Sensetime Group, Inc. Class B · HKEX:20
Sensetime Group, Inc. Class B — FY2023 Earnings Presentation
32 slides from Sensetime Group, Inc. Class B's earnings presentation, March 2024
Slide 1: 2023 Annual Results Presentation
SenseTime Group Inc. 2023 Annual Results Presentation —— TO CREATE AN AI-EMPOWERED FUTURE THROUGH INNOVATION 2024.03
Slide 2: Explosive Growth in Generative AI Revenue in 2023
Explosive Growth in Generative AI Revenue in 2023 Revenue Growth Operating Computing Power of Group Revenue 12,000P 35% Revenue 10% 2022 2023 Model-as-a-Service (MaaS) RMB 3.4bn RMB 3.8bn 4,500+ ( 11.1%) 10.6% RMB 5.5bn ( 2.2%) Group Revenue Receivables collection Total Employee Operating EBITDA (lo
Slide 3: Strong Foundation in SenseCore and Large Model Commercialization
Strong Foundation in SenseCore and Large Model Commercialization 12,000P Upgrading every 3 months Extensive Computing Rapid Iteration of Power Resources Foundation Model AI Model Capabilities Generative AI Applications Intelligence Perception Generation Content Natural Language Text-to-images Digita
Slide 4: Rearranged Business Segment - Generative AI as Core Growth Driver
Rearranged Business Segment - Generative AI as Core Growth Driver Original Business Segments Business Segments Post Rearrangement 2022 2023 yoy Revenue Revenue % Smart Business Generative AI RMB 395bn RMB 1.18bn 200% Smart City Traditional AI * RMB 3.12bn RMB 1.84bn 41% * Smart City's revenue accoun
Slide 5: Broad Commercialization of Generative AI
Generative AI Broad Commercialization of Generative AI First 8 approved Dozens of customers with Consumer usage volume Growth Index and Innovation Index in Large Models in China orders over RMB 10m increase in 6 months 2023 China AI Development Platform Market Report1 Public Cloud Private Deployment
Slide 6: SenseCore Facilities Production of Large Models with Trillion-Parameter
Generative AI SenseCore Facilities Production of Large Models with Trillion-Parameter Operating computing power* Shanghai Lingang AIDC Integrated Intelligent Computing Network in China Super-large cluster Linear speedup interconnection ratio Jinan Ankang Chongqing Fuzhou Training for Recovery time Y
Slide 7: Integration and Deployment of Domestic Chips
Generative AI Integration and Deployment of Domestic Chips Open computing ecosystem … Standard operator interfaces Mainstream AI algorithm frameworks provided by DeepLink Compatible with AI Large Model computing needs wide-range of hardware supported by CUDA covered Mainstream domestic AI chips 7
Slide 8: Milestones of SenseNova Large Model Iteration
Generative AI Milestones of SenseNova Large Model Iteration Initial release of Large First China foundation Performance closely Language Model with model surpassing comparable with GPT4 100bn+ parameters GPT3.5-Turbo across wide range of Performance performance evaluation benchmarked GPT4-Turbo Supp
Slide 9: SenseNova 4.0 – Surpassed GPT-3.5 and Comparable with GPT-4
Generative AI “SenseNova 4.0” – Surpassed GPT-3.5 and Comparable with GPT-4 Exam 90 80 77.4 76.4 69.9 70 68.1 Coding 60 Language 50 40 4.0 GPT4 4.0 GPT4 30 Math Knowledge 85.7 84.6 66.0 66.1 Reasoning SenseChat V4 GPT3.5 GPT4 4.0 GPT4 4.0 GPT4 * According to the OpenCompass LLM evaluation platform 9
Slide 10: World-leading Performing Medium-size Open-sourced Models
Generative AI World-leading Performing Medium-size Open-sourced Models Ranked 1st Ranked 2nd Release Ranked 3rd 2024.1.17 2023.7.18 2024.2.21 2022.11.30 2023.3.14 date InternLM2- InternLM2- 评测集 Chat-7B Chat-20B Llama2-7B Llama2-13B Gemma-7B ChatGPT GPT-4 AGIEval 47.2 50.3 21.8 30.9 -- 39.9 55.1 BBH
Slide 11: Small-size On-device Models Achieve Large Model Functioning
Generative AI Small-size On-device Models Achieve Large Model Functioning On-device Model Series 7B 1.8B … Supporting smartphones empowered by SenseTime AI UniAD and Multimodal Large Models Industry leading inference speed first deployed on on 7B model reaching 11
Slide 12: Best Code Interpreter in China with Steady Function Calling
Generative AI Best Code Interpreter in China with Steady Function Calling Leading code interpreter Accuracy on 1,000+ data analysis evaluation sets vs GPT-4 84.62% Ranked in SuperClue coding evaluation Supporting: Multiple file types Cross-file analysis Multiclass analysis, statistics & visual
Slide 13: Strong Multimodal Processing Capability
Generative AI Strong Multimodal Processing Capability 13
Slide 14: SenseMirage Text-to-image Model - Best in China and Top 3 Globally
Generative AI SenseMirage Text-to-image Model - Best in China and Top 3 Globally 2023-01 2023-04 2023-07 2024-01 SenseMirage 1.0 SenseMirage 2.0 SenseMirage 3.0 SenseMirage 4.0 0.8B parameters 1.4B parameters 7B parameters 10B parameters Input prompts to generate Realistic lighting effects and Profe
Slide 15: Text-to-image Evaluation Examples
Generative AI Text-to-image Evaluation Examples SenseMirage Competitor A Competitor B Prompt： Wise old elephant, full body, in a lush jungle, calm and gentle, detailed and lifelike, photography realism, National Geographic style Prompt： The European-style palace is located on the top of the mountain
Slide 16: Ultra HD Text-to-Video Model
Generative AI Ultra HD Text-to-Video Model 16
Slide 17: Large Model Commercial Application
Generative AI Large Model Commercial Application Finance Healthcare Enterprise Copilot Anthropomorphic Smart Device Interaction Established generative AI application value and leading position across multiple sectors 17
Slide 18: Finance & Healthcare: Large Model Driving Industry Transformation
Generative AI Finance & Healthcare: Large Model Driving Industry Transformation “Digital employees" for internal services and "digital customer service" for client interactions No.2 MedBench professional pharmacist Accurate search in large-scale document evaluation databases based on Retrieval-Augme
Slide 19: Anthropomorphic Interaction: Broadening To-C Application
Generative AI Anthropomorphic Interaction: Broadening To-C Application Character creation and customization High-speed, high-quality IP character customization Long dialogue memory Characters accurately recall lengthy conversations 60%+ Next-day user retention rate 2,000+ Max. number of dialogue int
Slide 20: Traditional AI is the Foundation of Generative AI Development
Traditional AI Traditional AI is the Foundation of Generative AI Development % of Total Revenue Traditional AI Traditional AI Generative AI 传统AI业务 54% and Smart Auto 82% Generative AI 2022 and Smart Auto 2023 (Smart City accounts for 29%) (Smart City accounts for less than 10%) 20
Slide 21: Smart Auto: Expanding Mass Production and Empowered by Large Models
Smart Auto: Expanding Mass Production and Empowered by Large Models Smart Auto Steady revenue growth & AGI bringing a new paradigm for Smart Auto Significant increase in mass production Revenue (RMB mn) Application Cumulative delivered +31% 384 Closed-loop vehicles Smart Pilot Smart Cabin V2X data s
Slide 22: Smart Pilot: Integrated Cabin-Pilot Driving Experience
Smart Pilot: Integrated Cabin-Pilot Driving Experience Smart Auto UniAD: The first autonomous driving Large Model integrated with perceptual decision-making, DriveMLM Solution awarded the Best Paper at CVPR 2023 and achieving a qualitative leap of deliveries Closed-loop driving Interventions bas
Slide 23: Smart Cabin: Pioneering New Interaction Experience with Large Models
Smart Cabin: Pioneering New Interaction Experience with Large Models Smart Auto Four major smart cabin Smart cabin-based 11 industry-leading products Large Models new smart cabin features Domestic and global mainstream automobile companies with demand of Large Model services …… 23
Slide 24: Strategic Objectives in 2024
Strategic Objectives in 2024 Maintaining Driving Business Profitability for Technology Growth Core Business Leadership Upgrading SenseNova Large Model Series Large Models gradually penetrating into the Promoting the development of chips with high- production field and bringing revolutionary changes
Slide 25: SenseTime Tech Day Coming Soon
nM sensetime 2024.04 SenseTime Tech Day Coming Soon
Slide 26: Financial Review
CA Bin sensetime Financial Review
Slide 27: FY2023 Financial Overview
FY2023 Financial Overview Revenue Gross Profit EBITDA Net Loss (RMB million) (RMB million) (RMB million) (RMB million) 3,809 3,406 (5,570) (5,450) 2,542 (6,093) (6,495) 1,501 2022 2023 2022 2023 2022 2023 2022 2023 (19.0%) (10.6%) 66.8% 44.1% (66.5%) (2.2%) (64.5%) 6.6% yoy growth Gross margin yoy g
Slide 28: Operating Expenses
Operating Expenses Operating Expenses（1） Opex Growth (RMB million) Opex (excl. SBC) SBC (1.7%) 6,483 464 5,795 324 6,018 (10.6%) 5,471 2022 2023 2022 2023 yoy growth（incl. SBC） Note: 1. Opex = R&D + G&A + Selling Expenses 28
Slide 29: Working Capital
Working Capital Provision for Impairment as % of gross trade receivables2 Turnover Days – Period Ending Method(1) Net Impairment Losses on financial assets and contract assets3 (RMB million) Cash Conversion Cycle 33.1% 53.0% 2,000 60.0% 434 334 1,800 50.0% 1,600 1,400 1,200 1,668 1,835 40.0% 1,000 3
Slide 30: Capex and Cash Balance
Capex and Cash Balance Capex(1) Total Cash(2) Equity and Bond Investment(3) (RMB million) (RMB million) (RMB million) Other Capex Office Space Related Capex Net Cash Equity Investment Total Borrowings Bond Investment 49.0% 43.6% 15,080 3,229 11,409 6,677 6,527 4,473 1,523 5,167 1,636 11,851 3,301 6,
Slide 31: Slide 31
Slide 32: THANKS!
THANKS! TO CREATE AN AI-EMPOWERED FUTURE THROUGH INNOVATION 2024.03