Slide 1: NVIDIA
GI NVIDIA
Slide 2: Partner and customer logos
TON! asroce accenture § @T AFL CAtamai Qs amdocs SANpURIL Arg ABUS Gaver catolio aS BLUE ORIGIN >BNY Qecene Booz Allen BostonDynamics 2] BOYD oy Brookhaven cadence C&g0v carahsoft. CATERPILLAR cow) Pelt MS — Cognition CoolkT ~— CW CoreWeave ~—- \EROWDSTRIKE Crusoe : Ostaiku DataRobot = Qddn-—(DeALTe
Slide 3: CPU General Purpose to GPU Accelerated Computing Platform Shift
CPU General Purpose to GPU Accelerated Computing Platform Shift Microprocessor Trend Data Dennard Scaling Stops 10 7 # of Transistors GPU 10 6 10 5 Single-Thread Performance (SpecINT) 10 4 Frequency 10 3 Power 10 2 CPU 10 1 10 0 1970 1980 1990 2000 2010 2020
Slide 4: Accelerated Computing Begins with CUDA-X Libraries
Accelerated Computing Begins with CUDA-X Libraries
Slide 5: Slide 5
Slide 6: AI 6G Quantum Models Enterprise Robotics Factories
AI 6G Quantum Models Enterprise Robotics Factories
Slide 7: Slide 7
Slide 8: NVIDIA and Nokia Pioneer the AI Era of Telecommunications
Announcing NVIDIA and Nokia Pioneer the AI Era of Telecommunications
Slide 9: Nokia to Build AI-Native 6G on New NVIDIA ARC
Announcing Nokia to Build AI-Native 6G on New NVIDIA ARC “Aerial RAN Computer” Nokia MIMO Radio Aerial RAN Computer
Slide 10: NVQLink Launches Quantum-GPU Computing
NVQLink Launches Quantum-GPU Computing
Slide 11: Slide 11
Slide 12: NVIDIA NVQLink Quantum-GPU Interconnect
Announcing NVIDIA NVQLink ― Quantum-GPU Interconnect NVQLink NVLink QPU GPU CPU NVQLink CUDA-Q | cuQuantum | CUDA Oak Ridge National Laboratory and the oak leaf symbol are registered trademarks of the U.S. Department of Energy. Use of this mark does not constitute or imply its endorsement, recommend
Slide 13: Department of Energy Partners with NVIDIA to Build 7 New AI Supercomputers
Announcing Department of Energy Partners with NVIDIA to Build 7 New AI Supercomputers Argonne National Laboratory Solstice – 100K Blackwell GPUs Equinox Tara Minerva Janus Los Alamos National Laboratory Powered by Vera Rubin Mission Vision
Slide 14: AI - The New Industrial Revolution
AI - THE NEW INDUSTRIAL REVOLUTION APPLICATIONS : MODELS - | Le INFRASTRUCTURE GPUS ENERGY —|
Slide 15: AI - The New Industrial Revolution
AI - THE NEW INDUSTRIAL REVOLUTION APPLICATIONS 7 LLM VLM VIA vMLM ){ GPT | MODELS \ [ Jt Jt 4 : Jt — L J (om j(_ enn ]{ woe }( ss } Cem} INFRASTRUCTURE , ENERGY
Slide 16: AI - The New Industrial Revolution
AI - THE NEW INDUSTRIAL REVOLUTION sas ££ &€& CHATBOTS DIGITAL BIOLOGY ROBOTAX] ENTERPRISE Al AGENTS APPLICATIONS 8 Y é2 ; } SCIENCE ROBOTICS MANUFACTURING Al CODER : LLM VLM wa }{ vai ){ cet] MODELS \ C Jt ae) Ged) ee L J (om }(_ewn {woe )( ssw] (tem ] INFRASTRUCTURE , ENERGY
Slide 17: AI - The New Industrial Revolution
AI - THE NEW INDUSTRIAL REVOLUTION = £§ & B& IN me CHATBOTS DIGITAL BIOLOGY ROBOTAXI ENTERPRISE. Al AGENTS APPLICATIONS e Pad ee —— SCIENCE ROBOTICS MANUFACTURING Al CODER LLM vim |{ VLA VIMLLM ——- SSS (om }(_ewn_ J {- woe }C sem MC = INFRASTRUCTURE J AI FACTORIES > ) | | ) ENERGY | vA
Slide 18: Three Scaling Laws
THREE SCALING LAWS EXPONENTIAL GROWTH IN COMPUTE, ADOPTION AND USAGE INTELLIGENCE TEST-TIME SCALING \ Ut “LONG THINKING’ | | ost SCALING > THREE SCALING LAWS ADOPTION INTELLIGENCE y { & a COMPUTE COMPUTE / LF PRE-TRAINING SCALING } Ne / 7 COMPUTE PROFITS
Slide 19: Co-Design
CO-DESIGN EXTREME CO-DESIGN — BEYOND MOORE'S LAW SCALE-OUT | SCALE-UP iz
Slide 20: Grace Blackwell NVL72 "A Thinking Machine"
Grace Blackwell NVL72 “A Thinking Machine”
Slide 21: Extreme Co-Designed Blackwell NVL72 for Mixture-of-Experts Models
Extreme Co-Designed Blackwell NVL72 Game-Changer for Mixture-of-Experts Models 32 Experts 32 Experts 32 Experts 32 Experts 4 Experts 4 Experts 4 Experts 4 Experts Ethernet NVLink All-to-All Switch 32 Experts 32 Experts 32 Experts 32 Experts 4 Experts 4 Experts 4 Experts 4 Experts 1,800 GB/s GPU-to-G
Slide 22: Token Throughput per Megawatt
Extreme Co-Designed Blackwell NVL72 Game-Changer for Mixture-of-Experts Models 10X Token Throughput per MW vs. Interactivity DeepSeek R1 0528 · FP4 · 8K/1K · Source: SemiAnalysis InferenceMAXTM Perf-per-Dollar 6,000,000 InferenceMAX by 10X Token Throughput per Megawatt GB200 NVL72 4,500,000 Perf-per
Slide 23: Cost per Million Tokens
Extreme Co-Designed Blackwell NVL72 Game-Changer for Mixture-of-Experts Models 10X Cost per Million Tokens vs. Interactivity DeepSeek R1 0528 · FP4 · 8K/1K · Source: SemiAnalysis InferenceMAXTM Perf-per-Dollar $2.80 InferenceMAX by 10X $2.10 Cost per Million Tokens Perf-per-Watt H200 NVL8 ($/M Token
Slide 24: Cloud capex spend (Morgan Stanley Research)
$632 $549 $443 $270 $126 $133 2022 2023 2024 2025 2026 2027 Source: Morgan Stanley Research Amazon, CoreWeave, Google, Meta, Microsoft, Oracle
Slide 25: CSP Capex Spend
CSP CAPEX SPEND ACCELERATED COMPUTING GENERAL PURPOSE COMPUTING 2022 2023 2024 2025 2026 2027
Slide 26: Exceptionally Strong Demand for Grace Blackwell NVL72
$0.5T+ Exceptionally Strong Demand for 20M GPUs Grace Blackwell NVL72 Driven by Two Exponentials Visibility into more than half a trillion dollars in cumulative Blackwell and Rubin revenue through CY2026 (so far) More than 5X Hopper’s lifetime revenue Shipped 6M GPUs $100B 4M GPUs Blackwell and Rubi
Slide 27: Slide 27
Slide 28: NVIDIA Extreme Co-Design Delivers X-Factors on One-Year Rhythm
NVIDIA Extreme Co-Design Delivers X-Factors on One-Year Rhythm Full-Stack | One Architecture | CUDA Everywhere Rubin Feynman Volta Blackwell Rubin Rubin CPX Rubin Ultra Feynman Blackwell Blackwell Ultra Next–Gen HBM V100 8S HBM3e 8S HBM3e 8S HBM4 16S HBM4e HBM2 Grace Vera Vera CPU CPU CPU BlueField-
Slide 29: Vera Rubin Superchip
Vera Rubin Superchip Processor for Gigascale AI Factories 100 PF AI with 88 Custom Arm Cores 2 TB Fast Memory 6 Trillion Transistors
Slide 30: Vera Rubin Compute Tray
Vera Rubin Compute Tray 200 PF NVL72 Building Block 14.4 TB/s NVLink | 800 GB/s ConnectX-9 100% Liquid-Cooled 12 Trillion Transistors
Slide 31: Vera Rubin CPX Compute Tray
Vera Rubin CPX Compute Tray 1M+ Token Context Accelerator 440 PF Inference Performance 8 Additional Rubin CPX GPUs Converged and CPX-Only Configurations
Slide 32: Announcing NVIDIA BlueField-4
Announcing NVIDIA BlueField-4 800G SmartNIC for AI Factories 64 Core Grace CPU with ConnectX-9 AI Data Storage Acceleration 126 Billion Transistors
Slide 33: NVLink Switch Tray
NVLink Switch Tray Scale-Up Fabric with 3.6 TB/s Per-GPU All-to-All BW NVLink6 400G Custom SerDes 28.8 TB/s Total Bandwidth 432 Billion Transistors
Slide 34: Spectrum-X Ethernet
Spectrum-X Ethernet 102.4 TB/s Scale-Out Switch Infrastructure Integrated 200G Silicon Photonics 95% Effective Bandwidth at Scale 352 Billion Transistors
Slide 35: Slide 35
Slide 36: NVIDIA Omniverse DSX Blueprint for Gigascale AI Factories
Announcing NVIDIA Omniverse DSX Blueprint for Gigascale AI Factories RTX PRO Server Omniverse | DSX | CUDA-X
Slide 37: Developer Adoption of Open Models
Developer Adoption of Open Models 160M 140M 120M Qwen 100M Llama Mistral Total Monthly 80M Gemma Downloads DeepSeek Phi 60M gpt-oss 40M 20M 0M October 2023 April 2024 October 2024 April 2025 October 2025
Slide 38: NVIDIA Open Models, Data, Libraries Top Leaderboards
NVIDIA Open Models, Data, Libraries Top Leaderboards NVIDIA and Top OSS Models 23 Models on Leaderboards Cosmos GR00T Physical AI Robotics Nemotron Clara Agentic AI Biomedical AI LiveCodeBench AIME 2025 ViDoRe, LLM Safety OpenASR PAIBench MVPBench, VLM3D MTEB, MMTEB Leaderboard IntPHys, CasualVQA
Slide 39: AI Startups Build on NVIDIA
Al Startups Build on NVIDIA ABRIDGE Forest Labs © Coderabbit 8 Cognition @ cursor |[Figver Kexa § FieldAl + FI S URE Sz Fireworks Al Generalist Harvey [111 [somorphic | Labs | Lica ; , NSCALE : nuro Openkvidence) Of perplexity Phgsical Ineigence x oP Ree 3 |. fy ® poolside O@ Reflection runway Q sKi
Slide 40: AWS and NVIDIA
aws @INVIDIA —— CUSTOMERS Al built for your 7 i AMGEN (1)! databricks @® QPalantir ~~ peplexty —servicenow. $i snowtlake biggest ideas SERVICES 777? Envon ws ANNOUNCED TODAY Deepest MIA Blueprints NDIA Nelo ‘Amazon Bedrock fosters bc taeolay pee hee ones Raney ate Painigiaives A OA ch Win ormnere aE
Slide 41: Google Cloud and NVIDIA
Google Cloud | GYNVIDIA Deep engineering collaboration powers every level of the Al stack lo Al-Optimized hardware Cutting-edge performance, planet-scale deployment, and robust security. G4VMs Powered by NVIDIA RTX PRO 6000 ‘A4X Max VMs Powered by NVIDIA GB300 NVL72 ea Leading software, 1 ii open fr
Slide 42: Microsoft Azure and NVIDIA
© amdocs E Microsoft Azure | @nvipoia Ecosystem Partnership on Azure Epic @esri servicenow. NVIDIA Dynamo Inference Server NVIDIA Blackwell Ultra NVIDIA Blackwell Accelerated Azure Services b Microsoft Bing ta ‘Azure Al Foundry NVIDIA libraries integration throughout Azure g & © NVIDIA NVIDIA AL NVI
Slide 43: Oracle and NVIDIA New Announcements
New Announcements: Zettascale Supercluster + 800,000 NVIDIA GPUs + 16 ZettaFLOPS of Al Oracle Database 26ai + Using NVIDIA NeMo Retriever and NVIDIA NIM to build RAG pipelines Oracle Al Data Platform + NVIDIA RAPIDS for ‘Apache Spark + Faster analytics, ETL, and machine learning ORACLE | GINVIDIA QA
Slide 44: ServiceNow and NVIDIA
servicenow @INVIDIA ServiceNow Al Platform (getom 7 aE ‘Autonomous, Adaptive, Goal-DrivenInteligence Digital Services is crm ir HR Finance & Sup Chain Tools ue Worktiows Case Triage Agent: Where is my stuff? Seeurty Governance APIS Spokes Al Control Tower Guardian Idontity & ACLS hme. dasii ite Cont
Slide 45: SAP and NVIDIA
@INVIDIA ‘SAP Application 9 Joule Customer —~- Partner Apps Cloud ERP crm Spend x ak ‘Al Foundation - Al Operating System Development 4° Generative Al Hub Serving/inference Engine Tooling Trust Orchestration am Harmonized API ta NVIDIA NeMo NVIDIA NvioIA Playground Masking Prompt Registry Retriever
Slide 46: Synopsys and NVIDIA
synopsys" @INVIDIA sation 1 Genal Services (* Synopsys AgentEngineer” Technology 6 © Vertietion Digital ‘alg Test a @ a 3 weTeoine e@ a (ca) (cay (ca) 3 Toolkit F Nemotron Reasoning 5 pain 2 5) sameae Fon Cpt i ca ston Comy Ves, erd, Ve Formal, “ Primesim, 2 ive PrimeTime, TestMAx 5 ‘Ve Spyciess J c
Slide 47: Cadence and NVIDIA
MCP API's Cadence Solutions Digital Custom cr Verification PCB/3D-IC Multiphysics Data Center Blology cadence @INVIDIA Cadence* JedAl Platform Off-Prem Models Compute Data inference Pye @ CadeN Ce omcieus Refinement Embedding evn Vendor LLM Prony Server paaense Security Prompt a1 ‘On-Prem RAG index
Slide 48: CrowdStrike and NVIDIA
\GROWDSTRI KE @INVIDIA Endpoints, Workioads Third-Party Edge Solutions CrowdStrike Security Cloud g_= ‘os «6 3 an r RES Bea, ee ee aa 1 Falcon Threat Agent Prioritization Analysis Analysis. anaes H Data Lake Detections Agent ‘Agent Agent 1 ! Hl } oy a) Sg | O@) ea 1 CrowdStrike NVIDIA NeMo ‘Syntheti
Slide 49: Palantir and NVIDIA
QPalantir | GJnvipia ‘02 AT+HUMANTEAMS {ff} GE Witinata-) (E GomSance) (GE Common) © (Gentes -) @ Dowtma-) @ Are ANALYTICS & WORKFLOWS AUTOMATIONS PRODUCTS & SDKs ‘SYSTEMS OF ACTION MODELS
Slide 50: Palantir and NVIDIA
QPalantir | GJnvipia 92 AL-+HUMANTEAMS {if GE Mainsaa) (6 Opoitouce) (6 Commarea) + (@ Oowaios) @dowten) Grass) ANALYTICS & WORKFLOWS AUTOMATIONS PRODUCTS & SDKs BE Anvow ss \ oPyTo ‘SYSTEMS OF ACTION
Slide 51: NVIDIA and Global Leaders Build Digital Twin Platform for US Reindustrialization
NVIDIA and Global Leaders Build Digital Twin Platform for US Reindustrialization
Slide 52: NVIDIA and Global Leaders Build Digital Twin Platform for US Reindustrialization
NVIDIA and Global Leaders Build Digital Twin Platform for US Reindustrialization
Slide 53: Slide 53
Slide 54: Foxconn Builds New Blackwell Factory in Texas
Announcing Foxconn Builds New Blackwell Factory in Texas RTX PRO Server Omniverse | Isaac Sim | Cosmos | Metropolis
Slide 55: Caterpillar Builds Factory Digital Twin on NVIDIA
Caterpillar Builds Factory Digital Twin on NVIDIA RTX PRO Server Omniverse | Mega
Slide 56: Figure Builds Humanoid Robots on NVIDIA
Figure Builds Humanoid Robots on NVIDIA Jetson Thor GB200 RTX PRO Server Omniverse | Isaac Lab | Cosmos
Slide 57: Agility Robotics Builds Humanoid Robots on NVIDIA
Agility Robotics Builds Humanoid Robots on NVIDIA GB200 RTX PRO Server Jetson Thor Omniverse | Isaac Sim | Isaac Lab
Slide 58: Johnson & Johnson Builds Medical Robots on NVIDIA
Johnson & Johnson Builds Medical Robots on NVIDIA RTX PRO Server Omniverse | Isaac for Healthcare | Cosmos IGX Thor
Slide 59: NVIDIA and Disney Research Advance Robot Learning
NVIDIA and Disney Research Advance Robot Learning GB200 RTX PRO Server Omniverse | Isaac Sim | Newton Jetson AGX
Slide 60: Slide 60
Slide 61: Slide 61
Slide 62: Global Automakers Adopt NVIDIA DRIVE Hyperion
Announcing Global Automakers Adopt NVIDIA DRIVE Hyperion DRIVE AGX ECU
Slide 63: Slide 63
Slide 64: Global Robotaxi Ecosystem Built on NVIDIA DRIVE Hyperion
Announcing Global Robotaxi Ecosystem Built on NVIDIA DRIVE Hyperion
Slide 65: Blackwell NVL72
Al Blackwell NVL72 Quantum NV@Link @ oy ENERGY LLY Open Models Nemotron, Cosmos, GROOT, Clare BB russing race Made in America Omniverse i cexcon: Enterprise CUDA-X, Nemotron QPalantir Robotics DRIVE Hyperion Uber 6G NVIDIA ARC NO<IA. ‘y
Slide 66: NVIDIA
GI NVIDIA