Slide 1: NVIDIA
GI NVIDIA
Slide 2: Taiwan ecosystem partner logos
Accton acer ¥ ADATX A ADLINK AD\ANTECH ‘aetinoa AIRECO QIIXON @apmic APPRO.PHO peed 7sus A\atceNomix ALAN TAVC} AVerMedia Oa rxomrex BizLink wree <a CHENBRO- @ tx2@® CLEVO COMPAL COLORFUL a) deel Aeuta SWIM ET csuvoins P eaanen Bae verFocus aban DY rexcenn aac Prso wn GIGABYTE en tnventoc iStaging m
Slide 3: NVIDIA Transformation: From GPU Chip to AI Infrastructure
NVIDIA Transformation From GPU Chip to AI Infrastructure
Slide 4: Slide 4
Slide 5: Accelerated Computing Begins With CUDA-X Libraries
Accelerated Computing Begins With CUDA-X Libraries Parabricks Aerial cuPyNumeric cuEquivariance cuQuantum Earth-2 MONAI Megatron Sionna cuDF cuTensor CUDA-Q Genomics Numerical Warp Dynamo Weather Medical cuOpt cuML Analytics Imaging 5G/6G Computing Quantum cuDSS Physics NIXL Quantum Signal Processin
Slide 6: NVIDIA Building 6G on AI
NVIDIA Building 6G on AI NVIDIA Grace CPU NVIDIA PCIe GPU Aerial NVIDIA ConnectX-7 Sionna Compact AI-RAN 5G/6G Signal Processing
Slide 7: NVIDIA Building Hybrid Quantum-GPU Computing
NVIDIA Building Hybrid Quantum-GPU Computing cuQuantum CUDA-Q Quantum Computing GB200 NVL72
Slide 8: Agentic AI: 100-1000X More Compute Than One-Shot
Agentic AI “Understand-Think-Act” is 100-1000X More Compute Than One-Shot 你是我在台北的行程規劃師。 18:30 –21:30 的行程，捷運加步行，造訪台北五大夜市中的至少三個，品 稱、菜品、捷運路線與站名、步行距離、路線說明和每個資訊的 50 Reasoning Okay, let's tackle this. The user needs three search queries to help plan the sections of their Taiwan night market food tour repo
Slide 9: Physical AI: The Foundation of Robotic Revolution
Physical AI The Foundation of Robotic Revolution 內 Scenario 1 A shadow cast by a tree on the right briefly obscures a pedestrian stepping out from between packed cars, causing the ego driver to slow further.
Slide 10: Physical AI: The Foundation of Robotic Revolution
Physical AI The Foundation of Robotic Revolution 內 Scenario 2 A car is stopped on a quiet suburban road, with a clear blue sky overhead and lush green trees lining the streets. A pigeon enters the frame from the right, flying low over the street with graceful, fluid movements. It gracefully glides p
Slide 11: Physical AI: The Foundation of Robotic Revolution
Physical AI The Foundation of Robotic Revolution 內 Scenario 3 The car is stopping on the road. Dogs barking and darting into the street from a nearby yard.
Slide 12: General Robotics Opens the Next Trillion-Dollar Industry
General Robotics Opens the Next Trillion-Dollar Industry NVIDIA 屜 在開始之前，解釋你將採取的每個步驟。 Reasoning Okay, let's break down the problem. The task is to pick up the mug on the table. First, I need to confirm that the "mug" shown is the same as the "water bottle" shown in the video. The user might be using
Slide 13: Grace Blackwell NVL72 "A Thinking Machine"
GB300 NVL72 Grace Blackwell NVL72 “A Thinking Machine”
Slide 14: Grace Blackwell Ultra Superchip
Grace Blackwell Ultra Superchip 40 PFLOPS TE FP4 1.5X GB200 576 GB at 16 TB/s HBM3e 1.5X GB200 800 Gb/s CX8 2X GB200
Slide 15: Grace Blackwell NVL72 "A Thinking Machine"
GB300 NVL72 Grace Blackwell NVL72 “A Thinking Machine”
Slide 16: CoreWeave 220 MW USA, Scaling Globally
CoreWeave 220 MW USA, Scaling Globally Courtesy of CoreWeave
Slide 17: Oracle Cloud Supercluster 260 MW Indonesia
Oracle Cloud - Supercluster 260 MW Indonesia, Scaling Globally Courtesy of Oracle
Slide 18: xAI Colossus 150 MW Memphis, Tennessee
xAI - Colossus 150 MW Memphis, Tennessee Courtesy of xAI
Slide 19: Stargate 200 MW, Scaling to 1.2 GW
Stargate 200 MW, Scaling to 1.2 GW Abilene, Texas Courtesy of Crusoe
Slide 20: SSeS oes
SSeS oes peeeee
Slide 21: Foxconn, Taiwan, TSMC and NVIDIA Build AI Infrastructure for Taiwan
Announcing Foxconn, Taiwan, TSMC and NVIDIA Build AI Infrastructure for Taiwan
Slide 22: Grace Blackwell NVL72 "One Giant GPU"
Grace Blackwell NVL72 “One Giant GPU”
Slide 23: Announcing NVLink Fusion: AI Factory Plus Custom ASIC
Announcing NVLink Fusion NVIDIA AI Factory Plus Custom ASIC Custom ASIC with NVLink Fusion IP
Slide 24: Announcing NVLink Fusion: AI Factory Plus Custom CPU
Announcing NVLink Fusion NVIDIA AI Factory Plus Custom CPU Custom CPU with NVLink Fusion IP
Slide 25: NVIDIA NVLink Fusion
NVIDIA NVLink Fusion Semi-Custom AI Infrastructure NVIDIA Plus ASIC Hybrid AI Infrastructure Proven Scale-Up and Scale-Out Ecosystem Single Scalable Hardware Architecture
Slide 26: One Architecture from Cloud AI to Edge AI
One Architecture – from Cloud AI, Enterprise AI, Personal AI, to Edge AI Arm Platforms x86 Platforms AI Storage DGX Server RTX PRO B200/B300 GB200 / GB300 RTX PRO Server DGX NVL72 RTX PRO Workstation DGX Station Laptop Spark
Slide 27: Slide 27
Slide 28: Announcing DGX Station
Announcing DGX Station 1 Trillion Parameter Models Personal AI Machine 20 PFLOPS AI Performance GB300 Grace Blackwell Ultra Superchip 800 GB Unified System Memory Up to 1T Parameter Models ConnectX-8 SuperNIC 800 Gb/s Network Connection
Slide 29: NVIDIA Enterprise AI Platform Accelerates Trillion-Dollar Global IT Investment
NVIDIA Enterprise AI Platform Accelerates Trillion-Dollar Global IT Investment Agentic AI $1T AIOPS Enterprise IT and Industrial Automation Software Industry NEMO OMNIVERSE AI Infrastructure CUDA-X AI-Q DOCA COMPUTE STORAGE NETWORKING
Slide 30: Slide 30
Slide 31: Announcing NVIDIA RTX PRO Server
Announcing NVIDIA RTX PRO Server Enterprise AI and Industrial AI Computing Platform 30 PFLOPS FP4 AI Performance 8 x RTX PRO 6000 Blackwell Server Edition GPUs 3 PFLOPS RTX Graphics Performance Real-Time Omniverse Digital Twin Simulation 800 GB GPU Memory 13 TB/s Bandwidth ConnectX-8 SuperNIC Switch
Slide 32: Announcing NVIDIA RTX PRO Server
Announcing NVIDIA RTX PRO Server Enterprise AI and Industrial AI Computing Platform Llama 70B 8K ISL / 256 OSL
Slide 33: Announcing NVIDIA RTX PRO Server
Announcing NVIDIA RTX PRO Server Enterprise AI and Industrial AI Computing Platform DeepSeek R1 128K ISL / 4K OSL
Slide 34: Announcing NVIDIA RTX PRO Server: Available Now
Announcing NVIDIA RTX PRO Server Enterprise AI and Industrial AI Computing Platform Available Now
Slide 35: NVIDIA AI Data Platform
NVIDIA AI Data Platform Compute Prompt Critique Plan RTX PRO 6000 Blackwell Server Edition NVIDIA BlueField NVIDIA AI-Q Blueprint Response NVIDIA Llama Nemotron Reason Tool Use Networking NVIDIA Spectrum-X Calculator Web Semantic SQL Generate Search Query Query Summary Storage NVIDIA BlueField
Slide 36: NVIDIA Top Leaderboard Open Models and Datasets
NVIDIA Top Leaderboard Open Models and Datasets #1 15X Faster 50% Greater Faster Data Extraction Open Reasoning Models Higher Retrieval Accuracy NeMo Retriever Llama Nemotron NeMo Retriever
Slide 37: NeMo Retriever Agent Workflow
ie) rd User or Machine Prompt Vv DataEngine Pian Reflect NVIDIA NeMo Retriever NVIDIA NeMo Retriever NVIDIA NeMo Retriever NVIDIANIM, © . all (3 Extraction Embedding Reranking NVIDIA Llama Nemotron Report / Actions Reason Refine j ® Images Documents Videos Chat Email Data Catalog EOW Internet VJ oat
Slide 38: Build plan for top sales opportunities in APAC
Bul plan forthe top sles opportunites in APAC. Muttimodal Data Retrieval Bull Action Plan Reply to your agent @ seen (a Reson Tiling recent Create Action Plan + step Prompt: Key ules that apply provide data-driven, actionable guidance fr new APAC sales reps use evidence rom actual deals and testimon
Slide 39: Dell AI Data Platform With NVIDIA
Client Applications DELL Al Data Platform With NVIDIA. Agent Identity & Access: Management & B D@LTechnologies Data Platform Services 3 3 g Vector Data Dataset interface & Access Control Storage Access Control Federated SQL Engine A Data Pipelines Data Processing eS & Identity Management Engine Resp
Slide 40: Hitachi Vantara iQ Studio Agent
iQ Studio Agent User Response With Cites Source Link Data Load = versiones Enterprise ala VSP One NVIDIA NeMo_ Object & File Retriever Extraction NVIDIA NeMo Retriever Embedding © Hitachi Vantara iQ Studio Agent Runtime omar ee _. exam = eS mes el ‘a.stdio ueston Question & Context Time Agent NVIDIA
Slide 41: Enterprise Data Retrieval Pipeline
PDF NVIDIA Neto Retriever oan \ veers oe ng a . a= T™ iid. [api] - ; $ & ] i tt g ee . “ne _ \ NVIDIA Llam: S3INFS 12M Storage Katka CAS Extract Embed Vectorize Index Vector Vectorized CAS Search NERS Seale purine patabase —“Revrank csi ie : ! : cowat secs Data Control Por . | atc . (@ ————— NVIDIA
Slide 42: NetApp AI Data Compute Nodes
oO Data Compute Nodes Oh B B A Storage Compute Nodes t Sync & Detect Catalog, Data NVIDIA NeMo User Governance Retriever Reranking, Updates Q & | NetApp Al Agent Ai Agent Al Agent NVIDIA NeMo Vector DB Extract Classify and Retriever Embedding Updates Anonymize NVIDIA Llama Traceability and Nemotron
Slide 43: NVIDIA Enterprise AI Platform Accelerates Trillion-Dollar Global IT Investment
NVIDIA Enterprise AI Platform Accelerates Trillion-Dollar Global IT Investment Agentic AI Test and Data Curate Build Search Monitor Guardrail Secure Evaluate AIOPS Orchestrate NEMO OMNIVERSE AI Infrastructure CUDA-X AI-Q DOCA COMPUTE STORAGE NETWORKING
Slide 44: AI-Powered Falcon Next-Gen SIEM
‘Al-Powered Falcon Next-Gen SIEM Data Data » Data Ingestion Processing Analytics NVIDIA Morpheus Training Synthetic gretel Data ° NVIDIA NeMo + NVIDIA Llama Nemotron Reason Threat Detection Documents LA) ‘Threat Hunt Agent Vulnerability Agent Charlotte Al Detection Triage Agent Threat Rule Detection
Slide 45: Dataiku Platform Data Flywheel
User Dataiku Platform @ dataiku Data Flywheel Aue) Virtual Experimentation Application Dataiku Dita UM Guard ras Pert NVIDIA LUM NIM Information Retrieval NVIDIANeMo Curator curated > Davsets > HTL NUIDIANeMo Customizer | NVIDIA NeMo Evaluator NVIDIA NeWo Retriever Reranking NVIDIA NeMo Retriever Em
Slide 46: DataRobot Agentic Blueprint
Data Sources Inventory Forecast Data SS CCTV Cameras Realtime Video Enterprise Documents Agentic Blueprint Vector RAG Third-party LLM NIM aa Agent Runtime ; oo o+ Graph Context RAG Manager [eh e : ey CA Knowledge cA Engineering NVIDIA LLM NIM Ingestion Workflow = DataRobot Workbench (Develop) oy i)
Slide 47: DataStax Unified Knowledge Layer
DATASTA)(¢ NoSQL Vector-enabled Database A 2 >) Astra DB HCD OpenSearch (Cloud) (On-prem) Z% Langflow Low-Code Agent and Al IDE and Runtime > - App & 3" Parly API Services DATASTA)< NVIDIA Base Command RAG Workflow NVIDIANeMo. NVIDIANIM Retriever NVIDIA NVIDIA NeMo Blueprints Guardrails Enterprise M
Slide 48: Elastic Ingestion and Retrieval
Ingestion Retrieval &f — NVIDIA NeMo Guardrails Query NVIDIA NeMo Guardrails Lim Multimodal Enterprise NVIDIANeMo Retriever Reranking NVIDIANeMo Retriever Embedding NVIDIA NeMo NVIDIA NeMo Retriever Data Retriever Extraction Embedding yo GD astabase 4 NVIDIA cuvS Elasticsearch Vector @ fo} Search (C
Slide 49: Nutanix Enterprise AI
NUTAND<S OY Nutanix QZ Enterprise Al P oO Role Based Day2 Enterpr 4A Peres Controls Operations LLM Repository Al User Inference Endpoint Service NVIDIA NeMo NVIDIA NeMo NVIDIA NeMo NVIDIA Llama Nemotron Guardrails Retriever Reranking Retriever Embedding, Reason . RAG Inference Guardrail Embedding Re
Slide 50: Red Hat OpenShift AI
RedHat OpenShift Al ‘Agentic App Development Cc RedHat OpenShift fal Metrics Telemetry NVIDIA GPU Operator eo Enterprise Kubernetes, Enterprise Linux NVIDIA NIM vuM Optimized Profiles ing & Observal Accelerator Profiles [NVIDIA Network Operator OE Container & VM Managemert sGlang Logging Prompts and
Slide 51: Trend Vision One
Trend yn One™ Response (XDR) —) ran exten pera hj. = Posture Management cS oS NVIDIA RAPIDS Processing. 1 1 ‘cuDF i I NVIDIA Morpheus Data Pipeline Vulnerability Management Training —— Ax Events Alerts NVIDIA NIM Investigation Agent Detection Alerts NVIDIANIM Classification Agent g g Threat nolzence
Slide 52: Weights & Biases by CoreWeave
ii! Weights & Biases by CoreWeave W8B Guardrails User Front End Pian NVIDIA NeMo Plies Guardrails Feesinack Q. root W8B Feedback Inference Workflow Online Processing NVIDIANeMo Retriever Enterprise Data NVIDIA NIM NIM Logs Offline Processing Automated Dataset Curation i Human Customer Feedback Label
Slide 53: NVIDIA Enterprise AI Platform Accelerates Trillion-Dollar Global IT Investment
NVIDIA Enterprise AI Platform Accelerates Trillion-Dollar Global IT Investment Agentic AI Test and Data Curate Build Search Monitor Guardrail Secure Evaluate AIOPS Orchestrate NEMO OMNIVERSE AI Infrastructure CUDA-X AI-Q DOCA COMPUTE STORAGE NETWORKING
Slide 54: Newton: Disney Research, NVIDIA, Google DeepMind
‘Dienep Research GYNVIDIA J Google DeepMind Newton Fe |< Fe MuJoCo Disney Research Custom Solver Solver Solver MuJJoCo Playground a ro | Other Simulator NVIDIA Warp Isaac Lab
Slide 55: NVIDIA Isaac GR00T
NVIDIA Isaac GR00T Open Humanoid Robot Development Platform Sensor Input Action Prompt NVIDIA Isaac Jetson Thor
Slide 56: Announcing NVIDIA Isaac GR00T N1.5
Announcing NVIDIA Isaac GR00T N1.5 Open Humanoid Robot Foundation Model Sensor Input Action Prompt 6K 3.6K Model Downloads GitHub Stars
Slide 57: Announcing NVIDIA Isaac GR00T-Dreams
Announcing NVIDIA Isaac GR00T-Dreams A Small Human Demonstration Team Can Do the Work of Thousands NVIDIA Action Cosmos Labeling Pick up the object on the table that is used for drinking water
Slide 58: Slide 58
Slide 59: Omniverse Digital Twin
Omniverse Digital Twin re cel
Slide 60: Omniverse Digital Twin
Omniverse Digital Twin
Slide 61: Omniverse Digital Twin
Omniverse Digital Twin
Slide 62: Omniverse Digital Twin
pein VV : TB ? | ras (8 AI i A oe | ME { *\ WY ] Omniverse Digital Twin est
Slide 63: Omniverse Digital Twin
GIGABYTE Omniverse Digital Twin
Slide 64: Omniverse Digital Twin
Omniverse Digital Twin
Slide 65: Omniverse Digital Twin
al | Omniverse Digital Twin |
Slide 66: Omniverse Digital Twin
Omniverse Digital Twin
Slide 67: Omniverse Digital Twin
cadence @ Crusoe Avneuta tan sdddNd (lvaaati hi | RA um THAHANNAIHUUIN Omniverse Digital Twin
Slide 68: Kaohsiung City Omniverse Digital Twin
Kaohsiung City Omniverse Digital Twin
Slide 69: Slide 69
Slide 70: Slide 70
Slide 71: NVIDIA AI Factories, Agents, Robots | Amazing Partners | One Architecture
NVIDIA AI Factories, Agents, Robots | Amazing Partners | One Architecture
Slide 72: NVIDIA
GI NVIDIA