Slide 1: NVIDIA
GI NVIDIA
Slide 2: Slide 2
Slide 3: Slide 3
Slide 4: Partner and ecosystem logos
4x ao Amroboris QJ oxeee —Accion ccucrnzy” acer A ADL Adobe | AEE, ADWNTECH geting AFA RI SAsiity ABLE AIC Alytis AIM AIRES Gesu wwe C2AlbabeDbus —Alsemy — SVAnwances ANTHROP\C = @APMIC arm = Ainwis. rsneck — fA\Asterolabs ABLES fee ws Pimmom = GMM BEYONDA: QD Balink AE GD IntByedore @capasu cadence
Slide 5: Useful AI Has Arrived
Useful AI Has Arrived Source: GitHub, April 2026
Slide 6: Agent = LLM + Harness
AGENT = LLM + HARNESS CONTEXT OBSERVE __) REASON
Slide 7: Agents Go to Work
Agents Go to Work Make a JavaScript Newton's Cradle animation in a single HTML file.
Slide 8: Agents Go to Work
Agents Go to Work Create a gif of NVIDIA-green dots on black scatter, form Taipei 101 building, morph to “GTC TAIPEI 2026,” morph to the NVIDIA eye logo, then scatter and repeat.
Slide 9: Agents Go to Work
Agents Go to Work I lost my remote control battery clip, it looks like this. Create a CAD file ready for 3D printing to create a new one.
Slide 10: Slide 10
Slide 11: Prompt Orchestration
PROMPT ORCHESTRATION ae cPU AGENT = LLM + HARNESS CONTEXT REASON =) ACT | | MEMORY Qoewnut
Slide 12: Full-Stack AI Infrastructure Company
Full-Stack Al Infrastructure Company
Slide 13: NVIDIA DSX AI Factory Platform
NVIDIA DSX AI Factory Platform Power Optimization Infrastructure Software Platform Software DSX OS DSX Sim DSX MaxLPS DSX Flex Libraries and Frameworks Reference Designs
Slide 14: Slide 14
Slide 15: NVIDIA DSX AI Factory Ecosystem
NVIDIA DSX AI Factory Ecosystem AI Clouds AI Factory Software Design and Construction Energy and Cooling Infrastructure and Facilities Compute Systems
Slide 16: NVIDIA AI Platform
W&B W&B W&B W&B CoreWeave CoreWeave CoreWeave CoreWeave CoreWeave CoreWeave Models Training Inference Weave Inference Slurm on Kubernetes Service Compute Storage Mission Control Kubernetes NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB
Slide 17: NVIDIA AI Platform
Multi-Tenant AI Cloud Nebius Agentic Search NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB, Hopper Server Edition Grace Blackwell Computing Vera Rubin and BlueField Networking Spectrum-X Networking
Slide 18: Training and Inference
Training and Inference NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB, Hopper Server Edition Grace Blackwell Computing Vera Rubin and BlueField Networking Spectrum-X Networking
Slide 19: NAVER World Model and Cloud AI Factory
NAVER World Model NAVER Cloud AI Factory NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB, Hopper Server Edition Grace Blackwell Computing Vera Rubin and BlueField Networking Spectrum-X Networking
Slide 20: NVIDIA AI Platform
NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB, Hopper Server Edition Grace Blackwell Computing Vera Rubin and BlueField Networking Spectrum-X Networking
Slide 21: Training and Inference
Training and Inference NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB, Hopper Server Edition Grace Blackwell Computing Vera Rubin and BlueField Networking Spectrum-X Networking
Slide 22: Training and Inference
Training and Inference NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Confidential NVIDIA NVIDIA ConnectX NVIDIA Quantum IB, Hopper Server Edition Grace Blackwell Computing Vera Rubin and BlueField Networking Spectrum-X Networking
Slide 23: System Builders and OEMs
GMI GMI Model GMI Prime GMI Dedicated Agent Execution Agent Dedicated Agent Distribution GMI GMI Hybrid GMI Kubernetes Model Garden Fine Tuning Inference Endpoints Infra Isolation Ecosystem Compute Clusters Control System Builders and OEMs NVIDIA AI Platform NVIDIA NVIDIA RTX PRO Blackwell NVIDIA Co
Slide 24: Compute Is Revenue
COMPUTE IS REVENUE EXTREME CO-DESIGN — TTFT, TOKEN/WATT SCALE & EXPERIENCE — TTFT, MTBI CUDA ECOSYSTEM & INSTALLED BASE — USEFUL LIFE AI FACTORY REVENUE
Slide 25: Compute Is Revenue
COMPUTE IS REVENUE EXTREME CO-DESIGN — TTFT, TOKEN/WATT SCALE & EXPERIENCE — TTFT, MTBI CUDA ECOSYSTEM & INSTALLED BASE — USEFUL LIFE AI FACTORY REVENUE ATTFT ¢+----
Slide 26: Compute Is Revenue
COMPUTE IS REVENUE EXTREME CO-DESIGN — TTFT, TOKEN/WATT SCALE & EXPERIENCE — TTFT, MTBI CUDA ECOSYSTEM & INSTALLED BASE — USEFUL LIFE AI FACTORY REVENUE ZS TOKENS/WATT SS Sess ATTFT ‘¢----
Slide 27: Compute Is Revenue
COMPUTE IS REVENUE EXTREME CO-DESIGN — TTFT, TOKEN/WATT SCALE & EXPERIENCE — TTFT, MTBI CUDA ECOSYSTEM & INSTALLED BASE — USEFUL LIFE AI FACTORY ~ REVENUE + i} i i i i] H ZS TOKENS/WATT i | l ]
Slide 28: Compute Is Revenue
COMPUTE IS REVENUE EXTREME CO-DESIGN — TTFT, TOKEN/WATT SCALE & EXPERIENCE — TTFT, MTBI CUDA ECOSYSTEM & INSTALLED BASE — USEFUL LIFE Al FACTORY REVENUE a
Slide 29: Compute Is Revenue
COMPUTE IS REVENUE EXTREME CO-DESIGN — TTFT, TOKEN/WATT SCALE & EXPERIENCE — TTFT, MTBI CUDA ECOSYSTEM & INSTALLED BASE — USEFUL LIFE Al FACTORY REVENUE
Slide 30: Slide 30
Slide 31: Vera Rubin in Full Production
Vera Rubin in Full Production ODMs and OEMs
Slide 32: NVIDIA Vera – CPU for the Age of AI
NVIDIA Vera – CPU for the Age of AI Grace BlueField Vera Rubin Vera Vera BlueField GPU Compute Storage
Slide 33: NVIDIA Vera
NVIDIA Vera CPU for Agents NVIDIA-Custom Olympus Core 1.2 TB/s LPDDR5X ECC 88-Core / 176-Threads 40% Lower Loaded Latency Spatial Multithreading (vs x86) 10-Wide Instruction Fetch/Decode 3.4 TB/s Core-to-Core BW 2MB L2 per Core / 164MB L3 1.4 TB/s PCIe Gen6 250W - 450W TDP 1.8 TB/s NVLink C2C
Slide 34: NVIDIA Vera
NVIDIA Vera 3X Bandwidth per Core vs. x86 3X 2X 1X 1X Bandwidth Bandwidth per Core x86 CPU with DDR5 NVIDIA Vera with LPDDR5X
Slide 35: NVIDIA Vera
NVIDIA Vera 1.8X Agentic Sandbox Performance vs. x86 1.9X 1.7X 1X 1X Compilation Python x86 CPU NVIDIA Vera
Slide 36: Slide 36
Slide 37: NVIDIA Vera is the most formidable competition to Intel and AMD
“NVIDIA Vera is the most formidable competition to Intel and AMD x86_64 processors ever realized.” Michael Larabel Author, Phoronix CPU Benchmark Suite
Slide 38: Vera CPU 3X Faster SQL Data Processing
Vera CPU 3X Faster SQL Data Processing “With Starburst on NVIDIA Vera CPUs with Dell, agents can instantly query governed enterprise data at scale, unlocking the ground truth to power decisions at the speed the agentic era requires." ― Justin Borgman, CEO and Founder of Starburst Dell AI Data Platfo
Slide 39: Vera CPU 6X Faster Real-Time Streaming Data Processing
Vera CPU 6X Faster Real-Time Streaming Data Processing “In collaboration with Redpanda and HPE, using NVIDIA Vera CPUs, we will be scaling our capacity, while further optimizing latency to power a high-performance, resilient and AI-ready market infrastructure.” ― Lynn Martin, President of NYSE Group
Slide 40: NVIDIA Vera Opens Large Market
NVIDIA Vera Opens Large Market CPU for Agents Early Adopters Cloud Ecosystem Partners
Slide 41: Agent = LLM + Harness
AGENT = LLM + HARNESS CONTEXT __) OBSERVE =) REASON __) ACT
Slide 42: Announcing NVIDIA Agent Toolkit for Enterprise AI
Announcing NVIDIA Agent Toolkit for Enterprise AI NVIDIA AI Platform
Slide 43: Announcing Cadence Chip Design Super Agent
Announcing Cadence Chip Design Super Agent Inputs Outputs RTL RTL Coverage Metrics Proof Core Specs COI Coverage Functional Coverage Codex Agent Code Coverage ChipStack AI Super Agent Planning and Orchestration Unit Test Agent Simulate Chip Behavior Mental Model Agent Mental Model Bug Reports, RTL A
Slide 44: u_dut.core®.pipe.decode.mux.cycle[15:6] a BEA :
u_dut.core®.pipe.decode.mux.cycle[15:6] a BEA : 7 top.core2.lid.data.pll_locked 8 th.dut.gpe[2].mon.inn[7:6) 41 l top.coreB.irg.parity[63:6] B7EQ9ACAAI706800 | th. dut.way.clk_div2 1 u_dut.warp[26] .mux.user [63:6] CFeD446ABC2cBBe8 | u_dut.clk_gen.gpe[8].tpc[1].sm.warp[14] .ch[14].way.operand[15.. E
Slide 45: Announcing NVIDIA Nemotron 3 Ultra
Announcing NVIDIA Nemotron 3 Ultra Frontier Smart Nemotron 3 Ultra GLM 5.1 Kimi K2.6 Qwen3.5 (550B) (744B) (1T) (397B) Agent Productivity PinchBench 91% 84% 91% 89% Long-Horizon Planning EnterpriseOps-Gym 33% 40% 29% 30% Coding Terminal-Bench 2.0 54% 64% 67% 53% Instruction Following IFBench 82% 77%
Slide 46: Announcing NVIDIA Nemotron 3 Ultra
Announcing NVIDIA Nemotron 3 Ultra 5X Faster Intelligence Index 60 Kimi K2.6 GLM-5.1 Nemotron 3 Ultra 50 Most Qwen3.5 397B 40 Attractive Quadrant 30 20 50 100 150 200 250 300 350 Output Speed (Output Tokens per Second) Alibaba Kimi NVIDIA Z.ai
Slide 47: Announcing NVIDIA Nemotron 3 Ultra
Announcing NVIDIA Nemotron 3 Ultra 30% Lower Cost % Coding Tasks Completed 80 Nemotron 3 Ultra Kimi-K2.6 GLM-5.1 Qwen3.5 (1T) (744B) (397B) (550B) Cost Efficiency Frontier 60 40 20 $250 $500 $750 $1,000 $1,250 $1,500 $1,750 $2,000 Cost to Task Completion (US$) Alibaba Kimi NVIDIA Z.ai SWE-bench Veri
Slide 48: Enterprise Software Platforms Build Agents on NVIDIA
Enterprise Software Platforms Build Agents on NVIDIA RTL Engineer for EDA Detection Triage, Malware Analysis, Engineering and Scientific AI FDE Supply Chain Agent Threat Hunting Workflow Automation Expense Accounting, Supply Chain IT Services and Workflow Automation Autonomous Chip Design Autonomous
Slide 49: 40 Years of Personal Computing
40 Years of Personal Computing
Slide 50: NVIDIA and Microsoft Reinvent PC
NVIDIA and Microsoft Reinvent PC Local LLM DeepSeek Gemma GLM GPT-OSS Kimi MiniMax Nemotron Qwen
Slide 51: Slide 51
Slide 52: Announcing NVIDIA and Microsoft Reinvent PC
Announcing NVIDIA and Microsoft Reinvent PC Powered by RTX Spark Blackwell RTX GPU 1 Petaflop FP4 AI Performance 20 Core Grace CPU Custom Built with MediaTek 128 GB Unified Memory 600 GB/s NVLink C2C Full NVIDIA Stack CUDA | TensorRT | NVFP4 RTX Ray Tracing | DLSS | Reflex | G-SYNC
Slide 53: i
i
Slide 54: Accelerate Every AI. Every App. Every Game.
Accelerate Every AI. Every App. Every Game.
Slide 55: Announcing New Adobe Premiere and Photoshop for RTX Spark
Announcing New Adobe Premiere and Photoshop for RTX Spark Up to 2X Faster Creative Agent-Ready
Slide 56: Announcing NVIDIA RTX Spark Laptops
Announcing NVIDIA RTX Spark Laptops Smart. Powerful. Beautiful. Available Fall
Slide 57: Three Revolutionary Windows Machines
Three Revolutionary Windows Machines One Architecture. Agent-Ready.
Slide 58: A New Line. A New Beginning.
A New Line. A New Beginning. Feynman Rubin Blackwell Feynman HBM Next Rubin Blackwell Ultra 20 Petaflops HBM4 HBM3e Grace Grace Blackwell Spark LPDDR5x 1 Petaflop Vera Rosa Rosa Feynman LPDDR5x Vera Rubin Spark CPU LPDDR6 CPU Spark CX8 Grace Blackwell Spark 800G 400 Teraflops LPDDR5x CX9 Vera Rubin
Slide 59: Platform lineup
ISAAC GROOT SPACE-1 Ne ee, HYPERION Dee REReeeer VERA RUBIN \ Dex sTaTION FOR WINDOWS RIX SPARK _DGX SPARK GEFORCE RTX HOLOSCAN AERIAL ARC
Slide 60: Compute is Data
Compute is Data WORLD FOUNDATION MODEL ROBOT CAPABILITY Hl + INTERNET + 4 TELEOP COMPUTE AND DATA VOLUME
Slide 61: Announcing NVIDIA Cosmos 3 — The Frontier of Physical AI
Announcing NVIDIA Cosmos 3 — The Frontier of Physical AI Artificial Analysis Artificial Analysis PAI-Bench Physics-IQ Text to Image Image to Video World Generation Accuracy Physics Accuracy , R-Bench RoboArena RoboLab VANTAGE-Bench TAR Embodiment Video Generation Action Policy Score Manipulation Pol
Slide 62: Pick up the screwdriver and put it on the top rack
Pick up the screwdriver and put it on the top rack. Plan: 1. Lower the gripper to grasp the screwdriver. 2. Lift the screwdriver from the table. 3. Move the screwdriver horizontally toward the white rack. 4. Place the screwdriver on the designated shelf of the white rack.
Slide 63: Announcing AI Leaders Join NVIDIA Cosmos Coalition
Announcing AI Leaders Join NVIDIA Cosmos Coalition
Slide 64: NVIDIA DRIVE Hyperion Robotaxi Platform
NVIDIA DRIVE Hyperion Robotaxi Platform Announcing Alpamayo 2 Super Robotaxi Reasoning Model Alpa Simulators Alpamayo and OmniDreams Training Data Generation Open Models NVIDIA Halos OS Hyperion DRIVE AGX Thor
Slide 65: Global Automakers Build Robotaxis on NVIDIA DRIVE Hyperion
Global Automakers Build Robotaxis on NVIDIA DRIVE Hyperion
Slide 66: Slide 66
Slide 67: NVIDIA Isaac GR00T Humanoid Robot Platform
NVIDIA Isaac GR00T Humanoid Robot Platform Isaac Simulators GR00T Open Models Training Data Generation NVIDIA Isaac OS Jetson Thor
Slide 68: Announcing NVIDIA Isaac GR00T Reference Humanoid Robot
Announcing NVIDIA Isaac GR00T Reference Humanoid Robot Jetson Thor Onboard Compute 2070 FP4 TFLOPS | 14-Core Arm CPU | 128 GB memory 31-DOF Unitree H2 Plus Robot 22-DOF 5-Finger Sharpa Hands GR00T Software Stack Data Generation and Policy Validation Unitree Whole-Body Control | GR00T 1.7 Model Avail
Slide 69: Slide 69
Slide 70: Slide 70
Slide 71: NVIDIA Enterprise AI
SINvIDIA Enterprise Al — . iii iii i — .
Slide 72: Slide 72
Slide 73: NVIDIA
GI NVIDIA