NVIDIA GTC Washington, D.C. Keynote with CEO Jensen Huang : Speech Outline


 

The speech video: NVIDIA GTC Washington DC Keynote with CEO Jensen Huang

1. Opening & Welcome

  • Welcome to GTC Washington, D.C.

  • Thanks to sponsors and ecosystem partners.

  • “The Super Bowl of AI” and its pre-show presentation.

2. The Birth of a New Computing Model

  • The first new computing model in 60 years.

  • The end of Moore’s Law and Dennard Scaling.

  • The origin of accelerated computing: GPU + CUDA.

  • A turning point has arrived — the era of accelerated computing begins.

3. Challenges in Accelerated Computing

  • Fundamental differences in programming models.

  • Algorithm restructuring and library development.

  • Cross-generation compatibility: CUDA 13 → CUDA 14.

  • Application rewriting and domain expansion.

4. CUDA X Libraries: The Foundation of Innovation

  • cuLitho – Computational lithography (TSMC / Synopsys / ASML).

  • cuOpt – Numerical optimization (e.g., traveling salesman problem).

  • Warp – Python-based physics simulation engine.

  • cuDF – Accelerated data frame and SQL processing.

  • cuDNN & Megatron Core – Large-scale language model support.

  • MONAI – AI framework for medical imaging.

  • Aerial – Wireless communication acceleration.

  • cuQuantum – Quantum computing acceleration.

  • Over 350 libraries, each redesigned to unlock new markets.

5. The Beauty of Simulation

  • Fully simulated demonstrations (no CGI or artistic animation).

  • Industry coverage: healthcare, manufacturing, robotics, autonomous vehicles, graphics, and gaming.

  • From Virtua Fighter (1993) to today’s breakthroughs.

6. 6G and the Transformation of Telecommunications

  • Telecommunications as a pillar of national security and the economy.

  • The U.S. regains technological leadership.

  • Launch of NVIDIA ARC (Aerial Radio Network Computer).

  • Three core technologies: Grace CPU, Blackwell GPU, ConnectX networking.

  • Software-defined, programmable wireless communication + AI processing.

  • Strategic partnership with Nokia (compatible with AirScale base stations).

  • AI for RAN: boosts spectrum efficiency, reducing global electricity use by 1.5–2%.

  • AI on RAN: supports edge industrial robots and cloud computing.

7. Breakthroughs in Quantum Computing

  • Inspired by Richard Feynman’s 1981 vision.

  • Major milestone: logical qubits with error correction stability.

  • Quantum fragility and the need for robust correction.

  • Launch of NVQ Link – ultra-high-bandwidth QPU↔GPU interconnect.

  • CUDA Q: a hybrid quantum-GPU computing platform.

  • Ecosystem support: 17 quantum companies + 8 U.S. DOE national labs.

  • Collaboration with the U.S. Department of Energy: 7 AI supercomputers.

8. AI Beyond Chatbots

  • AI ≠ just chatbots.

  • AI is reshaping the computing stack — from hand-coded software to trained models.

  • New infrastructure: Energy → GPU → Data Center → Token Production.

  • Tokenization extends to language, images, video, 3D, chemistry, genes, and motion.

  • AI model diversity: CNNs, state-space models, GNNs, multimodal networks, etc.

  • AI as a worker, not just a tool (examples: Perplexity, Cursor, robotaxi).

  • Economic impact: entering a multi-trillion-dollar economy, easing labor shortages.

9. The AI Factory Revolution

  • An AI Factory ≠ traditional data center.

  • It produces “intelligent tokens.”

  • Three stages of AI learning:

    1. Pre-training

    2. Post-training – inference, coding, mathematical reasoning

    3. Thinking – continuous learning and reasoning

  • The virtuous cycle: smarter models → more users → willingness to pay → more compute → even smarter models.

10. Extreme Co-Design: The Solution

  • With Moore’s Law ending, exponential performance gains are needed.

  • NVLink 72: rack-scale “mega GPU” delivering 10× performance (vs. H200).

  • Lowest token generation cost.

  • Spectrum X Ethernet: AI-optimized Ethernet for data center scalability.

  • Scale-Across architecture: connects data centers at XGS terabit scale.

11. Blackwell & U.S. Manufacturing

  • Grace Blackwell NVLink 72: the first rack-scale AI supercomputer made in the USA.

  • Full U.S. production pipeline:

    • Arizona (wafer fab)

    • Indiana (HBM memory)

    • Texas (assembly)

    • California (networking)

  • Projected $500 billion revenue visibility (Blackwell + early Rubin, before 2026).

  • Expected shipment: 20 million Blackwell GPUs — 5× the Hopper generation.

12. Rubin Architecture Launch

  • Third-generation NVLink 72 rack-scale system.

  • Vera Rubin Superchip: 100 petaflops (equivalent to 100 DGX-1s).

  • Cable-free and 100% liquid-cooled.

  • Context processors with massive KV cache support.

  • BlueField-4 DPU: memory and context acceleration.

  • Spectrum X + Quantum Switch: multi-protocol, fabric expansion.

13. The Future of AI Factories: Omniverse DSx

  • Digital twin blueprint: from design to operations.

  • Partners: Jacobs, Siemens, Bechtel, Vertiv, and others.

  • 1 GW-scale AI factory optimization: billions in annual revenue uplift.

  • Digital twins as the new “operating system” of industrial AI.

14. The Importance of Open-Source Models

  • Explosion in open-source model capabilities: inference, multimodal, distillation.

  • NVIDIA leads open AI: 23 top-ranking models.

  • Open models are the lifeblood of startups, research, and industry.

15. Building the AI Startup Ecosystem

  • NVIDIA’s platform advantages: cloud, on-premises, and enthusiast PCs.

  • Emerging GPU cloud providers: CoreWeave, Lambda, Crusoe, etc.

  • Major cloud integrations: AWS, Google Cloud, Azure, Oracle.

  • Enterprise SaaS transformation: ServiceNow, SAP, Synopsys, Cadence.

16. New Strategic Partnerships

  • CrowdStrike: AI-powered cybersecurity (cloud + edge).

  • Palantir Ontology: real-time data processing and business intelligence.

17. Physical AI: The Three Computers

  • Training computer: Grace Blackwell NVLink 72.

  • Simulation computer: Omniverse (digital twins).

  • Inference computer: Jetson Thor (robots and autonomous driving).

18. The Future of Manufacturing (Robot Factories)

  • Foxconn Houston factory: fully digital twin design.

  • Partners: Siemens, FANUC, FII, Vision AI.

  • AI agents for monitoring, safety, and quality control.

  • Caterpillar’s century-old manufacturing digitally transformed.

19. The Robotics Revolution

  • Figure: humanoid robots (valuation ~$40B).

  • Agility Robotics: warehouse automation.

  • Johnson & Johnson: surgical robots.

  • Disney Blue: adorable robots based on Newton simulation.

20. Autonomous Driving & Robotaxi

  • Launch of NVIDIA DRIVE Hyperion platform.

  • Standard sensor suite: cameras, radar, LiDAR.

  • Supported by automakers: Lucid, Mercedes-Benz, Stellantis, etc.

  • Global partnership with Uber for robotaxi deployment.

  • Market potential: 1 trillion miles per year, converting 50 million taxis into robotaxis.

21. Summary: Platform Shifts & New Technologies

  • Two major platform shifts:

    1. General computing → Accelerated computing (CUDA X)

    2. Hand-coded software → Artificial intelligence

  • The new NVIDIA platform pillars:

    • 6G: ARC

    • Robotaxi: Hyperion

    • AI Factory: DSx

    • Robot Factory: MEGA

  • Return of U.S. manufacturing leadership.

     

    P.S.: When upgrading your AI training or inference systems, consider selling your used GPU graphics cards to recover value and reduce business costs.

     

     

Comments

Post a Comment

Popular posts from this blog

Unlock Top Value for Your High-End GPU: Strategies for Selling Fast and Secure

Image
High-end graphics processing units have evolved from niche enthusiast components into strategic assets . In today’s market, a single flagship GPU can represent thousands of dollars in recoverable value, whether it originated from a workstation upgrade, a research cluster refresh, or surplus inventory tied to changing AI workloads. At the same time, the secondary GPU market has become increasingly fragmented. Pricing can shift quickly, demand varies by buyer segment, and liquidation channels differ dramatically in terms of speed, risk, and net return.  Selling a high-end GPU  is no longer a matter of listing it and waiting; it requires understanding where value is realized, how risk is managed, and when speed outweighs price. This article examines the primary channels available for liquidating high-end GPUs and evaluates them through a practical, decision-oriented lens. Rather than promoting a single method, the goal is to provide a clear framework that helps sellers align...
Read more

Everything You Need to Know Before Installing More RAM in Your PC

Image
Upgrading your computer’s memory is still one of the simplest and most satisfying ways to breathe new life into your PC. Whether you’re editing 4K videos, building AI models, or just tired of Chrome eating all your RAM, adding more memory can make a huge difference. At  BuySellRam.com , we’ve seen thousands of systems come through our doors — from aging DDR3 desktops to brand-new DDR5 workstations. And no matter the year, memory upgrades remain one of the most cost-effective improvements you can make. In this post, we will illustrate how to upgrade your PC memory (RAM). But before you start shopping, here are a few key things to know about upgrading RAM in 2025. Why Upgrading RAM Still Matters Modern CPUs and GPUs are lightning-fast, but they can’t shine without enough memory to feed them data efficiently. RAM acts as the bridge between your processor and storage — and when it’s full, your computer starts swapping data to slower drives, causing those frustrating slowdowns. So wheth...
Read more