Slide 1: NVIDIA
GA NVIDIA
Slide 2: AI Scales Beyond LLMs
AI Scales Beyond LLMs “Everything Everywhere All at Once”
Slide 3: Open Models Reach the Frontier
Open Models Reach the Frontier we Kim ke Deeoseck v3.2 cptoss wen Nemotron Bade 80% of Startups Building on Open Models 1-in-4 OpenRouter Tokens Generated by Open Models Reasoning Models,
Slide 4: NVIDIA Leads Open Model Ecosystem
NVIDIA Leads Open Model Ecosystem Language Models Released in 2025 NVIDIA Google Alibaba Qwen IBM Granite AI2 Microsoft Mistral Z.ai DeepSeek Kimi Minimax Meta Llama OpenAI 0 25 50 75 100 Source: Hugging Face Count excludes quantized models, but includes variants (e.g., Nemotron reasoning vs. instru
Slide 5: NVIDIA Tops Leaderboards
NVIDIA Tops Leaderboards Artificial Analysis Artificial Analysis VLM3D OpenASR Intelligence Index Intelligence per Token Physical Reasoning OCRBench Retrieval Physical AI MVPBench, IntPHys, CasualVQA ViDoRe, MTEB, MMTEB PAIBench
Slide 6: Agents Are Multi-Model, Multi-Cloud, and Hybrid-Cloud
Agents Are Multi-Model, Multi-Cloud, and Hybrid-Cloud Prompt Response User or Machine Reason Critique Plan Tool Use App, Data Web Search Semantic Filesystem Memory Sandbox Platforms Search (MCP)
Slide 7: Agents Are Multi-Model, Multi-Cloud, and Hybrid-Cloud
Agents Are Multi-Model, Multi-Cloud, and Hybrid-Cloud Prompt Response User or Machine Reason Frontier Models Critique Plan Router Customized Open Models Tool Use App, Data Web Search Semantic Filesystem Memory Sandbox Platforms Search (MCP)
Slide 8: Slide 8
Slide 9: Enterprise Platforms Adopt NVIDIA AI
Enterprise Platforms Adopt NVIDIA AI Edge Video IT and HR Workflow Accelerated Data Analytics Automation Science and ML
Slide 10: Enterprise Platforms Adopt NVIDIA AI
Enterprise Platforms Adopt NVIDIA AI Code Real-Time Retrieval Augmented Review Security Analyst Generation
Slide 11: NVIDIA Full-Stack Physical AI Platform
NVIDIA Full-Stack Physical AI Platform
Slide 12: Compute is Data
Compute is Data Cosmos
Slide 13: Slide 13
Slide 14: Announcing NVIDIA Alpamayo
Announcing NVIDIA Alpamayo Open Reasoning VLA for Autonomous Vehicles Ego-Motion Driving History Alpamayo Decision Multi-Camera Causal Video Reasoning Optional Trajectory User Commands
Slide 15: Announcing NVIDIA Alpamayo
Announcing NVIDIA Alpamayo Open Reasoning VLA for Autonomous Vehicles “Parked cars mean the ego needs to be cautious of sudden movements” Alpamayo “Ball rolling could be a hazard-maybe a child or pet following it” “Slow down and prepare to stop if the ball or pedestrians pose a risk”
Slide 16: Slide 16
Slide 17: NVIDIA Ships Full-Stack AV on 2025 Mercedes Benz CLA
NVIDIA Ships Full-Stack AV on 2025 Mercedes Benz CLA Alpamayo Policy and Safety Evaluator Classical AV Stack Halos Safety OS
Slide 18: Global L4 and Robotaxi Ecosystem Building on NVIDIA
Global L4 and Robotaxi Ecosystem Building on NVIDIA AV Software OEMs and Mobility Platforms Tier 1 and Hardware Partners
Slide 19: Slide 19
Slide 20: Slide 20
Slide 21: Slide 21
Slide 22: Cadence Optimization AI and Agentic AI
Knowledge Sources Optimization AI Digital Custom Verification PCB/Package System 3rd Party Cadence Cerebrus® Virtuoso® Studio Verisium Allegro® X AI Optimality Agentic AI Silicon Agent Systems Agent LLM Vendor Cadence JedAI Platform Customer Electronic Design Automation System Design Automation Scie
Slide 23: Synopsys and NVIDIA Omniverse
Electronic Design Automation Computer Aided Engineering Electronics, Design Manufacturing Verification Structures Fluids Optics Electromagnetics PrimeLib Proteus PrimeSim Mechanical Fluent HFSS SPEOS Redhawk-SC S-Litho VCS LS-Dyna FreeFlow Icepak Zemax Totem SC S-Device Discovery Rocky Maxwell Lumer
Slide 24: Siemens Industries Infrastructure
Industries Infrastructure Chip Design Design / Engineering Operations Teamcenter Fuse Solido Industrial AI Suite BuildingX Digital Reality Viewer Calibre Tessent Designcenter Industrial Edge Gridscale X Totally Integrated Questa One Veloce Simcenter Electrification X Automation (TIA) Industrial Cybe
Slide 25: = _———
= _———
Slide 26: NVIDIA Leads Open Model Ecosystem
NVIDIA Leads Open Model Ecosystem Language Models Released in 2025 NVIDIA Google Alibaba Qwen IBM Granite AI2 Microsoft Mistral Z.ai DeepSeek Kimi Minimax Meta Llama OpenAI 0 25 50 75 100 Source: Hugging Face Count excludes quantized models, but includes variants (e.g., Nemotron reasoning vs. instru
Slide 27: Vera Rubin
Rotational P ‘ Velocity * Measured - + * Calculated 50,000 ‘ ‘ ‘ Distance From Center ‘ (Light Years) * s “we ‘ *
Slide 28: Insane Demand for AI Computing
Insane Demand for AI Computing Model Size Growing Test-Time Scaling “Thinking” Token Cost 10X Parameters Per Year 5X Tokens Per Year 10X Cheaper Per Year Kimi-K2 1T $100.00 1,000 100,000 Qwen3-480B GPT-4 Qwen-72B Reasoning GPT-3.5 Mixtral 8x22B GPT-4 Turbo 100 $10.00 GPT-4o Llama-2 70B Llama-13B 10,
Slide 29: Slide 29
Slide 30: NVIDIA Vera CPU
NVIDIA Vera CPU 88 NVIDIA Custom Olympus Cores 176 Threads with NVIDIA Spatial Multi-Threading 1.8 TB/s NVLink-C2C 1.5 TB System Memory (3X Grace) 1.2 TB/s LPDDR5X 227 Billion Transistors
Slide 31: NVIDIA Rubin GPU
NVIDIA Rubin GPU NVFP4 Inference 50 PFLOPS 5X Blackwell NVFP4 Training 35 PFLOPS 3.5X HBM4 Bandwidth 22 TB/s 2.8X NVLink Bandwidth per GPU 3.6 TB/s 2X Transistors 336 Billion 1.6X
Slide 32: NVIDIA ConnectX-9 Spectrum-X SuperNIC
NVIDIA ConnectX-9 Spectrum-X SuperNIC 800 Gb/s Ethernet with 200G PAM4 SerDes Programmable RDMA and Data Path Accelerator State-of-the-Art Security Line-Speed Encryption | Security Enclave and Attestation CNSA and FIPS Certified 23 Billion Transistors
Slide 33: NVIDIA BlueField-4
NVIDIA BlueField-4 800G Gb/s DPU for SmartNIC and Storage Processor 64 Core Grace CPU with ConnectX-9 2X Networking 6X Compute vs BF3 3X Memory BW 126 Billion Transistors
Slide 34: NVIDIA NVLink 6 Switch
NVIDIA NVLink 6 Switch Scale-Up Fabric with 3.6 TB/s Per-GPU All-to-All BW 400G SerDes In-Network SHARP Collectives 108 Billion Transistors
Slide 35: NVIDIA Vera Rubin NVL72
NVIDIA Vera Rubin NVL72 NVFP4 Inference 3.6 EFLOPS 5X Blackwell NVFP4 Training 2.5 EFLOPS 3.5X LPDDR5X Capacity 54 TB 3X HBM Capacity 20.7 TB 1.5X HBM4 Bandwidth 1.6 PB/s 2.8X Scale-Up Bandwidth 260 TB/s 2X Transistors 220 Trillion 1.7X
Slide 36: NVIDIA Spectrum-X Ethernet Co-Packaged Optics
NVIDIA Spectrum-X Ethernet Co-Packaged Optics 102.4 Tb/s Scale-Out Switch Infrastructure Co-Packaged 200G Silicon Photonics 128 Ports of 800 Gb/s 512 Ports of 200 Gb/s 352 Billion Transistors
Slide 37: Context is the New Bottleneck – Storage Must be Rearchitected
Context is the New Bottleneck – Storage Must be Rearchitected Model Size Growing Context Length Growing HBM Multi-Turn Conversations - Context Accumulates More Concurrent Users and Sessions Memory Rack SSD Network SSD
Slide 38: NVIDIA Context Memory Storage Platform
NVIDIA Context Memory Storage Platform A New POD-Level Context Memory Platform BlueField-4 Data and Storage Processor Spectrum-X Ethernet RDMA Infrastructure 5X Higher Tokens per Second 5X Higher Power Efficiency
Slide 39: Vera Rubin: Six New Chips – One Giant Leap to the Next Frontier
Vera Rubin Six New Chips – One Giant Leap to the Next Frontier
Slide 40: Six New Chips – One Giant Leap to the Next Frontier
Six New Chips – One Giant Leap to the Next Frontier Time to Train Factory Throughput Token Cost 1/4th Fewer GPUs Up to 10X More Tokens 1/10th Lower Cost DeepSeek++ Kimi K2-Thinking (32K/8K) Kimi K2-Thinking (32K/8K) $10.00 5M $8.00 4M $6.00 3M $4.00 2M $2.00 Throughput 1M 100T Tokens in 1 Month 0 $0
Slide 41: NVIDIA One Platform for Every AI
NVIDIA One Platform for Every AI Vera Rubin Isaac Open Models Alpamayo Agentic AI
Slide 42: NVIDIA
GA NVIDIA