Finding the blockbuster AI applications of 2025

Abstract

AI applications are undergoing changes. The iteration of models, reduction in costs, accumulation of scenario data, and increasingly urgent downstream demand are driving developers to continuously explore the empowerment and reshaping of various industries with AI. Currently, we see breakthroughs in different AI application directions on both the product and monetization sides. We have selected five directions: 1) Edge-side Agent, 2) Marketing Technology (Mar-tech), 3) Enterprise Solution Decision-making, 4) CRM, and 5) Real-time Interaction (RTE), to summarize the emerging companies and applications.

Edge-side Agent functionality has been initially realized. Similar to the edge-side AI paths of Apple and others, screen recognition, simulated execution, and reinforcement learning have supported the gradual realization of Agent functionality. In October 2024, Anthropic's newly launched Claude 3.5 Sonnet demonstrated excellent Computer Use capabilities. The AutoGLM launched by Zhipu also showcased Phone Use capabilities such as online shopping and food delivery. We expect that by 2025, more Agent products in different fields will achieve various functionalities.

Advertising technology drives precise matching and intelligent bidding. Applovin significantly improved advertising matching accuracy through its Axon recommendation engine, enhancing advertiser benefits while also driving Applovin's growth: in Q3 2024, the company's software platform revenue increased by 66% year-on-year. Mintegral from HuiLiang Technology is focused on improving delivery accuracy while also enhancing intelligent bidding models, helping advertisers optimize delivery costs and effectiveness, which in turn drives the company's revenue growth: in Q3 2024, HuiLiang Technology's advertising technology business revenue achieved a year-on-year growth of 55.4%.

Enterprise solution decision-making benefits from AI tool support. AI tools play a crucial role in data collection and decision generation for enterprises and governments. Palantir's AIP platform significantly boosted its growth in the U.S. commercial sector: in Q3 2024, its U.S. commercial revenue grew by 54% year-on-year, with the number of U.S. commercial clients reaching 321, a year-on-year increase of 77%. The "Prophet AI Platform" developed by Fourth Paradigm has created over 40 AI products through the "Prophet Inside Model," empowering more than 20 industries, achieving a year-on-year revenue growth of 50.2% in the first three quarters of 2024.

AI empowers customer relationship management. Salesforce's Agentforce can perform various agency tasks such as service representation and order management, with multiple official cases showcasing the immense application potential of Agentforce. Weimeng Group has 15 major AI Agents, covering over 58 business scenarios, demonstrating significant effects on new customer delivery, old customer retention, and commercialization acceleration RTE-PaaS makes real-time interaction smoother. Agora, as one of the three global partners of OpenAI's Realtime API, can provide developers with low-latency and highly available real-time audio and video transmission services. We anticipate that large-scale promotion and usage growth of application scenarios will occur in the future if prices drop significantly. Recently, the company released a panoramic view of RTE+AI capabilities, which also guides the direction for AI-enabled real-time interaction scenarios.

2025 is expected to usher in a blossoming moment for AI applications. AI is reshaping various scenarios on both the consumer and business sides, including edge-side agents, marketing technology, enterprise solution decision-making, CRM, and real-time interaction RTE. We expect that 2025 will become a year of flourishing AI applications.

Risk warning: The speed of model iteration may not meet expectations, the integration of scenarios and AI may progress slower than anticipated, and the prosperity of downstream industries may fall short of expectations. Main Text

1. Edge-side Agent

1.1 Anthropic: Claude 3.5 Sonnet brings a moment for Computer Use On October 22, 2024, Anthropic released a brand new large model—Claude 3.5 Sonnet, which was praised by Anthropic's product manager Michael Gerstenhaber as the smartest model to date; and launched a new model Claude 3.5 Haiku. The most striking feature of Claude 3.5 Sonnet is its computer usage capability—through the corresponding API, computers can simulate human operations to complete tasks such as screen reading, moving the cursor, clicking keys, and entering text. According to the company's official website, the realization of this capability mainly relies on the following steps:

• Screen recognition: Screen reading capability remains an important prerequisite.
• Logical reasoning: Based on the recognized screen content, reasoning when and how to perform the corresponding operations.
• Positioning execution: Accurately calculating pixel positioning is very important: how many pixels need to be moved vertically or horizontally to correctly click the corresponding position.
• Rapid generalization and continuous iteration: The research team observed that the model generalizes very quickly and will self-correct and retry tasks when encountering difficulties.

In the OSWorld benchmark test assessing computer usage capability, Claude 3.5 Sonnet achieved an accuracy of 14.9%, reaching 22.0% with more steps provided, ranking first in the industry 1.2 Zhipu: AutoGLM Innovates Edge Interaction, Phone Use Becomes Reality AutoGLM is an innovative Agent tool recently launched by Zhipu, capable of receiving simple text or voice commands to simulate human operations on mobile phones and complete a series of complex tasks. For example, AutoGLM can easily purchase items from historical orders on Taobao, order takeout on Meituan, and buy train tickets on 12306, among other tasks. According to the paper by the Zhipu AutoGLM team, the realization of this powerful function relies on the following core steps:

• Multimodal Screen Reading: In traditional RPA (Robotic Process Automation), GUI (Graphical User Interface) recognition is performed through OCR, but it is difficult to generalize. Large multimodal models can perform fuzzy matching and long task planning, which is very helpful for understanding GUIs.
• Simulated Operation: Behavior Cloning is very important for Agent training; however, a fundamental issue is that the agent may only be imitating rather than understanding.
• Curriculum Learning: Agent tasks are usually complex, so progressive curriculum learning is very important.
• Reward Modeling (RM) and Reinforcement Learning (RL): Compared to past specific task training, agents now need to perform generalized tasks in an open world, requiring more generalized reward modeling and reinforcement learning.

The power of AutoGLM comes from its two core technological innovations:

• First, through Intermediate Interface Design, AutoGLM achieves decoupling of the "task planning" and "action execution" phases, making the agent more precise and flexible in task planning and action execution.
• Second, AutoGLM employs a "self-evolving online curriculum reinforcement learning framework" to train and enhance the capabilities of large model agents. This framework achieves continuous self-examination, motivation, and improvement through self-evolving learning strategies.

In terms of performance, AutoGLM has achieved significant improvements in Phone Use and Web Browser Use On the AndroidLab evaluation benchmark, its success rate reached 36.2%, significantly surpassing the levels of both GPT-4o and Claude-3.5-Sonnet. In the WebArena-Lite evaluation benchmark, it achieved approximately 200% performance improvement relative to GPT-4o. We see AutoGLM as a typical case of AI Agent, possessing broad application prospects and potential. We expect to see more Agents applied in more scenarios by 2025.

2. Advertising Technology

2.1 Applovin (APP.O): Recommendation Engine Axon Aids Precise Advertising As a global leading mobile marketing platform, Applovin helps advertisers reach over 1.4 billion daily active users across multiple channels, including in-app, mobile, and CTV, providing more possibilities for business growth and achieving substantial advertising revenue growth. Applovin's main business is divided into software platform and applications. The software platform is supported by the company's AI-based recommendation engine AXON, enabling advertisers to automate marketing, engagement, and monetization. The applications include over 200 free mobile games primarily focused on casual, match-three, and card genres. The application business carries lower risks and typically has more predictable revenue streams and returns. Studios developing and operating applications also utilize the software platform for marketing, scaling, and monetization.

In 2023, Applovin's total revenue was $3.28 billion, a year-on-year increase of 17%, with software platform and application revenues at $1.84 billion and $1.44 billion, respectively, growing by 75.6% and declining by 18.4%. Compared to a loss of $190 million in 2022, the company turned a profit in 2023, achieving a net profit of $360 million.

In Q3 2024, Applovin's total revenue was $1.2 billion, a year-on-year increase of 39%. Among them, software platform revenue grew to $840 million, a year-on-year increase of 66%; application revenue was $360 million, a year-on-year increase of 1%. Profit reached $430 million, with a staggering year-on-year growth of 298%, and an adjusted profit margin of 60%.

• Software Platform: In Q3 2024, software platform revenue accounted for 70% of total revenue. Revenue primarily comes from fees paid by advertisers who achieve growth and profitability using the software platform. Clients include independent developer studios and some of the largest global internet platforms, such as Meta and Google.
• Applications: Revenue sources are divided into in-app purchases (IAP) and in-app advertising (IAA). In Q3 2024, application revenue increased by $3.25 million compared to the same period last year Among them, IAP revenue decreased by $1 million, mainly due to a 5% decrease in the number of in-app purchases, while the average price increased by 4%. IAA revenue increased by $4.2 million, or 4%, mainly due to an 8% increase in ad impressions, but CPM decreased by 4%.

Since the launch of the AI advertising recommendation engine AXON, AppLovin's software platform revenue has rapidly grown. In 2023, software platform revenue increased by 76% year-on-year, and in Q3 2024, software platform revenue grew by 66%. The software platform's business is supported by the AI-based recommendation engine AXON, enabling advertisers to automate their marketing, engagement, and monetization in three core ways:

• Matching: The provision of marketing technology allows advertisers to reach more of the most suitable users through personalized content, thereby increasing the number of users downloading and engaging with their content.
• Pricing: Providing monetization and analytics technology to advertisers to maximize the value of their ad inventory by obtaining high prices per impression.
• Process Optimization: Offering developers a set of features to optimize applications and streamline operations.

AI empowers AppDiscovery for precise targeting. As the main revenue source of AppLovin's software platform, AppDiscovery is a software platform that helps mobile applications reach a large number of high-quality users, discover new incremental audiences, and promote growth, supported by AXON's predictive algorithms. AXON's algorithms enable precise user targeting, allowing advertisers to match their applications with users who are more likely to download them and provide users with more content they may be interested in.

AXON's upgrades enhance the company's monetization capabilities. The improvement in AXON's data processing capabilities further refines the matching between advertisers, platforms, and users, thereby enhancing ad effectiveness and driving increased spending from existing advertisers, further propelling AppLovin's growth.

The company's scale effects feed back into AI efficiency improvements. From early 2018 to the end of 2022, the company invested nearly $4 billion, made 29 strategic acquisitions, and established partnerships with mobile app developers. The company's official website states that AppLovin has over 1.4 billion daily active users. The expansion of the company's scale and reach benefits from the compounded improvements to the AI advertising engine AXON, which in turn enhances the efficiency and growth of software platform solutions.

Through the AI-driven advertising recommendation engine AXON, AppLovin not only enhances advertisers' monetization capabilities but also optimizes user experience, leading to business growth. AppLovin is expected to continue leveraging its technological advantages, vast global user base, and comprehensive product ecosystem to further consolidate its leadership position in the global mobile marketing platform, creating more value for advertisers.

2.2 Mobvista (1860.HK): Upgrading Bidding Model for Programmatic Advertising Platform Mintegral Mobvista is a technology company dedicated to providing advertising technology services and marketing technology services necessary for the development of the mobile internet ecosystem for global clients, with a business model that includes providing advertising placement services and matrix-style marketing SaaS tool products The advertising technology segment is divided into:

• Programmatic advertising technology platform. As a globally leading AI-driven programmatic interactive advertising platform, Mintegral provides advertisers with one-stop programmatic advertising placement and traffic monetization services by linking a large number of mid to long-tail app traffic.

• Non-programmatic advertising business platform. After the advertising technology platform helps enterprise clients achieve user growth and commercial monetization, marketing technology is needed to help enterprise clients activate the accumulated data and optimize marketing plans during this process, thereby achieving high-quality growth.

In 2023, the total revenue of HuiLiang Technology was USD 1.054 billion, a year-on-year increase of 17.9% (excluding the impact of top media agency business, it increased by 19.1% year-on-year), of which the revenue from advertising technology business was USD 1.038 billion, a year-on-year increase of 17.8%, accounting for 98.6% of total revenue, and the revenue from marketing technology business was USD 15.6 million, a year-on-year increase of 23.9%. The adjusted profit for 2023 was USD 19.12 million, a year-on-year increase of 97.1%.

In Q3 2024, HuiLiang Technology's total revenue was USD 420 million, a year-on-year increase of 54.6%. Among them, the revenue from advertising technology business was USD 410 million, a year-on-year increase of 55.4%, accounting for 99% of total revenue; the revenue from marketing technology business was USD 4.002 million, a year-on-year increase of 1%. The adjusted profit for this quarter was USD 34.27 million, a year-on-year increase of 45.3%.

• Advertising technology segment: Net income accounts for over 90% of overall net income. The Mintegral platform achieved simultaneous growth in revenue and profit in Q3 2024 and has already achieved control over profit margins across various verticals. During the reporting period, the revenue growth of the light game category was lower than the overall growth of Mintegral, mainly due to the industry's gradual shift towards a mixed monetization model of advertising and in-app purchases, which somewhat affected the growth rate of light game revenue within the game category. In Q3 2024, non-game category revenue was USD 110 million, a year-on-year increase of 98.5%. Thanks to the support of the intelligent bidding system, Mintegral has also made new breakthroughs in other verticals such as mid to heavy games, e-commerce, and tools.

• Marketing technology segment: Still in the product refinement stage, with a year-on-year revenue growth of 23.9%, overall losses further narrowed.

We see that with the help of AI tools, programmatic advertising placement has achieved more precise matching, smarter bidding, and more automated material generation.

• Machine learning and intelligent precise placement: To provide advertisers with intelligent and precise advertising placement solutions, machine learning is needed to analyze a large amount of advertising data. Mintegral fully utilizes the powerful algorithms of machine learning to achieve personalized and intelligent advertising recommendations through deep learning of user behavior and predictive modeling. This not only improves the effectiveness of advertising placement but also creates a more attractive advertising experience for advertisers

• AI Optimizes Advertising Technology Product Bidding Model: Mintegral officially launched its smart bidding feature (Target ROAS) in May 2023. By enabling full-channel data feedback on Mintegral, advertisers can set in-app advertising ROAS targets on their self-service platform. After the product launch, it significantly boosted the platform's business revenue: by the end of 2023, the daily revenue contribution approached USD 1.5 million, and Mintegral's revenue increased by 18.8% year-on-year in 2023. Additionally, the company will launch the Target CPE (cost per engagement) feature in July 2024, optimized for in-app events (engagement), to help advertisers effectively control advertising costs and optimize ad performance.

• AI Restructures Marketing Technology Product Services: To assist clients in quickly creating highly attractive advertising creatives (such as introducing advanced image generation and processing models into the system) and creating efficient advertising campaigns, the platform can utilize LLM/AIGC to restructure relevant services and continuously iterate.

In the wave of globalization and digitalization, HuiLiang Technology will embrace the AI trend, helping more enterprises achieve improved advertising effectiveness and cost control through more data insights and model iterations: while helping enterprises gain more growth engines, it will also enhance its own monetization.

3. Corporate Solution Decision-Making

3.1 Palantir (PLTR.O): Large Model Empowering AIP Platform Capabilities, Accelerating Expansion of Commercial Clients

Palantir redefines the data-driven intelligent platform for modern government and enterprise services. Palantir's main software systems and platforms include:

• Gotham is an intelligence and defense tool platform for government and security agencies, capable of processing and analyzing complex intelligence data;

• Foundry is a seamless data management and operational platform for commercial clients, helping enterprises make real-time decisions;

• AIP is a real-time, AI-driven decision support solution that provides in-depth data analysis;

• Apollo is an automated deployment platform and operational tool that assists clients with software deployment.

Palantir was officially included in the S&P 500 index in September 2024.

The company's profit growth exceeded 103% in Q3 2024. For FY2023, the company's revenue was USD 2.225 billion, up 17% year-on-year; of which, government business revenue was USD 1.22 billion, up 14% year-on-year; and commercial business revenue was USD 1.0 billion, up 20% year-on-year. FY2023 marked the company's first annual profit, with a net profit of USD 220 million, compared to a net loss of USD 370 million in FY22.

In Q3 2024, the company's revenue was USD 726 million, up 30% year-on-year; of which, government business revenue was USD 408 million, up 33% year-on-year; and commercial business revenue was USD 317 million, up 27% year-on-year In Q3 2024, the company's commercial business revenue of USD 317 million included USD 179 million from domestic commercial operations in the United States, with a year-on-year growth rate exceeding 54%. The company's net profit for Q3 2024 was USD 149 million, up 103% year-on-year. In Q3 2024, the number of customers increased by a net of 36, of which 31 were commercial clients and 5 were from government agencies; thus, the total number of commercial clients reached 498, up 51% year-on-year.

• To G: High customization leads to high customer stickiness, continuing to maintain stable profitability. In the field of cooperation with government entities, the company adopts a highly customized model. Although the costs are higher, the high barriers to entry in this field (involving issues such as data regulation and data leakage) result in greater customer stickiness. As cooperation with government-related agencies deepens, the company's government business will become more solid. In 2023, benefiting from in-depth cooperation with the U.S. Army, the year-on-year growth rate of government business reached 14%, with a gross profit margin of 81% during the same period. In 2024, the company successfully secured two important orders from the U.S. Army again—providing systems for the TITAN "ground station" and expanding the MAVEN intelligent system, with order amounts reaching USD 178 million and USD 480 million, respectively. As of Q3 2024, the company's to G business still accounted for over 56%, and this business will continue to provide stable revenue for the company in the future.

• To B: AIP paves the way for serving commercial clients, and Bootcamp accelerates the expansion of commercial clients. AIP can assist enterprises in establishing and optimizing AI-driven data flow systems, AI-driven logic construction, workflow visualization, and workflow monitoring, among others. In terms of promotion channels, in addition to building direct sales and distribution channels, the company also provides trial opportunities for enterprises through Bootcamp to complete rapid system implementation, addressing the key issue of high barriers to promotion and accelerating the rapid promotion of AIP.

The transformation brought by AIP has led to rapid growth in the U.S. commercial business. As of Q3 2024, the number of the company's U.S. commercial clients reached 321, a year-on-year increase of 77%. This figure was 181 in Q3 2023. Currently, according to the official website, AIP has served commercial clients in various industries, including aerospace, automotive, infrastructure engineering, energy, banking services, government security, healthcare and life sciences, insurance, logistics, industrial manufacturing, media communications, retail, and chemicals. As of Q3 2024, U.S. commercial business revenue increased by 54% year-on-year, while the remaining deal value for U.S. commercial operations increased by 73% year-on-year.

The successful application of AIP has brought significant growth to Palantir and its clients. We believe that AI will play a greater role in enterprise solution decision support in the future.

3.2 Fourth Paradigm (6682.HK): Prophet AIOS Drives AI Implementation Across Various Industries Fourth Paradigm is a leading provider of artificial intelligence technology and solutions in China, offering platform-centered decision-making AI solutions to help enterprises achieve intelligent transformation.

In the first three quarters of 2024, Fourth Paradigm's total revenue was RMB 3.218 billion, a year-on-year increase of 26.1%. The gross profit was RMB 1.350 billion, a year-on-year increase of 12.5%. The gross profit margin was 42.0%, remaining stable overall. The total number of users was 224, with 98 benchmark users. The average revenue contribution from the benchmark user group was RMB 14.87 million:

• Prophet AI Platform: Revenue in the first three quarters of 2024 was RMB 2.171 billion, a year-on-year increase of 50.2%, accounting for 67.5% of total revenue.

• SHIFT Intelligent Solutions: Revenue in the first three quarters of 2024 was RMB 688 million, a year-on-year decrease of 14.1%, accounting for 21.4% of total revenue.

• Shishuo AIGS Service: Revenue in the first three quarters of 2024 was RMB 359 million, a year-on-year increase of 17.0%, accounting for 11.1% of total revenue.

In terms of AI research and development, Fourth Paradigm launched its core product, the industry large model platform "Prophet AIOS 5.0," in March 2024. The working principle of Prophet AIOS 5.0 can be understood as "Predict the Next 'X'," which means constructing an industry foundational large model based on X-modal data from various industry scenarios.

For example, in the field of health management, if a large amount of professional medical and health-related data is fed into Prophet AIOS 5.0, it can build a health management large model based on this data. This model can predict a user's future health examination report data based on their past examination report data and generate personalized health management recommendations. Similarly, in areas such as water and electricity management, water services, and acoustics, Prophet AIOS 5.0 can also construct corresponding large models based on historical data from their respective fields and predict future data changes.

As of Q3 2024, Fourth Paradigm has created over 40 AI products through the "Prophet Inside Model," including intelligent mouse solutions, intelligent meeting solutions, vGPU, AI Data Foundry, large model toolchains, integrated platform TianShu for search and promotion, large model management platform Model Hub, intelligent question answering, AI digital human video synthesis platform, document digitization management platform, AI+5G video marketing, AI quality inspection systems, 3D scanning and modeling equipment, etc. Additionally, the company disclosed that it has also integrated AI Agent capabilities into several intelligent hardware solutions such as AIGS Builder and AIGS CodeX Empowering products with more complex reasoning and execution capabilities. We look forward to the Fourth Paradigm achieving higher growth through its outstanding hardware and software AI capabilities across more industry tracks.

4. Customer Relationship Management (CRM)

4.1 Salesforce (CRM.N): Agentforce empowers intelligent interaction and automation

Salesforce is a leading enterprise cloud computing platform dedicated to providing comprehensive customer relationship management (CRM) solutions for businesses. The company offers customized solutions in sales, service, marketing, and business intelligence through its powerful CRM platform, helping businesses enhance customer experience and operational efficiency. Salesforce also provides rich data analysis and application development tools to help businesses achieve data-driven decision optimization, driving business innovation and growth.

The company's business is mainly divided into two segments:

1. Subscription and Support: Salesforce's core business, primarily providing cloud-based CRM software subscription services, as well as related technical support and maintenance services.

2. Professional Services and Others: This segment includes consulting services, custom development, training, and other value-added services provided by Salesforce.

The company achieved strong performance growth in Q2 2025. Salesforce's Q2 2025 revenue was $9.33 billion, a year-on-year increase of 8%. Among them, subscription and support revenue was $8.76 billion, a year-on-year increase of 9%; professional services & other revenue was $560 million, with a year-on-year change of -6%. The company's net profit for the quarter was $1.429 billion, a year-on-year increase of 12.8%. The subscription and support business can be divided into five parts: Sales Cloud, Service Cloud, Platform & Others, Marketing and Commerce Cloud, Integration & Analytics, with revenues of $2.071 billion / $2.257 billion / $1.786 billion / $1.308 billion / $1.342 billion for the quarter, representing year-on-year growth of 9% / 10% / 9% / 6% / 13%, respectively.

Agentforce reshapes the service experience. Agentforce is Salesforce's latest AI agent system, which enhances employee productivity and customer satisfaction by automating tasks in service, sales, marketing, and commerce. The Agentforce Reasoning Engine aims to enhance user interaction through multi-turn dialogue, topic classification, commands and operations, and knowledge retrieval, providing more precise and effective services.

Agentforce will demonstrate tremendous practical application potential. Many industries are beginning to recognize that incorporating artificial intelligence into business processes can not only improve productivity but also enhance customer experience. We can see in numerous cases that Agentforce has shown great practical application potential, for example:

• Publisher Wiley has chosen to adopt Agentforce and Service Cloud with Einstein AI to address the surge in service requests from students using educational resources at the start of the new semester. Agentforce's AI-powered agents enable customers to resolve common issues on their own, increasing the case resolution rate by over 40% in the initial weeks. Meanwhile, AI productivity tools have accelerated the onboarding of seasonal customer service staff by 50%, achieving a 213% return on investment;

• Luxury department store chain Saks utilizes Agentforce for order status management and return inquiries, effectively providing personalized services and strengthening customer relationships, enhancing the luxury shopping experience;

• Group meal ordering service company ezCater leverages Agentforce to transform customer experience, automate customer service, and streamline the food ordering process.

• These successful cases fully demonstrate Agentforce's exceptional ability to assist enterprises in leveraging AI capabilities for business upgrades. We believe that subsequent AI capabilities can empower enterprises to achieve more intelligent interactions in the CRM field.

4.2 Weimob Group (2013.HK): Fully Embracing the Era of Agentic AI

Weimob is a leading provider of cloud-based business and marketing solutions in China. The company 1) offers subscription solutions to merchants in e-commerce and smart retail sectors, providing SaaS services, and 2) provides precise marketing services to advertising clients through merchant solutions.

In 2023, Weimob's revenue was approximately 2.23 billion yuan, a year-on-year increase of 21%. Among them, the revenue from subscription solutions and merchant solutions was 1.35 billion yuan and 880 million yuan, with year-on-year growth of 4.5% and 60.5%, respectively. The adjusted net loss attributable to the parent company was 390 million yuan, significantly narrowing from a loss of 1.46 billion yuan in 2022.

• In terms of subscription solutions, as of 2023, Weimob had 96,000 paid SaaS merchants, a year-on-year decrease of 3.3%; ARPU was 14,007 yuan, an increase of 8%, mainly benefiting from the promotion of large customer accounts. Among them, smart retail grew by 19.5% to 613 million yuan, and the organic growth rate reached 32% after excluding Haiding, with the proportion of subscription solutions increasing to 45.5%. The number of brand merchants increased to 1,306, with an average order revenue of 195,000 yuan per merchant. The implementation of Weimob's large customer strategy and industry strategy has effectively strengthened merchant stickiness and ARPU, enhancing future monetization capabilities. We expect the revenue from subscription solutions to maintain steady growth in 2024.

• In terms of merchant solutions, Weimob's gross revenue from precise marketing in 2023 was 14.5 billion yuan, a year-on-year increase of 44.5%. This strong growth was mainly due to a robust recovery in advertising demand, growth in video account advertising, and new channel expansion. Additionally, in terms of new channels, the annual consumption on Kuaishou and Xiaohongshu increased by 164% and 259%, respectively In the field of AI exploration, Weimob Group began laying out AI applications in October 2022; in May 2023, it officially launched the AI application product WAI based on large models. Currently, Weimob has gradually formed four major dimensions of AI layout in the AI field, comprehensively creating an AI Agent product matrix:

• AI + SaaS: Helping SaaS merchants achieve intelligent operations

• AI + Marketing: AI-driven smart marketing to reduce costs and increase efficiency

• To B: Providing personalized AI solutions

• To C: Offering AI tools for small and micro enterprises and designers.

Taking AI + SaaS as an example, Weimob Group has 15 major AI Agents, covering more than 58 real business application scenarios. Weimob Group stated that AI Agent products have effectively driven customer activity and commercialization of core SaaS businesses:

• New customer delivery: After the Agent team went live, the usage during the new customer delivery period increased by over 217%.

• Existing customer activity: After the Agent team went live, the monthly active customers using AI significantly increased by 74%, with a renewal rate of 75% for small and medium-sized merchants.

• Acceleration of commercialization: Compared to Q3 and Q2 of this year, WAI's commercialization revenue increased by 237% quarter-on-quarter.

In Weimob Group's future blueprint, all SaaS should be rewritten with AI; the era of Agentic AI has just begun. Weimob Group will not only reduce costs and increase efficiency internally through AI but also assist enterprise clients in their intelligent advancement and help individual clients evolve into super individuals through AI tools.

5. Real-time Interaction

5.1 Agora, Inc (API.O): RTE-PaaS leader, leveraging OpenAI to expand AI advantages

Agora, Inc. is a leading company in the RTE-PaaS (Real-Time Interactive PaaS Services) sector, empowering developers to build various real-time audio and video interaction scenarios within applications through APIs. Its business in China is known as Shengwang, while its overseas business is Agora.

In 2018, the company launched the world's first real-time interaction quality diagnostic analysis tool "Crystal Ball Analytics." By 2019, the company's software-defined real-time network SD-RTN™ had over 200 data centers worldwide. In 2020, Agora, Inc. was listed on NASDAQ. As of December 31, 2023, the number of registered applications worldwide exceeded 742,000.

The company's total revenue in 2023 was approximately $142 million, a year-on-year decrease of 11.9% compared to 2022, with Agora's total revenue at $61 million and the revenue from the voice network segment at $80 million. In Q4 2023, the company achieved an adjusted net profit of $1.4 million, marking its first single-quarter profit in over three years.

In Q2 2024, the company achieved total revenue of approximately $34 million, a year-on-year increase of 0.5%, with Agora generating revenue of $15.6 million and the voice network segment generating revenue of 132 million RMB. The adjusted net loss for the second quarter of 2024 was $6 million, a reduction from the $6.6 million loss in the same period last year. The company's management stated during the Q2 2024 earnings conference call that they expect to achieve a full-year GAAP net profit turnaround in 2025.

As of June 30, 2024, the company had 5,446 active customers, with 3,774 from the voice network segment and 1,672 from the Agora segment, maintaining a high level. As of June 30, 2024, the company's cash and cash equivalents amounted to $371 million, and the company completed a cumulative stock repurchase of approximately $110 million.

In terms of AI exploration:

• The company launched the Conversational AI SDK, integrating with OpenAI's real-time API to power the development of natural voice-driven AI experiences. On October 1, 2024, OpenAI officially released the low-latency multimodal API interface for GPT-4o, called the Realtime API. Agora became one of three global partners recommended by OpenAI, integrating OpenAI's Realtime API into the Conversational AI SDK, providing developers with low-latency, high-availability, and excellent audio quality transmission services, while also offering echo cancellation and noise reduction features on the device side. Currently, OpenAI's Realtime API pricing remains high, but it is expected that significant price reductions in the future will lead to large-scale promotion and increased usage scenarios.

• The company released a panoramic view of RTE+AI capabilities, guiding the future direction of RTE. At the RTE conference in October 2024, the voice network officially released the panoramic view of RTE+AI capabilities The company envisions future RTE track solutions empowered by AI from five dimensions: real-time AI infrastructure, RTE + AI ecological capabilities, Agora AI Agent, real-time multimodal conversational AI solutions, and RTE + AI application scenarios.

1. Real-time AI infrastructure: A real-time AI infrastructure that ensures a wide range, low latency, excellent quality, and strong availability, capable of handling tens of millions of concurrent experiences - Network Infrastructure SD-RTNTM.

2. RTE + AI ecology: An RTE + AI ecology that provides a comprehensive and excellent experience at the levels of text, visuals, sound, and large models.

3. AI Agent: Agora AI Agent offers powerful orchestration capabilities and customization features, with ultra-low development thresholds and costs, high scalability, and comprehensive component capabilities.

4. Real-time multimodal conversational AI solutions: Real-time multimodal conversational AI solutions that provide ultra-fast responses, highly human-like interactions, and high-definition conversational experiences.

5. RTE + AI application scenarios: Multi-scenario application innovations achieved through GEN AI + RTE, covering five areas: social entertainment, education, IoT, healthcare, and digitalization.

Agora, Inc.'s development model, cooperative ecosystem, technological advantages, and market positioning have laid a stronger foundation for its long-term growth potential. The collaboration between Agora, Inc. and OpenAI not only enhances the underlying capabilities and overall performance of its products but also expands the range of effective services, thereby better meeting the continuously evolving market demands in the thriving AI era. With the rise of AI models and the ongoing deepening of related collaborations, Agora, Inc. is expected to continue its growth in the field of real-time interactive services, while its strong financial resilience also provides robust support for further business expansion in the future.

Author of this article: Xia Jun S0680519100004, Source: GS Overseas Research, Original Title: "Finding the 2025 Hottest AI Applications | Guosheng Overseas"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk