Mobileye Global Inc. · NASDAQ:MBLY
Mobileye Global — Investor Day (December 2024)
102 slides from Mobileye Global's investor day presentation, December 2024
Slide 1: INNOVATE 2024: Defining Success in a Fully Autonomous Driving System
INNOVATE 2024 Defining Success in a Fully Autonomous Driving System Prof. Amnon Shashua, CEO Prof. Shai Shalev-Schwartz, CTO © mobileye
Slide 2: Forward-Looking Statements
Forward-Looking Statements Mobileye’s business outlook, guidance and other statements in this presentation that are not statements of historical fact, including statements about our beliefs and expectations, are forward- looking statements and should be evaluated as such. Forward-looking statements
Slide 3: Mobileye's Vision: Solving Autonomy and Step-Change in Road Safety
Mobileye’s Vision: Solving Autonomy and Step-Change in Road Safety* Moving from Hands-On / Eyes-On to Hands-off / Eyes-off and No-driver ADAS SuperVision™ Chauffeur™ Drive ™ EYES-ON / HANDS-ON HANDS-OFF / EYES-ON EYES-OFF NO DRIVER IN THE CAR Front Camera (1V) • “Vision Zero” - comprehensive • Givin
Slide 4: Driving Demos
Driving Demos SuperVision™ Drive™ HANDS-OFF / EYES-ON ROBOTAXI Polestar / Zeekr ID Buzz • Production vehicle, deployed in both China and • Pre-production Europe • Project was launched less than 1 year ago • Point-to-point navigation (selected by investors) • Time difference of less than 10% compared
Slide 5: INNOVATE 2024
INNOVATE 2024 Defining Success in a Fully Autonomous Driving System Prof. Amnon Shashua, CEO Prof. Shai Shalev-Schwartz, CTO © mobileye
Slide 6: Forward-Looking Statements
Forward-Looking Statements Mobileye’s business outlook, guidance and other statements in this presentation that are not statements of historical fact, including statements about our beliefs and expectations, are forward- looking statements and should be evaluated as such. Forward-looking statements
Slide 7: Approaches to Solve Autonomy
Approaches to Solve Autonomy AI Geographic Sensors Cost Modularity Deployed How to Solve Autonomy Approach Scalability High quality Lidars, stereo cameras, Lidar-centric CAIS ? L4 radars, … Camera alone would be able to achieve Camera only End-to-end L2+ autonomy Sensor redundancy which requires: •
Slide 8: Requirements For Success: From Eyes-on to Eyes-off
Requirements For Success: From Eyes-on to Eyes-off 01 02 03 04 05 Productization Scalability Technology Cost Safety Getting from demo Geographic and Stack Solving autonomy Eyes-on: human to real product ODD while controlling serves as the safety CAIS system (compute, driver sensors…) and EyeQ6H Eyes
Slide 9: Productization: From Demo to Real Product (Consumer Car)
Productization: From Demo to Real Product (Consumer Car) Why aren’t more companies being seriously considered as players in solving autonomous driving, when many showcase nice demos? The path to productization presents significant challenges Geographic scalability Multiple car models and OEMs: Meeti
Slide 10: Productization: From Demo to Real Product (Consumer Car)
Productization: From Demo to Real Product (Consumer Car) Mobileye Advanced products deployed in Europe, China, and soon in the U.S. In production (SuperVision52) Pre-production SuperVision62 17 car models Chauffeur63 10 car models Drive64 © mobileye
Slide 11: Requirements For Success: From Eyes-on to Eyes-off
Requirements For Success: From Eyes-on to Eyes-off 01 02 03 04 05 Productization Scalability Technology Cost Safety Getting from demo Geographic and Stack Solving autonomy Eyes-on: human to real product ODD while controlling serves as the safety CAIS system (compute, driver sensors…) and EyeQ6H Eyes
Slide 12: Scalability: Geographic and ODD
Scalability: Geographic and ODD Geographic ODD The ability for a system to operate across diverse The ability of an autonomous system to expand its ODD regions over time • Semantic data (e.g., traffic signs/lights, traffic laws, public • Road types (e.g., highway, rural, arterial, urban) transportat
Slide 13: Geographic Scalability: REM
Geographic Scalability: REM 56.6B Miles Harvested in 2021-2024 Total miles harvested so far 7.71B 7.94B 5.9B 5.5B 29.6B 5.05B 4.75B 3.5B 3.2B 2.6B 2.1B Miles harvested in 2024 1.3B 1.4B 1B 0.2B 0.3B Q1 2021 Q3 2021 Q1 2022 Q3 2022 Q1 2023 Q3 2023 Q1 2024 Q3 2024 62.4M Miles collected daily © mobiley
Slide 14: Geographic Scalability: REM Coverage and Features
Geographic Scalability: REM Coverage and Features Crowd Speed Traffic light associations Priorities Drivable Path Lane level accuracy … © mobileye
Slide 15: Requirements For Success: From Eyes-on to Eyes-off
Requirements For Success: from Eyes-on to Eyes-off 01 02 03 04 05 Productization Scalability Technology Cost Safety Getting from demo Geographic and Stack Solving autonomy Eyes-on: human to real product ODD while controlling serves as the safety CAIS system (compute, driver sensors…) and EyeQ6H Eyes
Slide 16: Gen I: Solving Autonomy at Scale, Gen II: Cost Reduction
Gen I: Solving Autonomy at Scale, Gen II: Cost Reduction Gen I With Mobileye's current technology stack, including EyeQ6H, highly efficient AI as part of our CAIS approach, Imaging-radar, and more Mobileye is well-positioned to solve autonomy through its ongoing projects EyeQ6H Mobileye BSR Gen II E
Slide 17: Inference chip (EyeQ6H)
Inference chip (EyeQ6H) Design for efficiency © mobileye
Slide 18: Hardware Architectures Tradeoff: Flexibility vs. Efficiency
Hardware Architectures Tradeoff: Flexibility vs. Efficiency efficiency Fixed-function High Efficiency GPU efficiency Low CPU Special purpose Flexibility General purpose © mobileye
Slide 19: EyeQ6H: Distinct Architectures
EyeQ6H: 5 Distinct Architectures EyeQ6H • Address Mobileye’s • Enable accelerating high efficiency and range of parallel flexibility needs compute paradigms efficiency High XNN PMA VMP Efficiency MPC efficiency Low MIPS Special Flexibility General purpose purpose © mobileye
Slide 20: EyeQ6H vs. EyeQ5H: 2x in TOPS, But 10x in FPS!
EyeQ6H vs. EyeQ5H: 2x in TOPS, But 10x in FPS! EyeQ5H EyeQ6H 16 TOPS (int 8) 34 TOPS (int 8) 27W (max) 33W (max) © mobileye
Slide 21: EyeQ6H vs. Orin: It's Not All About TOPS
EyeQ6H vs. Orin: It’s Not All About TOPS Theoretical TOPS Factor x8 34 TOPS (int8) 275 TOPS (int8) Frames per Second for ResNet50 only factor x2 Conclusion: TOPS are a poor measure for compute capabilities © mobileye
Slide 22: Mobileye's Imaging Radar
Mobileye’s Imaging Radar BSR / BSRC Support high speeds, dense traffic, arterial, rural and urban scenarios BSR (Front LRR/ MRR) Designed to enable L4 capabilities Short range capabilities designed to replace Short- Range Lidars and enable autonomous parking scenarios SOP 2025 BSRC (corner radar) ©
Slide 23: Mobileye's Imaging Radar
Mobileye’s Imaging Radar Hundreds of tests Hundreds of tests Western OEM requirements Mobileye’s Imaging Radar performance (done by the OEM) conducted by OEMs over Lying dummy facing vehicle direction (open space) >130m 181m-214m the past 2 years Lying dummy facing vehicle direction (near guard rail
Slide 24: Mobileye's Imaging Radar
Mobileye’s Imaging Radar A brief overview of one scenario showcasing the tests and performance evaluations conducted by the OEM on our radar Detection of a dummy with a helmet lying on-road facing our direction At a distance of 8 meters © mobileye
Slide 25: Mobileye's Imaging Radar
Mobileye’s Imaging Radar A brief overview of one scenario showcasing the tests and performance evaluations conducted by the OEM on our radar Detection of a dummy with a helmet lying on-road facing our direction At a distance of 100 meters © mobileye
Slide 26: Mobileye's Imaging Radar
Mobileye’s Imaging Radar A brief overview of one scenario showcasing the tests and performance evaluations conducted by the OEM on our radar Detection of a dummy with a helmet lying on-road facing our direction At a distance of 220 meters © mobileye
Slide 27: Imaging Radar Enabling Autonomy at Scale
Imaging Radar Enabling Autonomy at Scale Redundancy • An independent sensor with different failure modes from cameras and lidars Performance • Density of cameras BSR (Front LRR/ MRR) • High accuracy in detecting objects in extended distances • Enables very close 360° coverage Cost-optimized for scal
Slide 28: Hardware in the Loop (HIL) Validation
Hardware in the Loop (HIL) Validation HIL enables a quick overnight validation cycle Real-time Playback of 10,000 hours Each session reproduces all data Efficient compression of sessions to of real world data of mission-collected received by the ECU during the reduce HIL server costs critical sessio
Slide 29: Mobileye Extremely Efficient AI
Mobileye Extremely • Transformers at x100 efficiency Efficient AI • Efficient labeling by Auto Ground Truth • Meeting Safety Goals • Sufficiently high MTBF • No ‘unreasonable risk’ Will be covered by Shai © mobileye
Slide 30: Requirements For Success: From Eyes-on to Eyes-off
Requirements For Success: From Eyes-on to Eyes-off 01 02 03 04 05 Productization Scalability Technology Cost Safety Getting from demo Geographic and Stack Solving autonomy Eyes-on: human to real product ODD while controlling serves as the safety CAIS system (compute, driver sensors…) and EyeQ6H Eyes
Slide 31: Mobileye's Vision: Solving Autonomy and Step-Change in Road Safety
Mobileye’s Vision: Solving Autonomy and Step-Change in Road Safety* Moving from Hands-On / Eyes-On to Hands-off / Eyes-off and No-driver ADAS SuperVision™ Chauffeur™ Drive ™ EYES-ON / HANDS-ON HANDS-OFF / EYES-ON EYES-OFF NO DRIVER IN THE CAR Surround Camera (optional radar) Surround Camera + Radar
Slide 32: Requirements For Success: From Eyes-on to Eyes-off
Requirements For Success: From Eyes-on to Eyes-off 01 02 03 04 05 Productization Scalability Technology Cost Safety Getting from demo Geographic and Stack Solving autonomy Eyes-on: human to real product ODD while controlling serves as the safety CAIS system (compute, driver sensors…) and EyeQ6H Eyes
Slide 33: Safety: How Safe is Safe Enough?
Safety: How Safe is Safe Enough? Why? Is a self-driving system with Human-level MTBF • Humans are expected to properly respond to events even when those sufficient on its own? are extremely rare (e.g., baby lying on the road) Answer • Human driving statistics are heavily influenced by illegal or irr
Slide 34: Safety: Mobileye Safety Architecture to Meet the Safety Goals
Safety: Mobileye Safety Architecture to Meet the Safety Goals Absence of “unreasonable risks”, with a transparent definition of the boundary between reasonable and “unreasonable risks” Planning Identifiable Errors Reproducible Errors Black Swans Errors due to mis-interpreting the Hardware failures:
Slide 35: Summary Requirements for Success
Summary Requirements • Productization: From demo to real product for Success • Scalability: Geographic and ODD From Eyes-on to Eyes-off • Cost: Controlling system and development cost • Technology stack: CAIS, HIL, EyeQ6H and imaging-radar, … • Safety: Safety bars for autonomous driving systems © mo
Slide 36: Navigating The Path to Autonomous Mobility with Extremely Efficient AI
Navigating The Path to Autonomous Mobility with Extremely Efficient AI © mobileye
Slide 37: Prologue: 6 AI Revolutions
Prologue 6 AI Revolutions Machine Learning Mobileye was the First to Utilize Deep Learning Generative AI Universal Learning Sim2Real Reasoning © mobileye
Slide 38: Machine Learning and Deep Learning in Mobileye
Machine Learning and Deep Learning in Mobileye Machine Learning Revolution ML • Shifting from expert systems to learning algorithms • SVM and AdaBoost on hand-tuned features DL Instead of hand-tuned features, let the machine learn the features as well • 2012: Krizhevsky-Sustkever-Hinton Imagenet pap
Slide 39: Prologue: 6 AI Revolutions – Transformers
Prologue 6 AI Revolutions Machine Learning Deep Learning Generative AI Transformers Universal Learning Sim2Real Reasoning © mobileye
Slide 40: Pre-Transformers: Object Detection Pipeline
Pre-Transformers: Object Detection Pipeline Clustering and max suppression 2D to 3D © mobileye
Slide 41: Three Revolutions of Generative Pre-trained Transformers (GPTs)
Three Revolutions of • Tokenize everything Generative Pre-trained • Generative, Auto-regressive Transformers (GPTs) • Transformer architecture: ’Attention is all you need’ © mobileye
Slide 42: Three Revolutions of Generative Pretrained Transformers
Three Revolutions of Generative Pretrained Transformers 1 Tokenize everything Input: Transcribe each input modality (e.g., text, images) into a sequence of tokens Output: Transcribe each output modality as a sequence of tokens and employ generative, auto-regressive models with suitable loss function
Slide 43: Three Revolutions of Generative Pretrained Transformers
Three Revolutions of Generative Pretrained Transformers 2 Generative, Auto-regressive Previous approach: Classification or regression with fixed, small size, outputs (e.g., ImageNet) Current approach: Learn probabilities for sequences of arbitrary length (e.g., sentence generation) Key Features: Cha
Slide 44: Three Revolutions of Generative Pretrained Transformers
Three Revolutions of Generative Pretrained Transformers Example: Consider a 1000x1000 pixel image containing 4 vehicles, with the image divided into 10x10 pixel patches. What are the probabilities for identifying vehicle positions when not using the chain rule compared to when using the chain rule?
Slide 45: Three Revolutions of Generative Pretrained Transformers
Three Revolutions of Generative Pretrained Transformers 3 Transformer architecture: ’Attention is all you need’ Tailored for problem of predicting * +,-./%&! +,-./% , +,-./%'! , … , +,-./( ] ! Transformer layer ... Self-reflection FCN FCN FCN FCN FCN FCN FCN ... Self-attention ... © mobileye
Slide 46: Transformers Layer: Group Thinking Analogy
Transformers Layer: Group Thinking Analogy Imagine a team discussing a project - Each person has their own area of expertise - they all contribute to the overall outcome - Everyone is working simultaneously rather than one after another Self-reflection Self-attention Each participant takes time alon
Slide 47: Transformers Layer: Self-Reflection
Transformers Layer: Self-Reflection - Each token individually processes its ‘knowledge’ using a multi-layer-perceptron, without interacting with other tokens ! Input " 9 FCN FCN FCN ! " ... ... ... ... Output © mobileye
Slide 48: Transformers Layer: Self-Attention
Transformers Layer: Self-Attention - Each token send ‘query’ to the other tokens, which respond with values if their ‘key’ match the ‘query’ - The querying token then averages the received values, facilitating inter-token connectivity Relevancy ... "#$%&! ... ... '$&" i,j Questions 9 " ! ... ... ...
Slide 49: Transformers Layer: Self-Attention
Transformers Layer: Self-Attention Normalizes Scores: It converts raw attention scores into normalized probabilities Probability Distribution: Each set of attention scores is transformed so that their probabilities sum to 1 Focus Mechanism: This allows the model to weigh different parts of the input
Slide 50: Transformers: Complexity
Transformers: Complexity 5 ∗ (8 9+ + 8+ 9) #layers Self Self reflection attention Cost per layer for alternative architectures: Fully Connected Network (FCN) Recurrent Neural Network (RNN) Flatten #$ values ‘Talks’ only with previous token ... Input ... ... ... ... ... ... ... ... Output ... ... Con
Slide 51: 'Effective Sparsity' of Transformers
‘Effective Sparsity’ of Transformers Transformers Denser, but effectively selects Sparser ) ) Any modality only a few past tokens for $ #+# $ communication Long-Short-Term-Memory Fully Connected Network Convolutional Neural (LSTM) (FCN) Networks (CNNs) Recurrent Neural Networks (RNN) # # Markov spar
Slide 52: The 3 Revolutions Enable a Universal Solution
The 3 Revolutions • Handle all types of inputs Enable a Universal Solution • Deals with uncertainty (by learning probability) • Enables all types of outputs The ultimate learning machine? © mobileye
Slide 53: End-to-End Driving: From Pixels to Control Commands
End-to-End Driving: From Pixels to Control Commands Input: images Output: Control commands © mobileye
Slide 54: Network Architecture: Vanilla Transformer
Network Architecture: Vanilla Transformer - CNN backbone for creating image tokens: - ! = 32 high resolution images are converted to 32 images of resolution 20x15 yielding %& = 300 “()*+,-” per image, and / = 256 channels - Encoder: - We have % = ! ∗ %& = 9600 “)456+ 789+:-”, each at dimension / = 2
Slide 55: © mobileye
x1699 y:489
Slide 56: © mobileye
x210 y:-407
Slide 57: © mobileye
x1-363_y:307 S [BM x: 744 y:759 :-567 y:546 World Map +
Slide 58: Limitations of End-to-End
Limitations of End-to-End Will it reach the desired MTBF? 1 - ‘Rare and correct’ vs. ‘common and incorrect’ - Fail to learn important abstractions (calculator as a study case) - The 'Shortcut Learning' Problem - Long tail problem 2 Eliminating ‘unreasonable risk’ 3 Brute force © mobileye
Slide 59: Mobileye Extremely Efficient AI
Mobileye Extremely • Transformers at x100 Efficient AI Efficiency • Efficient labeling by Auto Ground Truth © mobileye
Slide 60: Transformers for Sensing State and Control
Transformers for Sensing State and Control Input: images Output: All objects & trajectory © mobileye
Slide 61: STAT: Sparse Typed Attention
STAT: Sparse Typed Attention Vanilla transformer: @ @ n d+d n STAT: - Token Types: Each token has a “type” - Dimensionality: of embeddings and self-reflection matrices may vary based on the token type. - Token Connectivity: The connectivity between tokens is sparse and depends on their types - Link
Slide 62: STAT: Sparse Typed Attention
STAT: Sparse Typed Attention Vanilla transformer: !! " + d! n * = 32 ./01$2 STAT Encoder for Object Detection: ... ... ... 300 image tokens 300 image tokens 300 image tokens - Token types: - Image tokens: recall, we have , = 32 images each with ... ... ... ... 32 Link tokens 32 Link tokens 32 Link t
Slide 63: Parallel Auto-Regressive (PAR)
Parallel Auto-Regressive (PAR) We need to detect all objects in the scene: What is the order? Auto-Regressive: It doesn’t matter due to the chain rule ! Price of sequential decoding - Sequential decoding is costly on all modern deep learning chips (due to IO) ”Truck and trailer” problem - We added u
Slide 64: Parallel Auto-Regressive (PAR)
Parallel Auto-Regressive (PAR) Input images - The decoder contains query heads which perform cross attention with the encoder’s link CNN Tokenization tokens entirely in parallel - Each query head outputs, auto-regressively, 0/1/2 objects (independently and in parallel to the other query heads) STAT
Slide 65: Intermediate Summary
Intermediate Summary Transformers revolutionized AI - The good Machine Learning - Universal, generative, AI Deep Learning - The bad - Can’t separate “correct & rare” from “wrong & common” - Miss important abstractions Generative AI - Questionable when very high accuracy is required Transformers - Th
Slide 66: Mobileye Extremely Efficient AI
Mobileye Extremely • Transformers at x100 Efficient AI Efficiency • Efficient labeling by Auto Ground Truth © mobileye
Slide 67: Automatic Ground Truth: CAIS vs. End-to-End
Automatic Ground Truth: CAIS vs. End-to-End Compound AI System End-to-end solution - Injecting abstractions: Sensing State, RSS, - Much more data PGF, etc. - Unsupervised - Need to label data: Normally does through supervised learning High End-to-end Data needed Using AutoGT CAIS Low Low Amount of s
Slide 68: Automatic Ground Truth: How to Reduce #Labels
Automatic Ground Truth: How to Reduce #Labels Easier problem to solve The future is known - Since the future is known - Kinematics become easier - Circumvent temporary occlusions - Can focus on short range + tracking - Powerful (expensive) sensor (e.g., 360° Lidar) Offline compute - Train foundation
Slide 69: Automatic Ground Truth: Foundation Model
Automatic Ground Truth: Foundation Model © mobileye
Slide 70: Automatic Ground Truth Final Product
Automatic Ground Truth Final Product © mobileye
Slide 71: Safety Goals
Safety Goals Super-human MTBF No ‘unreasonable risk’ • ‘No lapse of judgement’ • ‘No reproducible errors’ • ‘Identifiable errors’ © mobileye
Slide 72: The Boundary Between 'Reasonable' and 'Unreasonable'
The Boundary Between ‘Reasonable’ and ‘Unreasonable’ Planning Hardware “AI Bugs” Black Swans The RSS model defines what is FuSa standard requires SOTIF requires • No single point of failure – ‘reasonable’ to assume on the each system failure must • No single point of failure • Hunting and fixing fut
Slide 73: Examples of PGF as Part of Our Architecture
Examples of PGF as Part of Our Architecture Physical Objects and Ego Motion Primary – 3 low-level sub-systems, fused by GNN, fed into an RSS-based driving policy module Guardian – Approves P if its command adheres with RSS according to 2oo3 individual sub-systems Fallback - If 2oo3 state we violate
Slide 74: Summary: Mobileye Extremely Efficient AI
Summary Mobileye Extremely - Transformers at x100 efficiency Efficient AI - Efficient labeling by Auto Ground Truth - Safety Goals
Slide 75: INNOVATE 2024
INNOVATE 2024 Defining Success in a Fully Autonomous Driving System Prof. Amnon Shashua, CEO Prof. Shai Shalev-Schwartz, CTO © mobileye
Slide 76: Mobileye's Growth Story
INNOVATE 2024 Mobileye’s Growth Story Nimrod Nehushtan EVP Strategy & Business Development December 2024 © mobileye
Slide 77: Forward-Looking Statements
Forward-Looking Statements Mobileye’s business outlook, guidance and other statements in this presentation that are not statements of historical fact, including statements about our beliefs and expectations, are forward-looking statements and should be evaluated as such. Forward-looking statements i
Slide 78: Mobileye's Vision: Solving Autonomy and Step-Change in Road Safety
Mobileye’s Vision: Solving Autonomy and Step-Change in Road Safety* Moving from Hands-On / Eyes-On to Hands-off / Eyes-off and No-driver ADAS SuperVision™ Chauffeur™ Drive™ EYES-ON / HANDS-ON HANDS-OFF / EYES-ON EYES-OFF NO DRIVER IN THE CAR Front Camera (1V) • “Vision Zero” - comprehensive • Giving
Slide 79: Mobileye's Growth Story: Base ADAS
Mobileye’s Growth Story: Substantial 2023-2024 progress lays Base ADAS foundation for strong future growth Progress / Tailwinds Headwinds Won 95%+ of RFQs issued by Top 10 customer base during 2023-2024, • China – Emergence of a “race to the bottom” price dynamic extending our core ADAS position wit
Slide 80: Mobileye's Growth Story: Advanced Products
Mobileye’s Growth Story: Substantial 2023-2024 progress lays Advanced Products foundation for strong future growth Progress / Tailwinds Headwinds Winning 1st SuperVision, Chauffeur & Drive ™ programs with a global • Temporary delay (mid-’23 to mid-’24) in decision making OEM (VWG) among global OEMs
Slide 81: Market Evolution & Business Development Progress
Market Evolution & Business Development Progress © mobileye
Slide 82: Product Segments Evolution
Product Segments Evolution What is the OEM vision for ADAS / AD product segmentation in the near-to-mid-term? Today 2026+ Segment Standard Premium Minimal Mid Premium Point-to-point Hands Off In-lane Hands Off on Highways 5-StarSafety Rating Key Value Regulation-Compliant ADAS Regulation-Compliant ADA
Slide 83: What Are the Core Principles OEMs Look for?
What Are the Core Principles OEMs Look for? Clear Path to Cost-Efficient Autonomy Technical and Cost Leadership • State-of-the-art technology • Optimized system cost © mobileye
Slide 84: What Are the Core Principles OEMs Look for?
What Are the Core Principles OEMs Look for? Clear Path to Cost-Efficient Autonomy Technical and Cost Leadership • State-of-the-art technology • Optimized system cost Trusted, Long-Term Partner • Fastest time to market • Execution proficiency • Trusted functional performance • Controllability Proven
Slide 85: Key OEM Decision Criteria: How Do Competitors Stack Up?
Key OEM Decision Criteria: How Do Competitors Stack Up? Trusted Fastest Time Optimized Execution State-of-the-art Functional Controllability Modularity to Market System Cost Proficiency Technology Performance Legacy Tier 1s ? Silicon Actors ? Start-Ups ? ? ? ? © mobileye
Slide 86: Design Win Process Overview
Design Win Process Overview What are the key steps in OEM’s technology partner selection process? Negotiations & Interest Recurring Engagement Due Diligence Nominated Decision Making • What is the exact vehicle architecture? • What is the project execution plan? • What are the system requirements? •
Slide 87: Business Development Engagement Status – December '23
Business Development Engagement Status – December ‘23 Where were we last year? Negotiations & Interest Recurring Engagement Due Diligence Nominated Decision Making SuperVision/Chauffeur Surround ADAS Drive © mobileye
Slide 88: Business Development Engagement Status – December '24
Business Development Engagement Status – December ‘24 Where are we today (highlighted in green are positive progressions through the stages) Negotiations & Interest Recurring Engagement Due Diligence Nominated Decision Making SuperVision/Chauffeur Surround ADAS Drive © mobileye
Slide 89: Product Segmentation Impact on Future Revenue
Product Segmentation Impact on Future Revenue Illustrative Large OEM Case study – Based on design win projections Volume Split Revenue Split 2-3% 5-10% ASP: ~$200 Chauffeur 30-35% SuperVision ASP: ~$55 55-60% Surround ADAS Base / CE ~55% ADAS Base / CE ADAS 2024 Late-Decade 2024 Late-Decade Base / C
Slide 90: Mobileye's Growth Story
Mobileye’s Growth Story Mobileye’s long-term growth story unfolds in three stages, each defined by its strategic Forecasted revenue figures below are estimates and based on several assumptions, objectives, growth drivers, and revenue distribution including pricing and volume, and many factors may ca
Slide 91: Mobileye's Growth Story
Mobileye’s Growth Story The table below outlines the current volume pipeline and potential opportunities across all product ramps: Unprecedented Volume and Revenue Potential Current Booked Future Current and Advanced Incremental Advanced Engagement Product ASP Volume (units) Engagement OEMs RFQ Volu
Slide 92: Thank You.
Thank You. © mobileye
Slide 93: INNOVATE 2024: Finance Update
INNOVATE 2024 Finance Update Moran Shemesh Rojansky Chief Financial Officer © mobileye
Slide 94: Forward-Looking Statements
Forward-Looking Statements Mobileye’s business outlook, guidance and other statements in this presentation that are not statements of historical fact, including statements about our beliefs and expectations, are forward-looking statements and should be evaluated as such. Forward-looking statements i
Slide 95: Key Company Metrics
Key Company Metrics Last twelve months (through Q3 2024) 31.3 million Systems revenue of $1.8B System-on-chip (SoC) and Super VisionTM Systems shipped last twelve months Our Systems-on-chip have been deployed in Over 3,900 about Employees operating across 8 countries 190 million vehicles Operating C
Slide 96: Mobileye's Diverse, Global, Blue Chip Customer Base
Mobileye’s Diverse, Global, Blue Chip Customer Base 2024 Forecasted Base ADAS 2024 Forecasted Revenue by Region* Revenue by OEM (ADAS and Advanced Products) Japan/Korea CH OEMs Other 5% 13% 4% OEM 1 OEM 8 18% N. America 6% 17% OEM 7 8% China(**) 26% OEM 6 9% OEM 2 18% OEM 5 9% Rest of world Europe 6
Slide 97: China Exposure Diminished Meaningfully in 2024
China Exposure Diminished Meaningfully in 2024 EyeQ Volume Split, China OEM vs Foreign JV in China* 11.6 9.4 9.0 8.3 8.5 8.1 7.6 7.9 7 7 5.6 6.4 7.1 6.1 3.5 2.5 1.7 1.6 2.7 1.4 1.4 1.3 1.1 0.6 0.8 0.5 0.8 1.2 0.2 0.2 0.3 0.8 Q1'23 A Q2'23 A Q3'23 A Q4'23 A Q1'24 A Q2'24 A Q3'24 A Q4'24 F China- Chin
Slide 98: Strong Pipeline Across Customer Base, with Good Visibility
Strong Pipeline Across Customer Base, with Good Visibility Top 9 Anonymized OEM Shipped Awarded RFQ More than 190M EyeQ Chips Shipped to 103 Customers (thru Q3'24) 17 250M EyeQ Chips awarded future volume(*) 70 10 65 6 55 54 80M EyeQ Chips in concrete RFQ discussions (**) 57 52 5 14 7 29 36 35 35 27
Slide 99: Summary of future ADAS adoption opportunities (as of 2024)
Summary of future ADAS adoption opportunities (as of 2024) ADAS penetration rates by region (% of production) Regional distribution of 23.3m “no ADAS” produced units 29.3 17.1 15.5 12.0 5.6 3.9 2.9 2.2 NA ME/Africa JP/KR 5% 8% 4% 9% 9% 8% 6% 5% 3% S. America S. Asia(excl India) 2% EU 33% 11% 46% 56%
Slide 100: Individual OEM Design Win Example – Illustrative Financial Impact
Individual OEM Design Win Example –Illustrative Financial Impact Illustrative Annual Peak Surround ADAS SuperVision™ Chauffeur™ Year Figures Average annual RFQ Volume 800K-1,200K 200K-300K 50K-100K ASP $150 –$200 $1,250 –$1,500 $2,500 –$3,000 Revenue (at midpoint) ~$170mm ~$350mm ~$200mm Gross Profi
Slide 101: Scaling Revenues to Unlock Significant Operating Leverage
Scaling Revenues to Unlock Significant Operating Leverage R&D expenses breakdown by product 2024 Forecast 75% of our non-GAAP R&D supports development of 80% of our advanced products development Advanced Products cost is related to >50% (Supervision, Drive and internal resources Chauffer) (payroll e
Slide 102: Thank You.
Thank You. © mobileye