NVIDIA Video Summaries

We demonstrate how an on-device local agent powered by NVIDIA RTX Spark can help developers debug and fix issues in an open-source badminton scheduling website. The demo covers the entire workflow, from monitoring communication channels to automatically running QA tests and merging fixes.

■ Local agent workflow
- Monitoring communication channels and prioritizing issues
- Investigating and fixing a booking failure

■ Hardware capabilities
- 128 GB unified memory running a local Quant 3.6 model and speech models

■ Delegating QA
- Rebuilding the site and using a computer-use agent to test the UI
- Merging the fix automatically

■ Broader impact
- Offloading repetitive tasks to free developers for feature work

This is for developers interested in local AI agents and hardware-accelerated workflows. Viewers will learn how RTX Spark can handle codebase-wide context, speech recognition, and automated QA, and how to apply similar agent-based workflows to their own projects.

We present the latest Vera Rubin platform updates, focusing on measured performance gains and production readiness. The update covers the Vera CPU's efficiency, the scale-up NVLink fabric switch's advantage, and power efficiency improvements demonstrated on the CoreWeave platform.

■ Platform Performance Updates
- Vera CPU: 1.8x performance per loaded core versus traditional x86-based systems
- Scale-up NVLink fabric switch: 2x performance compared to off-the-shelf Ethernet
- Power efficiency: 10x measured performance per watt versus Blackwell on CoreWeave

■ Production Readiness
- Real silicon measurements on CoreWeave, not projections or estimates
- Full production status with ecosystem readiness to bring the platform to market

This update is intended for HPC and AI infrastructure professionals seeking verified performance data. Viewers will gain a clear understanding of the Vera Rubin platform's measured advantages and its deployment readiness. The content does not include detailed architectural specifications or pricing. To evaluate the platform, consider reviewing the measured results and engaging with the NVIDIA ecosystem.

At NVIDIA GTC, we present how physical AI is transforming healthcare, building on decades of accelerated computing and AI innovation. The talk covers the three foundational pillars—world models, robotic policy models, and physics simulation—and how they enable surgical robotics, hospital automation, and medical imaging.

■ Physical AI foundations for healthcare
- World foundation models (Cosmos), robotic policies (Groot H), and physics simulation
- Open Age dataset: 750+ hours, 35+ partners, 11 embodiments

■ Isaac for Healthcare platform
- Sensor simulation libraries for ultrasound and X-ray
- Synthetic data pipelines, digital twins, and post-training recipes

■ Deployment and medical imaging models
- Holoscan 4.0 and IGX Thor for edge AI deployment
- Raw-to-insights models for ultrasound and MRI, NV Generate for CT/MR

This talk is intended for healthcare technology developers, roboticists, and medical device researchers. Viewers will learn how NVIDIA's open models, simulation tools, and deployment platforms can be applied to accelerate physical AI development in healthcare.

This session from GTC examines how physical AI and foundation models are transforming healthcare robotics, from surgical training and autonomous procedures to hospital automation. It covers the core challenges of data curation, simulation, and model deployment, with presentations from NVIDIA and leading medical robotics companies.

■ Opening: Physical AI in Healthcare
- Data flywheel: curation, synthetic data, simulation, training
- Three shifts: generative physics, reasoning foundation models, accelerated simulation

■ Surgical Data and Simulation
- Filipo (Northwell Health) on surgical video annotation, ontologies, and Cosmos-based simulation
- Axel (Johns Hopkins) on imitation learning, autonomous suturing, and the Open H dataset

■ Industry Systems
- Moon Surgical's Maestro: collaborative robot, scope pilot, and simulation for arm deployment
- LEM Surgical: upper-torso humanoid architecture for hard-tissue surgery

■ Future Directions
- Path to supervised autonomy, safety architecture, and collaboration with NVIDIA tools

This session is valuable for engineers, researchers, and clinicians seeking an overview of current applications and research directions in healthcare robotics, including data generation, simulation, and autonomous functions.

AI has become one of the most rapidly adopted technologies, and its growth is driving unprecedented demand for computing power. A keynote announcement by NVIDIA, alongside Firebird, outlines plans to build national AI infrastructure in Armenia and Kazakhstan. It explains why every country must develop its own AI capabilities rather than relying solely on imported intelligence.

■ AI's impact and adoption
- AI is transforming healthcare, finance, manufacturing, and science
- Every query and reasoning step generates compute demand

■ AI as new infrastructure
- AI will become as essential as electricity and the internet
- Countries need domestic AI capacity, not just imported services

■ Announcement: 250 MW AI infrastructure
- Firebird and NVIDIA to deploy 250 MW across Armenia and Kazakhstan
- Infrastructure will serve researchers, startups, industries, and governments

■ Expected outcomes
- Attract global companies and catalyze local startups
- AI factories convert energy into intelligence

Policymakers, business leaders, and technology professionals will gain insight into the scale and rationale behind sovereign AI infrastructure projects and the role of compute in the AI era. Viewers will understand the expected impact on local researchers, startups, and the broader AI economy.

Infosys presents its AI strategy and the value of its partnership with NVIDIA. The presentation covers two main pillars—AI Augmented Services and AI First Services—and explains how NVIDIA's AI stack improves performance, governance, and cost efficiency.

■ AI Strategy Pillars
- AI Augmented Services: applying AI to existing offerings
- AI First Services: helping customers transform with AI

■ NVIDIA Collaboration
- AI Center of Excellence established in 2021
- Use of NeMo Curator, Guardrails, Retriever, and NiM Microservices

■ Ongoing Initiatives
- Integration with Nemotron-3 and industry-specific models
- Physical AI, AI factories, and sovereign AI cloud

This overview is suited for technology leaders and professionals interested in enterprise AI adoption. Viewers will gain insight into practical AI implementation approaches and the potential of strategic AI partnerships.

We explore the system of models approach to building AI agents, where frontier and open models are combined to balance reasoning depth, speed, cost, and data control. Using Glean's Waldo model as a concrete example, we demonstrate how routing tasks between specialized and frontier models improves efficiency.

■ The System of Models Approach
- Frontier models for state-of-the-art reasoning
- Open models for fine-tuning, on-premises deployment, and lower latency
■ Glean's Waldo Model in Action
- Post-trained on NVIDIA NeMo Triton 3 Nano
- Gathers context from support tickets, Slack, and survey data
- Routes simple questions to one-shot models and complex tasks to frontier models
■ Measurable Results
- 10x faster enterprise context search
- 50% lower latency and 25% fewer tokens without quality loss

Designed for AI engineers, developers, and enterprise leaders seeking to optimize agent performance, cost, and data security. Viewers will learn how a system of models can make AI agents smarter, faster, and more efficient.