(Report producer/author: CITIC Securities)
Investment analysis
With ChatGPT strongly "out of the loop" around the world, the AI big model business model has successfully run through, and we believe that the era of AIGC is coming. In recent years, with the continuous optimization of AI large models and computing power costs, the continuous decline in training and inference costs has provided prerequisites for the commercial application of AIGC. AIGC has derived a rich capability matrix, with three core capabilities of twin, editing and creation, and in the long run, AIGC has the application prospect of subversive cost reduction and efficiency improvement in the whole industry.
According to GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (2023) (by Tyna Eloundou et al.), AIGC may affect more than 80% of workers. At the same time, AIGC can also catalyze new industrial opportunities in frontier fields, and AIGC can greatly accelerate the creation of digital content and the development of the digital human industry, and promote the accelerated landing of the metaverse industry. Considering that AIGC needs strong hardware support and broad application space in the whole industry, we recommend three main investment lines: computing power hardware support, promoting subversive improvement of industry production efficiency, and giving birth to a new ecology for metaverse development.
The first main line is that AIGC computing power hardware supports the industrial chain, and the computing power revolution brought by AI large models will drive the development of computing power hardware to support the industrial chain, and the performance of derivative hardware is expected to continue to improve; The second main line is that AIGC promotes the subversive improvement of industry production efficiency. In the short term, AI already has the ability to assist humans to complete some work tasks, and we believe that AIGC will be the first to be applied in software development, daily office, film and television entertainment, education, e-commerce and other fields. In the long run, as AI gradually becomes more capable of professional creativity comparable to humans, AIGC will bring disruptive production methods across the industry, and we expect that related companies in entertainment, media and software development will be the first to benefit from the cost side. The third main line is that AIGC has spawned a new ecology for the development of the metaverse. Game companies are expected to take the lead in applying AI models in content creation and game setting, and occupy a favorable competitive position in the metaverse industry.
AIGC quietly rises, computing power reduces costs and consolidates the "AI foundation"
AIGC quietly rises, ChatGPT is "out of the loop"
As more and more artificial intelligence is used in content creation, the concept of AIGC is quietly emerging. AIGC (AI Generated Content) refers to the use of AI (Artificial Intelligence) technology to automatically generate content that matches user needs. Simply enter the requirements and AIGC helps creators automatically generate the content they need, allowing creators to spend more time on topics and less time actually creating, increasing productivity and quality. Since the 50s of the 20th century, deep learning algorithms and device computing power have developed rapidly, and AI research has made great progress. AI is not only able to interact with humans, but also do creative work such as writing, arranging, drawing, video production, and more. The development of AIGC can be roughly divided into three stages: early germination, accumulation and rapid development, and has entered the rapid development stage.
ChatGPT has a wide range of applications and is expected to take the lead in the field of AIGC. ChatGPT (Chat Generative Pre-trained Transformer) is a chatbot developed for OpenAI, which is built on the GPT-3 large language model developed by OpenAI and fine-tuned using supervised learning and reinforcement learning (human supervision) techniques. In the field of AIGC, ChatGPT can independently create high-quality content according to user needs, which not only lowers the threshold of creation, but also greatly improves the efficiency of user creation. The decisive factors for AIGC's product capabilities are interaction, quantity and quality, with ChatGPT significantly improving its ability to generate and understand, and GPT-4 greatly optimizing content quality by driving content formats from single to diverse.
ChatGPT exceeded 100 million MAUs in just two months, making it the fastest growing consumer app in history. In November 2022, OpenAI launched ChatGPT, which became the "top stream" in the AI industry once released. According to SimilarWeb data, ChatGPT has reached 100 million MAU (Monthly Active User) just two months after its launch, making it the fastest growing consumer app in history. According to Sensor Tower data, TikTok took 9 months to reach 100 million MAUs, and Instagram took 2 and a half years.
The scale of global intelligent computing power is growing rapidly, and training costs are expected to continue to be optimized
The scale of global intelligent computing power is developing rapidly, and the scale of China's intelligent computing power is expected to increase to 1271.4EFLOPS in 2026. According to Huawei's predictions, in the future, AI will move from perception to cognition, creativity will continue to increase, AI will enter daily life and empower all things with intelligence, and the demand for computing power will grow rapidly, and the global intelligent computing power is expected to reach 105ZFLOPS (10 21 floating point calculations per second) in 2030, an increase of 500 times compared with 2020. At present, domestic leading technology companies are making efforts to lay out AI models, relying on the construction of high computing power systems, and it is expected that the scale of domestic intelligent computing power will maintain a high increase. According to IDC data, the scale of China's intelligent computing power will reach 268.0EFLOPS (10 18 floating-point operations per second) in 2022, and it is expected that the scale of intelligent computing power will enter the ZFLOPS level in 2026, reaching 1271.4EFLOPS.
AI large model parameters are growing exponentially, and a new "Moore's Law" is about to emerge. In recent years, with the rapid development of computing power and data sets, NLP (Natural Language Processing) algorithms based on the Transformer model have developed rapidly. AI large models have "emergence capability", and when the training volume exceeds a certain threshold, the model will unlock the "emergence capability", that is, the model accuracy will suddenly increase. As a result, the number of parameters in state-of-the-art NLP and CV (Computer Vision) models continues to grow and has grown exponentially in recent years.
According to AI and Memory Wall (2021) (by Amir Gholami et al.), the computing power required by AI models in CV, NLP, and speech learning has increased 15x every two years over the past decade, while Transformer models have grown even faster, growing 750x every two years. OpenAI expects the computing resources required for AI research to double every 3-4 months. Sam Altman, CEO of OpenAI, also tweeted in February 2023 that "a new version of Moore's Law is coming soon, and intelligence in the universe will double every 18 months."
NVIDIA data center GPU AI inference capabilities continue to increase, and the cost per unit of computing power continues to be optimized. In recent years, NVIDIA data center GPUs have been iterating for an average of 2-3 years, and the GPUs currently used in AI large model training include V100, A100 and H100. From the performance perspective, NVIDIA allows multiple networks to run on a single GPU at the same time through multi-instance GPU (MIG) technology to maximize the utilization of computing resources; And by optimizing the GPU architecture and instructions, the training speed of large models is improved. According to NVIDIA's official website information, the GPU AI inference capability of NVIDIA data center has been greatly improved, and the AI inference throughput of H100 is 30 times higher than that of the A100 ultra-large model. At the same time, GPU prices have risen, but at a much lower rate than hash power. According to Zhongguancun online quotation and NetEase technology information, the price of NVIDIA H100 is more than 240,000 yuan, which is about 3 times that of A100. Therefore, we believe that as the GPU architecture continues to be optimized with system design, the cost per unit of computing power will become a long-term trend.
Refer to The Economics of Large Language Models (2023) (by SUNYAN) and Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM (2021) (by Deepak Narayanan et al.), The researchers have proposed a cost estimation model based on the number of training words (Tokens), large mode parameter quantity (Parameters), chip computing power (FLOPS), chip price and other indicators as parameters. We will calculate the training and inference costs of GPT-3 and the cost optimization brought by NVIDIA chip iteration according to the relevant models.
Model and hardware optimization is expected to save more than 80% of costs, and algorithms and datasets are expected to be the core of the competition. According to estimates from The Economics of Large Language Models (2023) (by SUNYAN), the authors estimate GPT-3 training costs around $1.4 million and inference costs $0.0035/1000 tokens. We believe that benefiting from the optimization of large model parameters, the gradual improvement of computing power utilization, and the continuous decline in unit computing power cost with chip iteration, the cost of AI large model training and inference is expected to continue to decrease, and it is expected to save more than 80%. According to Training Compute-Optimal Large Language Models (2022) (by Jordan Hoffmann et al.), DeepMind believes that the effect of expanding the number of model parameters may be marginally decreasing, and we believe that high-quality datasets and high-quality algorithms are expected to become the core competitiveness of AI large models.
The commercial use of AI large models is imminent, and AIGC liberates productivity
Multimodal AI large models are developing rapidly, and AIGC has been implemented in multiple fields
Transformer models bloom. Since 2017, Transformer has brought significant performance improvements to the CV and NLP fields, setting new records in object detection and semantic segmentation tasks, and CV and NLP are expected to be unified under the Transformer structure. According to TRANSFORMER MODELS: AN INTRODUCTION AND CATALOG (2023) (author: Xavier Amatriain), more than 60 AI large models have been developed based on Transformer, including the famous GPT series models.
Pre-training is trending towards "unification", and the Transformer architecture extends to multimodal scenarios. In recent years, pre-training in areas such as NLP, CV and multimodality has begun to show a trend of big convergence. In 2022, Microsoft Research Asia launched the BEiT-3 pre-training model, which achieved SOTA (state-of-the-art, best/most advanced) migration performance in tasks such as object detection, instance segmentation, semantic segmentation, visual reasoning, and image description generation. Large-scale pre-training on large amounts of data makes it easier to migrate models to multiple application areas, and Microsoft Research Asia believes that the trend of unification has gradually emerged in three aspects: backbone, pre-training tasks and scale improvement. We believe that AI big models based on Transformer architecture will continue to develop to multi-modality, promoting the all-round application of AIGC in the whole industry.
GPT-4 ushered in a huge performance upgrade, and OpenAI took the lead in AIGC development. According to the GPT-4 Technical Report (2023) (OpenAI), GPT-4 can accept image input and has the ability to interpret images "like humans"; GPT-4 achieved 5 out of 60% of AP exam subjects, an increase of more than 30% over GPT-3.5; GPT-4 also achieved a score of 339+4 in the Graduate Entrance Examination GRE, surpassing 95% of test takers. The introduction of multimodal in GPT-4 lays a solid foundation for subsequent generation of audio, pictures, and videos. With the help of Microsoft's full ecosystem and ChatGPT's global "out of the loop" performance, we believe that OpenAI will accelerate iteration in the field of AIGC, continue to improve model content generation and logical reasoning capabilities, and take the lead in the development of AIGC.
AIGC is gradually landing in many fields, and AI is expected to have professional-level creative capabilities in the long run. Benefiting from the rapid development of AI large models, AI is gradually evolving from content generation to content creation. From the perspective of application fields, AIGC can be applied to text, images, music, video, 3D modeling, architecture and other fields, and can play an auxiliary or even creator role in office, media, art and other scenes. From the perspective of industrialization, AIGC has developed rapidly in the field of text and code, and now has the ability of long text writing and basic software development, which can assist white-collar workers and technicians to complete some of the work; In the field of artistic creation, AIGC is still in the initial exploration stage, and there is still much room for improvement in creative ability. We believe that with the continuous enhancement of multimodal AI large model capabilities and the continuous optimization of algorithms for images and videos, AIGC is expected to have creative capabilities that surpass professionals in the field of artistic creation.
AIGC derives a rich capability matrix to promote the whole industry to reduce costs and increase efficiency
AIGC includes three core competencies, which have derived a rich capability matrix. AI models have mature commercial applications in single-modal domains such as CV and NLP. In recent years, the development of multimodal AI models has accelerated, Transformers have developed into a large "family", and multimodal commercial applications are maturing. Referring to the research of the China Academy of Information and Communications Technology, we believe that AIGC mainly includes three core capabilities: digital twin capabilities, digital editing capabilities, and digital content creation capabilities. The three core competencies mean moving real-world content to the digital world (twin capabilities), establishing the interoperability of digital and real-world content and assisting real-world content generation (editing capabilities), and eventually evolving from digital imitation to human-like reality creation capabilities (creative capabilities). Based on the three core capabilities, AIGC has derived a rich capability matrix, according to the information of JD Exploration Institute, AIGC is developing a capability matrix from recognition to generation in the fields of text, speech and image and video.
AIGC has broad application prospects, and we believe that it will be implemented quickly in daily office, media, film and television entertainment, e-commerce and other scenarios. Stable Diffusion and GPT-4 respectively let the public feel the creative ability of AIGC close to humans in the field of images and text, and major technology giants have increased the code of AI large models. Starting from 2022, platform technology giants such as Google, Microsoft, Meta, Amazon, Baidu, Alibaba, and Tencent have increased their efforts to deploy AIGC, and the speed of integration with existing business models is expected to accelerate. At present, the high degree of digitalization in the fields of media, film and television entertainment, and e-commerce provides a good soil for AIGC, and AIGC is expected to take the lead in landing in related industries. We believe that with AI's continuous evolution of text, image, and video content generation capabilities, it can help related industries significantly reduce costs and increase efficiency, and the market potential is huge.
AIGC+ Text & Audio: Empower AI customer service to reduce costs and increase efficiency. AI customer service has become popular around the world, iterating from text conversations to voice conversations. Benefiting from the rapid development of AIGC, AI customer service reasoning capabilities continue to increase, and logical reasoning can be carried out and appropriate answers can be given on the basis of fully understanding customer demands. According to IBM Institute for Business Value research, using virtual customer service technology can save $5.50 per customer conversation. At the same time, companies with leading AI customer service technology are more satisfied: customer and customer service employee satisfaction has increased by 12% and 9%, respectively, and processing time has been reduced by 15%. We believe that AI has a high maturity in the generation capacity of text and audio, and the commercialization process is fast, and AI customer service is expected to penetrate the whole industry. According to GrandView Research, the global AI customer service market was valued at $1.38 billion in 2022 and is expected to reach $7.08 billion by 2030, corresponding to a CAGR of 22% from 2022 to 2030.
AIGC+ Image and Video: Promote the film and television entertainment industry to reduce costs and increase efficiency. In the field of film and television entertainment, AI has the ability to create images, videos, and 3D modeling. In the field of images, according to 6pen forecasts, if 10%-30% of image content is generated by AI in the next five years, the market size is expected to exceed 60 billion yuan. In the video field, in January 2023, Netflix JP Japan, WIT Studio Japan, and Rinna, the Japanese division of Microsoft Xiaoice, jointly produced the animation "Dog and Boy", becoming the first commercial animation in history to use AI to generate a background. Excluding characters and animal characters, most of the drawing work is done by AI.
In the field of 3D modeling, Tencent AI Lab demonstrated the process of using AI to quickly build a 3D virtual city from scratch, with an area of 25 square kilometers, including 130 kilometers of road network, 4,416 buildings, and more than 380,000 indoor maps. According to Tencent's AI Lab release at the 2023 Game Developers Conference, modeling such a large city used to take multiple artists to complete in years, while combining AI took only weeks. We believe that AI has the ability to assist professional creators in the field of animation and games to complete some of their creations, greatly improving creative efficiency and reducing labor costs, thereby promoting the entire film and television entertainment industry to reduce costs and increase efficiency.
AIGC may have an impact on 80% of the workforce, China by 2030
The AIGC market is expected to exceed one trillion yuan of AIGC or affect 10% of the work tasks of 80% of the workforce, and greatly improve the work efficiency of workers. AI large models have certain creative capabilities in text, images, videos, etc., and can assist or even replace human work tasks in non-manual operations and interpersonal communication workplaces. According to GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (2023) (by Tyna Eloundou et al.), about 15% of the work tasks in an average occupation in the United States will be affected by AI large models. About 80% of U.S. workers may have at least 10% of their work tasks affected by AI large models, and about 19% of workers may have at least 50% of their work tasks affected. According to Xinhua, the latest research by Stanford University and MIT on a technology company shows that AI can improve the average labor productivity of technical support employees by 14%, and the work speed of "novice and low-skilled employees" can increase by 35%.
The AIGC industry is about to mature, and the size of China's AIGC market is expected to exceed one trillion yuan in 2030. According to research by OpenAI and the University of Pennsylvania, as the reasoning ability of AI large models continues to improve, even if development is stopped, its influence will continue to expand, and a global AIGC investment boom has been set off, and the commercial application of AIGC in various industries is accelerating. According to Gartner's forecast, AIGC is expected to enter an industry maturity period in the next 2-5 years. We believe that as domestic enterprises accelerate the introduction of AIGC in R&D, production and sales to reduce costs and increase efficiency, the domestic AIGC industry will also usher in a period of rapid development. According to the prediction of Qubit Think Tank, China's AIGC industry is in the cultivation and exploration period from 2023 to 2025, and the average annual compound growth rate is expected to be 25%; From 2026 to 2030, the industry will usher in a stage of rapid growth, and the scale of the Chinese market is expected to reach 1,149.1 billion yuan in 2030.
Imagination is productivity, and AIGC builds a bridge to the metaverse
Imagination is productivity, and AIGC revolutionizes content production
AIGC will go through three stages of development, and the era of human-machine collaboration is coming. At present, content creation has shifted from PGC (Professional Generated Content) to UGC (User Generated Content), and AI-assisted content generation (AIUGC) and AIGC are also becoming popular. According to the judgment of Baidu CEO Robin Li, AIGC will go through three stages of development: the first stage, known as the "assistant stage" of AIGC, AIGC is used to assist humans in content production, such as the production of audiobooks, assistance in video creation, etc.; The second stage, known as the "collaboration phase" of AIGC, appears in the form of virtual humans that coexist with virtual and real to form a situation of human-machine symbiosis. The third stage, AIGC's "original phase," will complete content creation independently. We believe that as AI large models have multimodal content creation capabilities and AIGC has the conditions for large-scale application, the era of human-machine collaboration is coming.
The threshold for content creation is lowered, and imagination is productivity in the AI era. AI has been upgraded from traditional analysis of data discovery patterns to analysis of perceptual data and content production. Compared with the human brain, which can only process information in a few directions based on its own knowledge graph, AI has the ability to process information in multiple directions from a larger knowledge system, which can provide more creative ideas. Review the history of content creation, in the PGC era, content creation needs to use professional creation tools for content creation, such as photography enthusiasts need to learn to use professional PS (Adobe Photoshop) tools for retouching; In the UGC era, photography enthusiasts only need to learn to use the Meitu App with a low threshold to achieve retouching effects comparable to professional PS tools. We believe that in the era of AIGC, anyone will become a content creator, just give full play to the imagination and describe the requirements to AI tools, and AI can complete creative tasks with professional thresholds such as code, drawing, modeling, etc., and the completion effect even exceeds the creator's expectations.
AIGC empowers multimodal digital content generation and accelerates the advent of the metaverse era
The metaverse can bring immersive experiences, with digital humans, digital reconstructions of the physical world, and software agents as key to the digital world. The metaverse is the convergence of the physical world, augmented reality (AR) and virtual reality (VR) in a shared digital space, and the metaverse has the potential to become a universal platform for digital social interaction in the future. We believe that building a highly available metaverse platform requires both mature mixed reality technology (MR) to enable good human-computer interaction, and massive digital content to enable digital world experiences comparable to the physical world. Combined with SenseTime's understanding of the metaverse, we believe that AIGC can support immersive and interactive metaverse experiences in three core ways: 1) Digital humans: AIGC can create avatars for humans and provide an entrance to the virtual world; 2) Digital reconstruction of the physical world: AIGC's application in 3D reconstruction can realize the digital reconstruction of the physical world and build the connection between the physical and digital worlds; 3) Software agents: AI large models already have strong analysis and reasoning capabilities, and AIGC can create highly intelligent software agents to communicate smoothly with humans in the digital world.
AIGC unleashes digital content creation productivity and builds the foundation of the metaverse. Referring to the research results of the Gyro Research Institute, we believe that the creation of digital content in the metaverse is similar to the creation of games, and both require a large number of professional and technical personnel to carry out large-scale development of resources including text, images, 3D models, audio, video, code and so on. For a long time, the creation of digital content has been too complex to create a high barrier to entry. AIGC greatly lowers the threshold for content creation, and ordinary users can become "professional creators" with the help of AI, which will completely liberate digital content productivity and provide massive digital content for the metaverse. According to IDC data, the total amount of global data reached 84.5 zettabytes in 2021, and the total amount of structured and unstructured data is expected to reach 221.2 zettabytes by 2026; AIGC is expected to generate 10% of all data in 2025 (no more than 1% in 2021), with a CAGR of 127% for 2021-2025.
Digital humans have entered the AIGC era, and the market size of AI digital humans is expected to exceed 10 billion yuan in 2026. At the end of the last century, the creation of digital humans basically relied on hand-drawing, which had a long creative cycle and high labor costs. With the development of CG and motion capture technology in the early 21st century, digital human creation entered the computer age, but it was still limited by the capacity of professional creators. With the rapid development of multimodal AI models, digital human creation has entered the AIGC era, and the digital human industry has entered a period of vigorous development. Combined with the research results of Tencent Research Institute, we believe that AIGC can not only pipeline the production of digital humans with "good-looking" skin bags, but also continuously promote the development of digital humans in the direction of having "interesting" souls, and the creation cycle of digital humans has been greatly shortened. According to IDC forecasts, in the future, digital humans will gradually transition to pure AI-driven, and the AI digital human market will enter a stage of rapid development, with the size of China's AI digital human market reaching 10.24 billion yuan in 2026, with a CAGR of about 83% from 2022 to 2026.
Key enterprise analysis
Main line 1: AIGC computing power hardware supports the industrial chain
Tongfei Co., Ltd.: Founded in 2006, the company is a pioneer of domestic industrial temperature control and domestic substitution, with strong comprehensive competitiveness. As a leader in liquid cooling solutions, the company actively deploys its data center business, and joined the data center industry association in 2022 to exchange technologies and products with many companies. With the implementation of solutions with customers, we believe that 2023 is expected to become the first year of the company's data center liquid cooling product volume, and the company will fully benefit from the rapid expansion of AI servers. In 23Q1, the company achieved revenue of 270 million yuan and net profit attributable to the parent of 30 million yuan.
Helin Microna: Founded in 2012, Helin Microna's main business is semiconductor test probes and MEMS related micro components. With the advent of ChatGPT, leading technology companies have set off a new round of "AI arms race" around the world. NVIDIA is the global leader in AI chips and is expected to directly benefit from the strong demand for AI chips brought about by AI large model training and inference. The company has long-term cooperation with NVIDIA in the semiconductor chip test probe business, and the rapid increase in demand for AI chips is expected to directly drive the company's semiconductor test probe performance. In 23Q1, the company achieved revenue of 40 million yuan and net profit attributable to the parent of -10 million yuan.
Dingtong Technology: Founded in 2003, Dingtong Technology is mainly engaged in precision components of communication connectors and precision components of automotive connectors. The demand for computing power for AI pre-trained large models has increased significantly, which is expected to promote the continuous expansion and upgrading of related hardware infrastructure. The company's communication connector business firmly adheres to the strategy of key customers, digs deep into the incremental needs of customers, and in recent years, CAGE series new products continue to be introduced to core customers, and newly entered the Tyco CAGE series supplier directory; At the same time, the products continue to be upgraded, and the 2x6 cage products and 112GB/s structural parts products are also upgraded in line with the increase in transmission rate, which is expected to contribute to new increments, especially 112GB/s mainly to meet the needs of 800G data communication. In addition, the company has also set up a subsidiary in Malaysia, and is expected to expand the overseas business of major customers such as Tyco and Amphenol based on the Malay factory. In 23Q1, the company achieved revenue of 160 million yuan and net profit attributable to the parent of 30 million yuan.
Gao Lan Co., Ltd.: Founded in 2001, the company has deep technical accumulation in the field of liquid cooling, and through forward-looking layout, it has a cold plate liquid cooling server thermal management solution, immersion liquid cooling server thermal management solution, container liquid cooling data center solution. The business covers from liquid cold plates, CDUs of various models and heat exchangers, multi-power tanks, multi-size containers and other components to data center design, equipment integration, system commissioning, equipment operation and maintenance system integration. In 2022, the company's IDC thermal management products achieved revenue of about 107 million yuan, a year-on-year increase of 9835.1%. In 23Q1, the company achieved revenue of 150 million yuan and net profit attributable to the parent of -10 million yuan.
Main line 2: AIGC promotes the subversive improvement of industry production efficiency
Jichuang Intelligence: Founded in 2008, the company is a domestic smart city and smart security solution provider, providing security, information system, data perception collection and other products and solutions for downstream customers such as government and enterprise and the Internet. The company has increased R&D investment in core technologies such as high-performance computing, signal and protocol analysis, and provides underlying computing support for smart cities, smart and secure big data analysis applications, and data protocol analysis applications. The company has relatively comprehensive business qualifications, such as special grade A for building intelligent system design, first-class professional contracting for electronic and intelligent engineering, and qualification for the Ministry of Public Security. In 23Q1, the company achieved revenue of 190 million yuan and net profit attributable to the parent of 100 million yuan.
Fengyuzhu: The company was founded in 2003, the main business is digital exhibition and display business. As a leading company in the domestic digital display industry, the company has been deeply engaged in offline immersive cultural experience for many years, and has rich experience in the application of ARVR, virtual anchor and other technologies; In addition, the company has a rich talent pool in art and design, which is expected to promote the diversified development of digital collections and other formats. In 23Q1, the company achieved revenue of 420 million yuan and net profit attributable to the parent of 40 million yuan.
Main line 3: AIGC spawns a new ecology for metaverse development
Perfect World: Founded in 1999, the company is currently in a stable operation cycle of many main game products, and games such as "Phantom Tower", "Fantasy Xinxian", "Perfect World", "Xianxian Mobile Game" and other games maintain strong long-term operation capabilities. The company continues to deploy cutting-edge technology applications such as AIGC, which is expected to improve the efficiency of game research and development. The company has established an AI center, which is led by the CEO of the game business and led by the middle office technology department, and the project producers are deeply involved in researching and implementing AI technology learning and application. At present, some of them have been applied to intelligent NPC, scene modeling, AI painting, AI plot, AI dubbing, etc. With the maturity of AIGC technology application tools such as ChatGPT, Midjourney, and Stable Diffusion, AI technology will be applied to more scenarios in the company's game development, distribution and operation, further improving game development efficiency and optimizing player experience. In 23Q1, the company achieved revenue of 1.90 billion yuan and net profit attributable to the parent of 240 million yuan.
Gibbit: Founded in 2004, the company's main games "Ask the Road Mobile Game" and "One Thought Getaway" have maintained a stable operation as a whole, and the product pipeline has obtained version numbers and self-developed products is relatively rich. The company continues to deploy cutting-edge technology applications such as AIGC, and AI-related tools have helped the company's actual work, such as artists using AI painting tools to quickly construct materials, planners using AI painting tools to quickly construct prototypes, so that artists can understand planning intentions and reduce communication costs. AIGC tools such as ChatGPT have also been applied to planning and design work, effectively improving the company's production efficiency. In 23Q1, the company achieved revenue of 1.14 billion yuan and net profit attributable to the parent of 310 million yuan.