Frost & Sullivan releases the 'China AI Basic Software Market research report (2023)'

AI2.0 is composed of the integration of big data, cloud computing, and artificial intelligence technologies, forming a huge platform opportunity that will provide more in-depth and comprehensive solutions for various industries in the future. With the advent of the AI 2.0 era, basic models no longer require manual annotation data and can learn and read vast amounts of text on their own. Moreover, models can be fine-tuned to train for tasks in different fields at low cost. Among them, AI2.0-related applications, platforms, and infrastructure will become hotspots including industry and investment. AI basic software, as an important platform for training, managing, and applying large AI models, will evolve into huge industrial opportunities with the long-term development of AI2.0.

Source: Frost & Sullivan report

On one hand, the explosion in data volume has driven the development of the basic data service industry and expanded the scope of services provided. China's data volume increased from 2.3 ZB (ten trillion bytes) to 23.88 ZB between 2017 and 2021, and is expected to reach 76.6 ZB by 2026, ranking first in the world and continuing to grow explosively in the future. With the explosive growth of big data, enterprises and organizations are also experiencing a sharp increase in their demand for processing, managing, and analyzing this data. This has brought about huge market demand for the basic data service industry and promoted its development.

On the other hand, synthetic data accelerates the supply of high-quality data, providing a foundation for the development of large AI models. Synthetic data is expected to lift the data constraints on AI, driving artificial intelligence towards the 2.0 phase. At this stage, synthetic data can not only train AI models but also enable self-learning and evolution through data simulation.

Source: Frost & Sullivan report

AI foundational software includes a series of platform software products and solutions required for enterprise-level AI applications. It is the main efficiency support for the implementation of large model applications. The development of AI foundational software determines the depth, height, and breadth of artificial intelligence development. It catalyzes the rapid development of large model applications and promotes the large-scale application of AI in government and enterprises.

The advent of the AI2.0 era has led to an increasing demand for large model applications across various industries, and there is a greater emphasis on the supporting capabilities of large models for business operations. However, most enterprises face issues such as insufficient engineering and technical capabilities.

As the top priority of AI infrastructure, AI basic software provides enterprises with comprehensive AI scheduling and model services. It includes one-stop model platforms such as machine learning platforms, as well as data platforms serving AI like data intelligence platforms, real-time decision centers, data lakes, and data warehouses. This reduces the threshold for industry customers to train their own AI models, achieving cost reduction and efficiency improvement.

Source: Frost & Sullivan report

At the data level, the quality and scale of model training data are crucial for the effectiveness of model iteration. Especially in current issues such as internal and external data sharing and co-creation, unbalanced data categories, and missing data in extreme scenarios, the industry calls for technological exploration in the field of structured data synthesis using AI generative computers (AIGC).

At the technical level, the reasoning capability of generative AI models is becoming increasingly important. At the same time, there is also a growing demand for the credibility and explainability of AI, calling for improvements in the performance of foundational software in areas such as automatic machine learning, deep learning, and causal learning.

At the business model level, as large models gradually mature, general-purpose large models + industry-specific small models will become the adopted implementation model for more and more enterprises. Therefore, AI basic software that helps enterprises build their own AI models is an inevitable trend.

Source: Frost & Sullivan report

At the national regulatory level, China is actively deploying in the artificial intelligence industry, competing for the 'future track.' With the growing demand for artificial intelligence across various industries and fields, new models that are deeply integrated with the real economy are emerging continuously, forming a R&D system and application ecosystem with Chinese characteristics. The technological explosion represented by 'large models' has accelerated the development of the artificial intelligence industry. Multiple AI industry support policies introduced by the state and local governments will boost the development of the industry, further promoting the integrated development of the digital economy and the real economy.

The construction and improvement of the open-source ecosystem have a tremendous impact and drive on the field of AI foundational software. It not only promotes technological innovation and progress but also fosters broader cooperation and knowledge sharing. Specifically, open source reduces the development threshold for AI foundational software, while facilitating knowledge sharing and collaboration among developers. Moreover, it expands the application fields of AI foundational software, providing a highly flexible and customizable foundation for a wide range of developers.

Source: Frost & Sullivan report

AI basic software is mainly located in the midstream of the industrial chain. Its industrial chain consists of upstream infrastructure and resource providers, midstream AI basic software and one-stop AI development platforms, and downstream application domain enterprises.

The upstream is the foundation for AI software deployment, providing computing power support for AI foundational software. The midstream is the core of the industrial chain, starting from simulating human intelligence-related characteristics to construct application technology paths, mainly including AI foundational software and one-stop AI development platforms. The downstream primarily involves the application of AI foundational software in various sub-scenarios, mainly including manufacturing, security, finance, healthcare, retail, transportation, and other fields.

Cloud providers tend to offer end-to-end holistic solutions to customers, expecting them to be bundled as a whole. In contrast, leading AI foundational software can be embedded into customer cloud systems in a modular manner, fully meeting the requirements for autonomy and control.

Source: Frost & Sullivan report

Jiuzhang Yunji DataCanvas is a leading AI foundational software provider, with its core product series, AIFS AI Foundation Software and DataPilot, being innovative, flexible, and scalable. AIFS empowers users with the AI capability to "autonomously build large and small models," including a full set of open-source, highly automated, and collaborative software tools such as DataCanvas Alaya's "General Knowledge + Industry" white-box large model matrix, APS machine learning platform, BAP business-oriented automatic modeling platform, DAT automatic machine learning software, and YLearn causal learning software. This accelerates the accumulation of user-urgently needed autonomous AI capabilities and the scaling of models. As a new generation data architecture tool product based on self-developed large models, DataPilot includes tools such as DingoDB multi-modal vector database and RT real-time decision center platform, helping users autonomously build a secure and controllable "vector ocean," achieve intelligentization and automation across the entire chain, and empower users with future-oriented multimodal real-time data processing capabilities.

Amazon SageMaker, a machine learning platform under Amazon Web Services (AWS), provides developers, data scientists, and ML engineers with complete and rich features. Amazon SageMaker Studio Lab offers free resources, while Amazon SageMaker Jumpstart and Amazon SageMaker Canvas provide low-code/no-code quick-start capabilities, Amazon SageMaker Pipelines build fully automated ML processes, Amazon SageMaker Ground Truth Plus offers intelligent annotation services, Amazon SageMaker Data Wrangler includes more than 300 built-in data transformations, Amazon SageMaker Autopilot automatically executes AutoML, and has a comprehensive AI development software and hardware stack supply level, from dedicated infrastructure, AI platforms to ready-to-use AI service solutions for various scenarios. Combined with AWS's series of cloud services, it meets the different needs of various types of customers.

Huawei Cloud AI Development Platform ModelArts is a one-stop AI development platform for developers. It enables rapid model creation and deployment, manages the entire AI workflow cycle, and provides massive data preprocessing and semi-automatic annotation for machine learning and deep learning, large-scale distributed training, and automated model generation. It helps industries upgrade their intelligence level in various fields. Huawei Cloud AI Development Platform ModelArts has advantages such as providing an end-to-end toolchain for model development, training, and inference, offering cost-effective AI computing power, supporting tens of thousands of cards in large-scale clusters, and enabling fault tolerance and automatic recovery from training job failures.

Alibaba Cloud's related products include the machine learning platform PAI, model online service PAI-EAS, and more. Among them, the machine learning platform PAI is a machine learning/deep learning engineering platform for developers and enterprises. It provides a full-link AI development service including data annotation, model construction, model training, model deployment, and inference optimization. It has built-in more than 140 optimization algorithms and rich scenario plugins, offering users a low-threshold and high-performance cloud-native AI engineering capability.

The Tencent Cloud AI Open Platform includes the Tencent Cloud TI platform, TI-Matrix application platform, etc. It is a full-stack artificial intelligence development service platform based on Tencent's ecosystem and AI development experience, providing services for developers and government and enterprise clients. It connects the entire industrial and AI implementation process chain, including data acquisition, data processing, algorithm construction, model training, model evaluation, model deployment to AI application development, helping users quickly create and deploy AI applications, manage full-cycle AI solutions, accelerate the digital transformation of government and enterprises, and promote the co-construction of the AI industry ecosystem.

The Databricks Lakehouse platform, under the Databricks umbrella, combines elements of data lakes and data warehouses. It offers the flexibility, cost-effectiveness, and scalability of data lakes while also providing data management transactions. Users can enable business intelligence and machine learning on all data. The Databricks Lakehouse platform provides a unified environment for data analysis and machine learning, enabling features such as business intelligence and SQL analysis, data science and machine learning, and real-time data applications. On the Databricks Lakehouse platform, enterprises can easily ingest and transform batch and stream data, orchestrate reliable production workflows, and improve productivity by leveraging built-in data quality testing and support for software development best practices.

Baidu's AI development platform consists of the Qianfan Large Model Platform, the Full-Function AI Development Platform (BML), the Zero-Threshold AI Development Platform (EasyDL), and the AI Development Training Platform (AI Studio), all empowered by the independently developed Baidu PaddlePaddle platform. Baidu's AI development platform covers developers with different needs, including data processing, algorithm development, model training, prediction service deployment, as well as resource management capabilities that span the entire lifecycle of models.