extensive error correction.
• NVIDIA is developing NVLink, an interconnect architecture to directly connect quantum processors with NVIDIA GPUs for error correction, calibration, and control.
• The CUDAQ platform allows for quantum GPU computing, enabling researchers to orchestrate quantum devices and AI supercomputers for hybrid simulations.
• Quantum computing will not replace classical systems but will work together in a fused accelerated quantum supercomputing platform.

"We now realize that it's essential for us to connect a quantum computer directly to a GPU supercomput so that we could do the error correction."

The Evolution of AI: From Chatbots to Universal Intelligence [30:52]

• AI is broadly defined beyond chatbots, encompassing fundamental science and AGI, requiring deep computer science and groundbreaking computing.
• The AI computing stack has been reinvented, moving from hand-coded software on CPUs to data-intensive machine learning training on GPUs.
• Energy is identified as a critical resource for AI growth, with credit given to a "pro-energy initiative."
• AI models learn by tokenizing various forms of information, including text, images, video, 3D structures, chemicals, and genes, enabling translation, response, and generation capabilities.
• The future of AI involves a vast array of models (Transformers, CNNs, state-space, graph neural networks) and tokenization methods, optimizing for different types of data and tasks.

"AI is machine learning training, data inensive programming, if you will, trained and learned by AI that runs on a GPU."

AI as Work, Not Just a Tool [36:38]

• A profound distinction is made between AI as a "tool" (like Excel or Word) and AI as "work."
• AI "workers" can use tools and perform tasks autonomously, fundamentally changing the economy by addressing a segment previously untouched by traditional IT tools.
• This new paradigm has the potential to engage the entire trillion-dollar global economy, making it more productive and driving growth.
• AI is also a new industry in itself, requiring "AI factories" to produce tokens at high rates and cost-effectively.

"AI is not a tool. AI is work."

The Exponential Demands of AI and the Virtuous Cycle [47:27]

• The growth of AI is driven by two exponentials: the increasing compute requirements from new scaling laws and the increasing usage of smarter models.
• This dual exponential growth is occurring at a time when Moore's Law has largely ended.
• NVIDIA has achieved a "virtuous cycle" for CUDA over 30 years, and now a similar cycle for AI is emerging after 15 years.
• This cycle involves smarter AI models leading to more usage, generating profit, which funds more compute, making AI smarter, and so on.
• Driving down costs is crucial to maintain this cycle, improving user experience and enabling continued innovation.

"These two exponentials, one is the exponential compute requirement of the three scaling law. And the second exponential, the more pe the smarter it is, the more people use it, the more people use it, the more computing it needs."

Extreme Co-Design and the Future of AI Factories [49:49]

• "Extreme co-design" is NVIDIA's approach to overcome the limitations of traditional chip design by architecting new computer architectures, chips, systems, software, and applications concurrently.
• This approach allows for compounding exponential improvements rather than incremental percentage gains.
• NVIDIA is building entire "AI Factories" rather than just data centers, designed to produce valuable tokens at incredible rates and cost-effectively.
• These AI factories are massive, highly integrated systems that scale from single racks to entire data centers.
• The GB200 NVL72, a rack-scale AI supercomputer, is highlighted as a testament to this extreme co-design, delivering significant performance gains and lower token generation costs.

"We call it extreme code design. Nvidia is the only company in the world today that literally starts from a blank sheet of paper and can think about new fundamental architecture, computer architecture, new chips, new systems, new software, new model architecture, and new applications all at the same time."

Re-Industrialization and AI's Role in Manufacturing [01:06:30]

• The transcript emphasizes America's return to manufacturing and re-industrialization, reignited by the age of AI.
• NVIDIA AI infrastructure systems, including Blackwell and future generations, are being built in America.
• The concept of the factory being born digital in Omniverse is explored, where digital twins are used for design, simulation, and operation of robotic factories.
• Partners like Foxconn, Siemens, and FANUC are collaborating to create state-of-the-art robotic facilities, leveraging AI for efficiency, safety, and worker onboarding.
• Physical AI, requiring three distinct computing systems (training, simulation, and operation), is key to advancing robotics and factories.

"America's return to making and reindustrialization, reignited by the age of AI."

The Inflection Point of Autonomous Vehicles and Robotics [01:36:12]

• Robots on wheels, such as robo-taxis, are at an inflection point, poised for widespread adoption.
• NVIDIA DRIVE Hyperion provides a standardized architecture for autonomous vehicle development, designed into vehicles from Lucid, Mercedes-Benz, and others.
• Partnerships with companies like Uber aim to connect these autonomous vehicles into global networks, creating a new computing platform for transportation.
• Humanoid robots are also discussed as a significant future consumer electronics and industrial equipment market, with companies like Figure and Agility making strides.
• Disney's collaboration on a new simulation platform for robotic learning highlights the advancements in creating physically aware and based AI for robots.

"The age of US re-industrialization is here with people and robots working together."

The Convergence of AI, Industry, and Innovation [01:39:39]

• The core message centers on two simultaneous platform transitions: general-purpose to accelerated computing, and classical software to AI.
• This convergence is driving unprecedented growth, with NVIDIA providing platforms for 6G (Arc), robotics (Hyperion), AI factories (DSX), and advanced manufacturing (Mega).
• The commitment to manufacturing in America is reiterated, signifying a new chapter of domestic innovation and production.
• The transcript concludes by celebrating the collaborative spirit and shared vision driving these technological advancements, underscoring the concept of "friendship and business rolling as one."

"Two platform transitioning happening at the same time which is the reason why we're feeling such incredible growth."