Ambarella: Ambarella : Evercore ISI ADAS, AV and AI Summit Presentation

September 24, 2024 Evercore ISI ADAS, AV & AI Summit 55 East 52 nd Street, New York, NY Louis Gerhardy, VP Corporate Development AMBARELLA.COM COPYRIGHT AMBARELLA 2024 Ambarella Overview Advanced deep learning AI inference hardware and software Founded on the premise that video is a unique type of data requiring an optimized SoC architecture Santa Clara, CA Company HQ Detroit Munich Parma Shanghai Seoul VisLab Dayton Tokyo Hsinchu Shenzhen 2024 ~25M AI SoCs shipped cumulatively 2021 Acquisition of 2015 Acquisition of 2012 IPO, NASDAQ: AMBA 2004 Ambarella founded Ambarella's deep neural network AI inference processor integrated with a state- of- the-art video & radar perception yields highly optimized domain controller SoCs VisLab, a pioneer in CV and autonomous driving software, acquired in 2015 Oculii, a pioneer in high-definition radar perception technology, acquired in 2021 World-class engineering team ~500 software, algorithm and image processing engineers globally 2nm in development , 5nm in mass production , 10nm ~60% of revenue >10yrs of flawless VLSI execution with step "A" silicon into production Computer vision a majority of AI revenue today. New product waves, in particular for advanced AI networks, anticipated from CV5, CV7 and CV3 families, respectively. Low-power and high-performance SoCs for auto and IoT markets Scalable family allows SW to be reused across a customer's portfolio Significant R&D investment funded by operations and strong balance sheet >$1 billion cumulative R&D investment into AI 925+ HEADCOUNT AI revenue contribution begins AI R&D investment begins AMBARELLA.COM COPYRIGHT AMBARELLA 2024 AI Inference the Foundation for Our Growth Successfully leveraging perception heritage into multiple high value deep learning AI inference applications Advanced Fully % Total Revenue 50% Edge AI Inference AI Automated Networks ~70% Q2 F2025 revenue AI Partially Automated Human 0% Viewing Video Processor Material covered under NDA Computer Vision 2 nd generation AI inference e.g., CNN AI networks AMBARELLA.COM COPYRIGHT AMBARELLA 2024 Advanced AI Networks including GenAI Automotive Central Domain Controllers (3 rd generation AI inference e.g., Transformer and CNN networks) Ambarella's "Algorithm First" Approach Differentiated approach enables superior efficiency and scalability Superior Image Quality AI-based Computer Vision & Radar Processing "Algorithm First" Approach Programmable AI Platform with Optimized Acceleration Image Processing Autonomous Radar CVflow - Deep Neural Network AI & Compression Driving Processing processor 17 years of 25 years of Oculii CVflow ® - Patented Image Autonomous Superior AI Adaptive AI Processing and Driving Performance Radar Compression Experience Per Watt Algorithms Experience AMBARELLA.COM COPYRIGHT AMBARELLA 2024 Scalable Deep Learning AI Inference Processor Portfolio Superior AI performance per Watt 5nm Samsung CV3-AD family of central domain controllers 3 rd generation AI (e.g., Transformer and CNN AI networks) SoCs 2024S 2024S 2024S 2024S 2023S Inference ~1.5x CV22 AI ✓ 10nm Samsung ~13x CV22 AI ✓ ~27x CV22 AI ✓ ~80x CV22 AI ✓ ~6x CV22 AI ✓ AI 2 nd generation AI (e.g., CNN AI networks) Feb 2021P Jul 2019P Dec 2018P Apr 2021P Aug 2019P Apr 2021P ¼ of CV22 AI ½ of CV22 AI Baseline (CVflow DNN AI) 4x CV22 AI 2x CV22 AI Human Viewing AMBARELLA.COM COPYRIGHT AMBARELLA 2024 AMBARELLA.COM COPYRIGHT AMBARELLA 2024 2022S ~160x CV22 AI ✓ 2022P 3.5x CV22 AI ✓ "New Products " ✓ 20xx = Calendar Year P = Production S = Sample Large and Growing Markets Serviceable market ("SAM") revenue CAGR in the high teens (K) • F2024 revenue was ~34% Auto and ~66% IoT Market • F2028 revenue SAM estimate ~70% Auto and ~30% IoT Serviceable • We are focused on AIoT edge applications, including both edge endpoints and edge infrastructure systems (e.g., AI box, edge server) • We address the megatrends of security, safety, and automation Ambarella's • Key driver is to enable electronic systems to perceive the world and make intelligent decisions ; human viewing business expected to decline as a proportion of revenue • AI has triggered new product cycles in existing IoT markets and opened entirely new opportunities in the auto and IoT markets • Advanced AI networks, including GenAI, are expected to be an important new product cycle • SAM estimates do not include autonomous driving software, advanced networks nor Generative AI ("GenAI") $10,000 Other SAM Robotics $7,500 Id/Access Smart Home AIoT Enterprise/ Public Security Auto Radar $5,000 L4/L5 SAM L2+ $2,500 Auto Front ADAS In-cabin, emirror, $- recorders F2024 F2027 Source: Ambarella, ABI, Gartner, IHS, Strategy Analytics, TSR AMBARELLA AMBARELLA .COM .C COPYRIGHT M COPYRIGHT AMBARELLA AMBARELLA 2024 2024 Automotive Applications Goal: (~34% of fiscal 2024 revenue) With CV3 Intersect Majority of Auto Revenue Today at L2+ (CV2 family + CV7 commence C2025) Visualization ADAS Driving Human Viewing + Recording Sensing + Viewing + Recording Part-time Full-Time Application Products Dataloggers, Drive Front ADAS Other ADAS L2+ L4 Recorders (emirror, in-cabin (DMS, OMS) (front facing camera L0 to L2) (pre-install,dealer-fit, aftermarket) access, personalization) C2023 market ~15% ~70% ~1% other ~3% ~0.1% penetration (% of new ~4% emirror vehicles produced globally) ~15% in-cabin C2023 → C2027 $275M → $400M $1.6M → $1.3B $200M → $500M $500M → $2.5B $200M → $900M Ambarella SAM* *does not include SW stack Ambarella Human Viewing 2 nd Generation AI 3 rd Generation AI Processors CV3-AD Platform includes: Other Ambarella CV3-AD SoCs Products • 4D radar perception SW • End-to-end AI SW Stack AMBARELLA AMBARELLA .COM .COM COPYRIGHTCOPYRIGHTAMBARELLAAMBARELLA2024 2024 Ambarella's Oculii 4D Imaging Radar Technology Oculii Edge 4D Imaging Radar Oculii Centralized 4D Imaging Radar CV3 AMBARELLA.COM COPYRIGHT AMBARELLA 2024 6-Year Automotive Revenue Funnel 4.0 ~$2.4B Consistent methodology: Won = notified of award with 1 discount factor Pipeline = in the bidding process with 2 discount factors (1) confidence in customers revenue forecast for a project and (2) probability of winning a project Our auto business generated about $77 million of revenue in F2024 and our six-year $2.4 billion funnel is an indication of the strong revenue growth we anticipate from the auto market >90% funnel is AI and a large portion of the AI is CV3 Significant revisions between funnel 4.0 and funnel 3.0 There were a significant number of revisions to the prior funnel including changes in OEM/T1 forecasts and project delays, projects won or lost and the addition of new projects Distribution of funnel revenue is exponential in shape Assumptions for auto funnel ASP to rise with time Rising adoption of new technologies in vehicles produced • Series production SoP can be ~3 years from award 3.0 = $2.3B 2.0 = $1.8B 1.0 = $600M $800 $200 $700 $400 $1,100 $1,500 4.0 = $2.4B $800 won $1,600 pipeline November 2020 November 2021 November 2022 November 2023 AMBARELLA.COM COPYRIGHT AMBARELLA 2024 AIoT Applications (non-auto AIoT applications represented ~66% of fiscal 2024 revenue) AIoT (Cells scaled to proportionally approximate F2024 AIoT revenue distribution ) Other "Security" -Access Control -Drone -Handheld Smart Imaging -LLMs -Multi- Home modal Enterprise "Security" -Robotics Security -Videocon. -VR -Wearable IoT historically camera-based human viewing and endpoints 2 nd gen AI SoCs now a majority of revenue and typically process CNN networks, enabling machine sensing + human viewing → ~CV2 ASP 2x video processor ASP 3 rd gen AI SoCs add transformer network processing support. Expected to enable entirely new applications, potentially expanding SAM deeper into the network → further increase in ASP anticipated Enterprise currently the largest security market Global installed base in C2023 estimated to be 1.1 billion cameras with 10% based on embedded deep learning SoCs Installed base growing low single digits Y/Y (AI a key driver) Annual unit demand heavily influenced by replacement units (AI a key driver) Consistent with our revenue, we estimate a majority of the installed base is represented by enterprise + public capital expenditures, the balance home spending Other AIoT includes emerging advanced AI networks (e.g. CLIP, VLM, LLM, GenAI, multi-modal) and legacy 2 nd gen and 3 rd gen AI SoCs are enabling entirely new applications, including access control, consumer cameras, fixed and mobile robotics and virtual reality GenAI represents both a new product cycle and an extension into new markets as larger models can be condensed into smaller, more efficient models at the edge N1 serves the "missing middle" for 3B to 50B parameter models such as Llama-v2, Gemma, Phi-2 and CLIP in network infrastructure HW CV7 family addresses 3B parameter models including within an endpoint AMBARELLA AMBARELLA .COM .COM COPYRIGHTCOPYRIGHTAMBARELLAAMBARELLA2024 2024 Attachments Original document Permalink Disclaimer Ambarella Inc. published this content on October 10, 2024 , and is solely responsible for the information contained herein. Distributed by Public, unedited and unaltered, on October 10, 2024 at 07:51:09.191 .