Communication Plus· Industry ｜ Big Model, Generative AI Pushes the Coming of AI 2.0 Era
2023 is a watershed in the development of artificial intelligence. The rapid development of large model and generative AI has driven the paradigm shift in the field of artificial intelligence, and the AI 2.0 era has arrived. According to the latest data, in the past four years, the parameters of the AI large model have increased at a compound annual rate of 400%, and the demand for AI computing power has increased by more than 150,000 times, far exceeding Moore’s Law. However, the traditional computing infrastructure centered on CPU can no longer meet the new requirements of large model and generative AI.
Recently, Shangtang Science and Technology Intelligent Industry Research Institute, Cloud Computing and Big Data Research Institute of China Information and Communication Research Institute, China Intelligent Computing Industry Alliance and Artificial Intelligence Computing Industry Ecological Alliance jointly released the White Paper on the New Generation of Artificial Intelligence Infrastructure (hereinafter referred to as the White Paper). The "White Paper" pointed out that the AI 2.0 era has arrived, clarified the definition, characteristics and value of the "new generation AI infrastructure" in the AI 2.0 era, and proposed the "new generation AI infrastructure evaluation system" for the first time, providing an important reference for the development of the intelligent computing industry in the AI 2.0 era.
The era of AI 2.0 has arrived.
Generative AI drives the large-scale development of AI market and brings new economic benefits (data source: Gartner, McKinsey, IDC).
The white paper shows that the AI 2.0 era has arrived. Prior to this, artificial intelligence helped to analyze data and make predictions by pattern detection or following rules, which was more like a "classifier", while the AI 2.0 era opened a new stage: generative AI based on large models. Generative AI can imitate the creative process of human beings through data training, and evolve artificial intelligence from a traditional "classifier" to a "generator". In this regard, Gartner predicts that by 2027, the rapid growth of generative AI will contribute 42% of the global artificial intelligence expenditure, and the scale will exceed 180 billion US dollars, with a compound growth rate of 169.7% from 2023 to 2027.
In addition, as the basis of the development of generative AI, large models are also developing at a high speed. According to IDC data, by the end of November 2023, more than 300 large models have been released in China market. The subversive potential of generative AI has been recognized by more and more enterprises. Gartner predicts that by 2026, more than 80% of enterprises will use the API or model of generative AI, or deploy applications supporting generative AI in production environment, compared with less than 5% in early 2023.
Technological changes have driven the expansion of the scene, and generative AI has also moved from heated discussion to application, and its value creation potential is extremely amazing. McKinsey predicts that generative AI is expected to contribute about 7 trillion US dollars to the global economy and increase the overall economic benefits of AI by about 50%; China is expected to contribute about $2 trillion, nearly one third of the global total.
A new generation of AI infrastructure has become a "new infrastructure" in the AI 2.0 era.
The new generation AI infrastructure is mainly composed of computing power, MaaS and related tools.
The White Paper shows that intelligent computing power in the AI 2.0 era has become a key supporting factor for the development of the AI industry. The demand for data quality and efficiency in large-scale model training, and the demand for MaaS (big model as a service) in enterprise application of generative AI have all increased. The new generation of infrastructure should support the training and reasoning of large models and the large-scale landing of generative AI applications. Its core elements, such as computing, storage, network and data services, should be finely designed and reconstructed, rather than simply stacking servers or GPUs.
The "White Paper" defines the new generation AI infrastructure: taking large model capability output as the core platform, integrating computing resources, data services and cloud services, and specially designed to maximize the performance of large model and generative AI applications: data preparation and management, large model training, reasoning, model capability call, and generative AI application deployment. Enterprises develop and run generative AI services and customer applications through a new generation of AI infrastructure, as well as training and fine-tuning basic models and industry models.
The "White Paper" pointed out that the construction of a new generation of AI infrastructure will lower the threshold for the development and application of large models, and create greater social value in government and enterprise services, industry and scientific research innovation: industry-oriented, it will accelerate the intelligent transformation of all links in the upstream and downstream of traditional industries, and promote the emergence of new formats and new models. Facing scientific research, it can accelerate the automation and intelligence of scientific experiments and stimulate a new paradigm of AI for Science. Facing government affairs, we will apply the scattered and fragmented government affairs to improve the quality and efficiency of government affairs services through "one model for all".
A new generation AI infrastructure evaluation system is proposed for the first time.
The "White Paper" first proposed the "New Generation AI Infrastructure Evaluation System". Through the three dimensions of product technology, strategic vision and market ecology, and twelve evaluation indicators, the comprehensive capabilities of AI infrastructure manufacturers are comprehensively evaluated qualitatively and quantitatively.
The White Paper selected 12 most representative AI infrastructure vendors. -SenseCore Shang Tang Big Device, a new generation AI infrastructure of Shangtang Technology, scored more than the average score of the manufacturer in all evaluation indexes, and got full marks in four evaluation indexes: market response, market cognition, product strategy and engineering construction, becoming the leader of the new generation AI infrastructure market.
According to the White Paper, Shang Tang has shown strong product strength and technology accumulation in product service capability, which not only laid out computing infrastructure in advance, but also formed a complete set of AI infrastructure product architecture with the empowerment of MaaS platform and its own large-scale model business, meeting the large-scale landing needs of customers for large-scale model training and generative AI applications.
SenseCore Shang Tang Large Device, as an AI infrastructure product and solution launched by Shang Tang, has an overall computing power scale of 6800P. It supports Shang Tang’s own large model research and development, as well as external customers’ training of large models and application deployment. Since 2023, more than 1,000 large models with parameters of billions to hundreds of billions have been trained on large devices, and dozens of generative AI applications have been supported.
Generative AI application has entered the era of great navigation.
In the AI 2.0 era, the artificial intelligence industry ushered in a more prosperous "great navigation era"
The White Paper also pointed out that generative AI applications are experiencing explosive growth. At the earliest, the applications of Wenshengwen and Wenshengtu, represented by ChatGPT and Midjourney, were introduced to the market and gained a rapid growth of user groups. Subsequently, applications such as audio generation, video generation and multimodal generation, as well as tool applications for different industries or user groups, such as code generation, Copilot, digital people, marketing tools and chat assistants, have been continuously introduced to the market.
In November 2023, OpenAI launched GPTs and plans to build GPT Store, which allows users to create customized versions of applications by combining their own instructions, external knowledge and capabilities without code. This customized mode and clear commercialization mode make the development subject of generative AI applications move from a small number of AI vendors to a large number of AI developers.
Finally, the White Paper points out that as the marginal cost of the new generation AI infrastructure continues to decline and the marginal benefits continue to grow, AI will benefit everyone.
Reporting/feedback