Due to the global popularity of ChatGPT, 2022 is known as the first year of AIGC, opening the wind for a new generation of artificial intelligence. The advent of ChatGPT has made the application of AI large models have a practical carrier, closer to people's production and life, from a realistic point of view, will greatly change the content industry such as games, film and television, and in the long run, will stimulate a large number of industry demand, profoundly affect the development of all walks of life.
1. About the AIGC industry chain
AIGC or AI-Generated Content is a technology that generates relevant content through data/information operations and appropriate generalization capabilities based on AI technologies such as pre-trained large models and adversarial generative networks (GANs). The AIGC industry chain can be understood as a four-layer architecture of the basic layer, network layer, platform layer and application layer.
The basic layer includes AI chips, servers, and data centers, as well as AI data collection, cleaning, annotation, and processing, providing computing power support for AIGC model training and application. The network layer is mainly responsible for network communication services under cloud computing architecture, which is the key support for efficient applications. The platform layer is mainly responsible for the development and training of general large models, as well as the vertical and scenario-based development of large models. The application layer mainly includes multimodal content generation and application in multiple industries, generally involving scenarios such as search, dialogue, and recommendation.
AIGC has a wide range of applications, including text, code, images, voice, video, 3D models, games, music, audio, etc. According to Sequoia Capital's forecast, text and code generation are expected to be maturely applied in 2023, and its Chinese generation can achieve accurate adjustment of vertical field copywriting, achieve the accuracy of scientific research papers, and code generation can cover multi-language and multi-vertical fields; The maturity of image, audio and video generation is relatively low, and it is still in the stage of generating the first draft of the foundation, and it is expected to be matured and applied in 2030.
Second, the hot opportunities of the AIGC industry chain
1. AI model
The ChatGPT launched by OpenAI has swept the world, and major manufacturers at home and abroad have followed suit. Microsoft has partnered with OpenAI in the field of artificial intelligence, and Google has launched a similar search tool, Bard. Domestic AI models include Baidu's "Wenxin", Alibaba's "Tongyi", Tencent's "Mixed Yuan", Huawei's "Pangu", JD.com's "Yanxi", iFLYTEK's "Spark", SenseTime's "Discussion", Kunlun Wanwei's "Tiangong" and so on. At present, 79 large models with a parameter scale of more than 1 billion parameters have been released in China.
In the future, the iterative upgrading of AI big models and the construction of platform ecology will continue to advance, and the industry opportunity is mainly to build regional large model innovation centers. For example, Wuxi Liangxi Talent Group cooperated with Baidu to build Baidu Intelligent Cloud Wenxin Qianfan Model (Wuxi) Innovation Center, as the first Baidu Wenxin Qianfan Model Empowerment Center base in China, to build Wenxin Qianfan Scientific Research Center, Intelligent Manufacturing Workshop, Talent Innovation and Training Base, Operation Service System, etc., providing a generative artificial intelligence full-process tool chain, bringing accurate intelligent manufacturing solutions to industrial enterprises and promoting the upgrading of intelligent manufacturing in the Yangtze River Delta.
2. AI chip
AI chips, also known as AI accelerators or compute cards, are modules specifically designed to handle a large number of computing tasks in AI applications. The development of AI large model applications will rapidly increase the demand for AI chips. At present, AI chips suitable for deep learning and large-scale commercial use mainly have three technical routes: GPU, FPGA, and ASIC. GPUs were first introduced into accelerated deep learning algorithms, and their applications are relatively mature, occupying most of the market share. FPGA, ASIC and other technologies are also developing rapidly, and the market share is expected to exceed 20% by 2025.
According to IDC's forecast, the scale of the mainland AI chip market will reach 43.2 billion yuan in 2021, a year-on-year increase of 47.2%; It will continue to maintain rapid growth in the next few years, and the market size is expected to exceed 100 billion yuan by 2025. AI chip industry mainly includes semiconductor materials, chip design, wafer manufacturing, packaging and testing and semiconductor equipment, semiconductor materials include silicon wafers, electronic gases, photomasks, polishing fluids and polishing pads, photoresist supporting reagents, photoresists, wet chemicals, sputtering targets, etc.
3. AI server
AI servers are equipped with acceleration chips such as GPUs, FPGAs, and ASICs, which can meet the needs of high-throughput interconnection and provide computing power support for AI application scenarios such as natural language processing, computer vision, and machine learning. The construction of large model platforms such as Google, Microsoft, Meta, Baidu, and Ali will greatly increase the market demand for AI servers. It is predicted that in the next five years, the share of AI servers in global data centers will grow from less than 10% today to 20%, and the share of such equipment in data center server capital expenditure will increase from 20% to 45%.
In the future, under the trend of gradually complicating AI large models, the development of AI servers will stimulate more memory usage, and simultaneously drive the growth of demand for Server DRAM, SSD, memory interface chips, etc. The development of AI servers will drive the development of high-end PCBs, server liquid cooling solutions and other markets, according to CCID Consulting, the mainland liquid cooling data center market size will reach 128.32 billion yuan in 2025.
4. Data center
The iteration of AI large models and the extension of industry application scenarios not only accumulate massive data, but also give birth to new data requirements. According to the McKinsey report, about 1/3 of the algorithm models are updated at least once a month, and about 1/4 of the algorithm models are updated at least once a day. At present, the large model used by ChatGPT has hundreds of billions of parameters, and as the main bearer of computing power, the demand for supporting construction of data centers has also increased. At the same time, the Al model will accelerate the transformation of traditional data centers into intelligent computing centers.
The AI big model will reconstruct the existing industry ecology of operators and form a new ecosystem with the big model as the core, including networks, cloud computing, deep learning frameworks, data annotation, SaaS applications, big model evaluation, and open source communities. IDC operators hold massive amounts of data, covering users' identity information, consumption information, Internet behavior, location information, access log data, and internal data, etc., deeply participate in data processing and technical services, and continuously invest in network computing power infrastructure, which is expected to transform into a comprehensive service provider of computing power platforms in the future.
5. Communication equipment
The wide application of AI large models requires high-performance computing, and also depends on the upgrading of network communication technology, mainly involving switches, optical modules, optical fibers, etc. The switch is one of the core equipment for building a network, and its main function is to expand the network coverage and connect more servers, computers and terminals. Optical module is an important hardware device in data center, consisting of optoelectronic devices, functional circuits and optical interfaces. The computing power layout of the mainland's "East Data and West Calculation" project has driven the demand for long-distance optical transmission cables.
Among them, high-speed optical modules have attracted special attention, and technologies such as silicon photonics, coherent and photoelectric co-packaging (CPO) will become solutions for reducing costs and increasing efficiency in high-computing power scenarios. These three technologies are iterative upgrades of optical module technology, of which silicon photonics and coherent technologies have developed considerably, and the hottest CPO technology has not yet reached the commercial level, and is still regarded as an auxiliary of silicon photonics technology. After the emergence of the large model, the market raised the shipment of 800G optical modules to more than 1 million units in 2023 and more than 2.5 million units in 2024. At present, the mainstream of the market has just moved towards 200G and 400G, and will be upgraded to 1.6T or even higher speeds in the future.
6. AIGC+ application software
AIGC has a natural connection point with the application software, and accessing AIGC in the application software can significantly improve the user's work efficiency. From the current Microsoft Office products that integrate ChatGPT, it can be seen that ChatGPT gives office software more convenience and automation functions, greatly reduces the threshold of use, and greatly improves office efficiency. Recently, domestic office software leaders DingTalk and WPS have announced that AIGC capabilities will be embedded in office software. With the blessing of ChatGPT, the value of office software is expected to increase significantly, and the industry is ushering in new opportunities.
Wu Taibing, chairman of Wondershare Technology, believes that as an infrastructure, large models can empower AI in thousands of industries and help the rapid development and iterative launch of software products, based on which three new trends will emerge: the future AIGC software ecology = several large models + massive small applications; The innovation of the interactive interface makes all software need to be done again; With AI empowerment, creative software will usher in unprecedented opportunities.
7. AIGC+ scenario service
Mainly involved in content generation, virtual people and virtual scenes and other fields, AI digital employees are expected to become an important productivity in the AI era, can be embedded in business workflow to assist production, in research and development can assist creative design, in marketing can assist marketing idea generation, in service can become anchors and customer service.
(1) AIGC+ Media: writing robot, interview assistant, video subtitle generation, voice broadcast, video collection, artificial intelligence synthesis anchor
(2) AIGC+ e-commerce: 3D models of goods, virtual anchors, virtual goods yards
(3) AIGC+ film and television: AI script creation, AI synthesis of faces and voices, AI creation of characters and scenes, AI automatic generation of film and television trailers
(4) AIGC+ entertainment: AI face changing application, AI composition, AI synthetic audio and video animation
(5) AIGC+ education: AI synthesizes virtual teachers, AI makes historical figures based on textbooks, and AI converts 2D textbooks into 3D
(6) AIGC+ Finance: Realize the automated production of financial information and product introduction video content through AIGC, and shape virtual digital human customer service through AIGC
(7) AIGC+ medical treatment: AIGC synthesizes speech audio for the voiceless, body projection for the disabled, and medical companionship for patients with mental illness
(8) AIGC+ Industry: Complete repetitive low-level tasks in engineering design through AIGC, generate derivative designs through AIGC, and provide inspiration for engineers
8. AIGC+ Internet of Things
The deep combination of AIGC and IoT can provide new ideas for the digital transformation of industries, one is to develop smart devices with edge computing power, and the other is to embed IoT solutions in all walks of life. AIGC+ intelligent terminals will reconstruct the human-computer interaction functions of future intelligent terminals, such as AR/VR/XR, tablets, mobile phones, audio, TVs, live broadcast equipment, smart retail terminals, smart cars, etc. For example, "Mobvoi" launched a smart watch with AIGC capabilities, and iFLYTEK launched the world's first AIGC headset.
As the Internet of Things enters the intelligent era, intelligent modules will integrate edge computing power such as connectivity, computing, perception, and expression, and become the core carrier supporting hardware intelligence and edge computing. The intelligent module has the characteristics of SoC heterogeneous computing and will gradually integrate functions such as CPU + GPU + DSP + NPU, which can undertake fragmented intelligent scenarios and meet the different needs of different industries.
Third, the layout of artificial intelligence industry in Hangzhou High-tech Zone
Hangzhou High-tech Zone (Binjiang) is the "core" of Hangzhou National Artificial Intelligence Innovation and Application Pilot Zone, which gathers high-tech parks such as Huawei Hangzhou Research Institute, New H3C Group, Supcon Software Park, Inventronics, Alibaba, and NetEase.
1. Industrial ecological construction
Relying on the advantages of the digital economy industry in the high-tech zone, a new "2+2+N" artificial intelligence industry system has been formed, with "intelligent computing and integrated circuits" as the basic layer (basic industry), "intelligent Internet of Things, digital communication" as the technical layer (advantageous industry), and "intelligent equipment, digital health, digital trade, digital cultural creativity, digital finance, and digital e-commerce" as the application layer (application industry).
Actively cultivate and supply rich application scenarios from the three energy levels of "city-level, industry-level, and enterprise-level", plan to lay out collaborative innovation platforms such as national manufacturing centers, industrial innovation centers, and industrial technology innovation alliances in the field of digital economy, rely on Huawei, Hikvision, NetEase and other leading enterprises to form industrial innovation consortiums, explore and carry out research on rules and regulations, regulatory policies, insurance protection, etc. in the field of artificial intelligence, and try a number of new industrial development policies first.
2. Hangzhou Artificial Intelligence Industrial Park
Hangzhou Artificial Intelligence Industrial Park, located in Binjiang District, Hangzhou, is an industrial development and innovation platform based on artificial intelligence, focusing on incubating start-up growth enterprises engaged in supercomputers, intelligent hardware, 5G communication networks, mobile Internet, big data, cloud computing and other fields, and creating artificial intelligence professional incubators, accelerators and multipliers. With a total area of more than 80,000 square meters, the park officially opened in November 2017.
The park is jointly built by Zhejiang Jingxin Smart City Planning and Research Institute, Zhejiang Artificial Intelligence Society, Binjiang Group and Zhejiang Zhixin Zedi Science and Technology Development Co., Ltd., and unites relevant industry associations, university research institutes, science and technology service institutions, financial capital forces and other social resources to form a service alliance to jointly provide entrepreneurial services and industrial resource support for innovative entrepreneurs, focusing on building an industrial development service system of "one main and three modernization and five platforms".
First: create an ecosystem conducive to the rapid development of the artificial intelligence industry;
Three modernizations: artificial intelligence technology achievements (incubator), artificial intelligence achievement industrialization (accelerator), artificial intelligence industry capitalization (multiplier);
Five platforms: artificial intelligence industry industry-university-research cooperation platform, artificial intelligence industry technology achievement trading platform, artificial intelligence industry public service platform, artificial intelligence industry entrepreneur exchange platform, artificial intelligence industry investment and development platform.
3. Hangzhou AIGC Industrial Park
Hangzhou AIGC Industrial Park is located in the Torch Innovation Center, the core area of Binjiang Block, adjacent to the Artificial Intelligence Industrial Park, with a construction area of 240,000 square meters. AIGC industry-related innovation and entrepreneurship projects can enter the AIGC Industrial Park after evaluation, and can enjoy a series of industrial support policies such as rent subsidies and R&D investment.
Rent subsidy: A three-year rent subsidy policy can be given, and the subsidized area is approved according to the standard of 15㎡/person.
R&D investment subsidy: According to the main creative team, financing and R&D investment of the project, a certain proportion of R&D funding will be given for a period of 3 years and up to 10 million yuan.
Application scenario support: In the stage of project industrialization, it can build a platform for the company to exchange and dock application scenarios between enterprises and government and enterprises in the region.
Talent policy: provide high-level talent identification, medical care, children's education, housing security, parental health care, government affairs green pass and other services for the core R&D team recruited by the company.
Professional platform: build a professional investment and financing platform and industry exchange platform.