Beyond CG and cryptos: NVIDIA, supporting scientific discovery, approaches the greatest ambition post-GenAI, humanoids.

During the keynote at GTC 2024 in San José, California, Jensen Huang, CEO of NVIDIA, made announcements reinforcing the company's leadership in key areas such as artificial intelligence (AI), high-performance computing, and robotics. The event started with a bang, showcasing impressive AI-generated animations in perfect sync with the auditorium lights, signalling how NVIDIA has evolved beyond its origins in graphics cards and cryptocurrency mining.

Huang introduced the Blackwell GPU, surpassing the already impressive H100 and H200 GPUs, positioning it as the most advanced chip for AI applications. This launch underscores NVIDIA's ambition to lead the generative AI revolution, promising exceptional performance for the data centre sector and marking a milestone in the industry that begins to define a new way to program computers.

The announcement of the availability of Blackwell, both independently and in combination with NVIDIA's Grace CPU to form the GB200 Superchip, promises performance up to 30 times superior to the Nvidia H100 GPU for large language model (LLM) workloads, with notably improved energy efficiency. This focus reflects a concern for sustainability and the energy consumption of its products.

The introduction of the DGX SuperPOD supercomputer system, which can scale to support a massive amount of GB200 Superchips and features an innovative liquid cooling system, demonstrates NVIDIA's commitment to efficiency and adaptability to customer needs.

Tech giants like Amazon, Google, Microsoft, and Oracle are already considering integrating Blackwell GPUs into their cloud platforms, evidencing the significant impact of NVIDIA on AI training and inference in the tech sector.

Towards the end of his presentation, Huang emphasized his leadership in robotics and AI with the announcement of Project GR00T, aimed at developing humanoid robots capable of understanding natural language and learning by observing. This project highlights NVIDIA's commitment to autonomous robotics and presents a vision where humanoids can learn like humans, leveraging the abundance of data from human activities.

NVIDIA's announcements reinforce its legacy in developing cutting-edge technologies. It anticipates a future transformed by AI, robotics, and accelerated computing, improving industries and daily life. The vision shared at GTC 2024 illustrates NVIDIA's central role in advancing toward new frontiers in AI and computing, setting the course for technological innovation.

The realm of robotics experienced a groundbreaking shift, a transformation that has been brewing for years. The anticipation reached its peak when NVIDIA's Jensen Huang recognized Gideon's pioneering efforts at #GTC21, hinting at the monumental impact of the convergence of AI and robotics.

From Vision to Reality: NVIDIA's Prophetic Insights

• Past Perspective: Jensen highlighted the immense potential of accelerated computing and AI at the data centre scale for enhancing simulation performance.
• Current Outlook: He now envisions a future where everything mobile will embody robotics, marking a significant evolution in technological integration.

NVIDIA's Robotics Milestones at GTC

NVIDIA Robotics unveiled several innovations at NVIDIA GTC, aiming to revolutionize AI robotics development with essential components and tools:

• Isaac Manipulator: Offers cutting-edge motion generation and AI capabilities for robotic arms.
• Isaac Perceptor: Introduces Visual AI for Autonomous Mobile Robots (AMRs), a crucial development for visionary AMR creators.
• GR00T: A versatile foundation model for humanoid robot learning.
• Jetson Thor-based Computer: A new computing solution for humanoid robots powered by the NVIDIA Thor SoC.
• Isaac Lab: A platform for advanced robot learning.
• Isaac OSMO: Facilitates hybrid-cloud workflow orchestration, enhancing development efficiency.

These developments push the boundaries of robotics technology and align with Gideon's vision over the past seven years. We've always believed in the power of advanced visual perception and AI to drive the future of flexible robotics.

Guiding Principles for Robotics Startups

For entrepreneurs aiming to make a mark in the robotics industry, the following guidelines are invaluable:

1. Customer First: Prioritize early customer feedback and add value from the outset. Forge lasting relationships and address their primary concerns.
2. Specialization: Concentrate on solving one specific issue at a time. Avoid creating universal solutions; instead, be the best at addressing a particular need.
3. Efficiency: Leverage readily available components to avoid unnecessary innovation.
4. Data Utilization: Harness industry-specific data and insights for development. This approach will serve as a competitive advantage and growth lever.
5. Integration and Cooperation: Ensure your robots can seamlessly communicate and interact with other systems.
6. Real-world Deployment: Acknowledge and prepare for the complexities of deploying AI robots in commercial settings. Invest in resources to manage these challenges effectively.
7. Strategic Partnerships: Collaborate with industry leaders to expedite your growth, like our partnership with Toyota Material Handling Europe.I was genuinely impressed by the technical level Huang spoke at during his conference. The event embraces the motto that NVIDIA is our gateway to the world of artificial intelligence, primarily aimed at developers and even scientists. From the beginning, with his characteristic sense of humour, Jensen Huang anticipated that the conference would be highly technical, making jokes about it.

This approach established a tone of camaraderie and openness and highlighted the depth and seriousness with which NVIDIA approaches technological advancement. The ability to balance complex technical details with moments of levity is undoubtedly a skill not many can wield as adeptly as Huang, arguably the most important CEO so far in the new spring of AI, which, without friendly hardware to make software, would be impossible.