Event Review

MatrixOne → MatrixOS: "AI Infra" and "AI Platform" Rewrite the Entrepreneurial History of MatrixOrigin

Posted by MatrixOriginPublished on

Under the impetus of the digital wave, the story of MatrixOne is like an epic tale of entrepreneurship in the tech world, beginning with a simple yet grand dream - to build an operating system capable of supporting the new generation of the digital world. Imagine, in the era of AIGC, where data flows like "blood," algorithms operate like "heartbeats," and user interfaces are like "breaths," a new generation operating system acts like a conductor, precisely handling diverse massive data and manipulating the allocation of computing power. How vibrant this digital world will become.

The story starts in 2021, with the founding team of MatrixOne standing at the entrance to the digital world, seeing a huge opportunity - data, the new oil of this era, waiting to be mined and utilized. To unleash the true potential of data, there must be powerful storage and processing capabilities. Thus, MatrixOne was born, initially positioned as a hyper-converged heterogeneous database, capable of handling various types of data to meet the needs of different industries.

However, as time passed, the team at MatrixOne realized that merely processing data was not enough. In this era driven by artificial intelligence and machine learning, computing power has become the new battlefield. Enterprises and developers are no longer satisfied with a single data storage solution; what they need is a platform that can integrate AI capabilities, meet computing power requirements, complete AI infrastructure construction, and balance business costs, efficiency, and technological advancement. Therefore, the R&D team of MatrixOne decided to expand its business to the fields of AI Infrastructure (AI Infra) and AI Platform, and deepen integration and close cooperation with VNET's AI Data Center (AIDC) business. In light of this, InfoQ conducted a special interview with Wang Long, CEO of MatrixOrigin, focusing on the business expansion plans of MatrixOrigin and discussing the future development of databases and AI technology.

MatrixOne upgrading to MatrixOS is a "response" to market demand under the wave of AI.

AI technology has rapidly become the core driving force behind the transformation of various industries, relying on powerful computing power and efficient data processing capabilities. Currently, there are many software platforms revolving around these two aspects, but behind this prosperity lies a challenge - the lack of standardization. Areas such as GPU cloudification, virtualization, and resource pooling have yet to establish unified standards, in contrast to the mature CPU public cloud services. In terms of data management, both traditional database solutions and emerging AI-driven data technologies are exploring how to more effectively manage and utilize data. The introduction of AI technology has brought new dimensions to data usage and management, leading many companies to innovate solutions such as vector databases, RAG models, and AI agent technologies to adapt to this change.

MatrixOrigin believes that "establishing effective connections between data and computing power is crucial." Therefore, when developing the MatrixOne cloud-native database, they focused on connecting "future data" and "past data," solving the data fragmentation problem caused by traditional databases, with the goal of integrating different types of data such as structured, unstructured, and semi-structured to support the practical applications of AI in various industries. This data integration capability, along with their compatibility strategy for computing power and data, provides them with a leading advantage compared to other data platform and computing platform vendors in the market.

Therefore, when MatrixOrigin recently completed a pre-A round of financing worth tens of millions of dollars, they decided to use this capital to develop MatrixOS, an ultra-simple unified, open-source AI-Native data intelligence platform. It is reported that the system will consist of three parts: MatrixDC, a large-scale heterogeneous computing power governance scheduling platform; MatrixOne, a hyper-converged heterogeneous data management platform; and MatrixGenesis, an AI intelligent application development platform. The goal is to create an AI Native software platform that links computing power, data, knowledge, models, and enterprise applications.

Matrix OS provides a unified platform for managing computing power and data, allowing different users to combine them according to their needs. For AI scientists, Matrix OS addresses the problem of moving data and computing power between different systems. They can train their models on the computing platform with their training data, and once training is complete, the models can be hosted on the same platform. Wang Long stated, "Matrix OS provides a unified operating platform, simplifying the connection and management between data and computing power."

At the same time, for developers who may not need to train models but need to use API calls to invoke large models or traditional models for inference, Matrix OS also provides cost-effective tools that simplify data connection issues, enabling developers to integrate API calls with traditional applications and workflows. For end-users, they can more tangibly experience faster application optimization, reduced latency, and improved cost-effectiveness. For example, the costs of AI services (such as token prices) and database services (such as node prices) are lower, applications are smoother, latency is lower, and usage costs are lower.

From a media perspective, the upgrade of MatrixOne to MatrixOS is a highly noteworthy industry development. It reflects MatrixOrigin's continuous improvement of existing services and active embrace of emerging technologies. It is evident that while maintaining the continuity and stability of its core business, MatrixOrigin is also actively exploring technological innovations to meet the evolving needs of the market and users. This upgrade of MatrixOne is a complement rather than a replacement for existing services. This business expansion by MatrixOrigin is not only a reinforcement of existing business but also a forward-looking layout for future trends.

AI Infra and AI Platform, a necessity for a one-stop AI solution

For enterprises looking to invest in AI development, the foundational AI Infra provides the necessary hardware and software infrastructure to support large-scale data processing and AI model training. Meanwhile, the top-tier AI Platform offers user-friendly services and tools to assist enterprises in rapidly developing and deploying AI applications. For companies originating from the database sector, offering a one-stop AI solution from the bottom to the top can enhance customer stickiness and expand technical service advantages.

Therefore, when we take a macro view of MatrixOS, the technological capabilities between the three sub-products — the large-scale heterogeneous computing scheduling platform MatrixDC, the hyper-converged heterogeneous data management platform MatrixOne, and the AI intelligent agent application development platform MatrixGenesis — are highly correlated. While there may be some functional overlap, their technical capabilities complement each other, collectively forming a powerful data intelligence platform.

In the past, MatrixOne could essentially be described as a data platform with a lake-warehouse architecture, but much simpler than Hadoop. Technologically, MatrixOne's architecture is unique, implementing multi-tier management including data tiering, read-write tiering, and storage-compute separation. This architecture allows flexible scheduling of computing resources and data resources, enabling MatrixOne to adapt to various application scenarios such as banking, the Internet, and IoT, like a Transformer. For example, to accommodate the development of large model technology, MatrixOne already has the capability of a vector database, which can support AI demands. Today, MatrixOne can naturally manage new data sources from the intelligent Internet of Things and AI, which will be the main source of "future data" and the foundation for the operation of MatrixOS, enabling MatrixOS to better connect with the AI world.

In the field of AI Infra, achieving disruptive technological changes requires addressing the two core issues of computing power and electricity. Currently, initiatives like Nvidia's Stargate project and the Chinese government's construction of intelligent computing centers are actively enhancing computing power at the hardware level, while software upgrades mainly revolve around GPU technology. A combined hardware-software solution is needed within the industry to complete the future AI infrastructure construction. Thus, MatrixOrigin regards MatrixDC as the resource base for MatrixOne and MatrixGenesis, using cloud-native technology to assist enterprises in building AI-native resource management and scheduling platforms.

The recent Pre-A round investment led by VNET allows MatrixOrigin to test its software on the Wanka cluster, providing them with valuable practical experience. Leveraging the advantages of a global talent team, MatrixOrigin iterates rapidly through practice on the Wanka cluster, building technological barriers and creating competitive MatrixDC, which provides large-scale heterogeneous GPUs that can rapidly meet the computing power needs of AIGC application scenarios. In terms of managing computing clusters, MatrixDC can effectively assist data centers in the unified management and resource pooling of heterogeneous chips and GPU servers, achieving resource sharing and dynamic scheduling to improve resource utilization.

Moreover, MatrixDC flexibly supports CPU computing resources, efficiently running structured transaction/analytical business applications; with AINet high-speed network architecture, it can effectively support workload training on thousands or even tens of thousands of card clusters, offering fully managed enterprise-level container services, enabling enterprises to fully enjoy the convenience of container and microservices architectures. Additionally, MatrixDC adopts a Serverless pay-as-you-go business model, helping enterprises significantly reduce the exploration costs of AI innovation applications. In the field of AI Platform, the primary technical challenges currently revolve around addressing the "illusion" of large models and the integration of large models with applications in traditional industries. There is an urgent need within the industry for a platform to integrate and connect all relevant capabilities while bringing together different technologies and solutions. This integration and collaboration will spawn new AI development paradigms, significantly lowering the development threshold for AI technology applications. Thus, MatrixOrigin aims to provide AI developers and users with a one-stop platform for data sharing, model training, and the development and operation of large-scale model applications through MatrixGenesis and MatrixOne. This will help AI technology to be quickly applied across various industries. Typical application scenarios include enterprise knowledge base management, intelligent customer service, complaint classification, intelligent report generation, policy consulting, and office assistants.

Among these, MatrixGenesis, as an AI intelligent agent application development platform, adopts the "Model as a Service" (MaaS) approach, focusing not only on the software level but also optimizing the entire process from data to computing power. By providing end-to-end development support and cost-effective large model services, it assists enterprises in building AI applications with zero threshold. It supports the selection, deployment, and inference of large models, seamlessly integrating with structured system data and knowledge bases in real-time. MatrixGenesis provides full lifecycle development support for AI application engineering flow orchestration and application docking integration. Enterprises can deploy open-source and closed-source large models with a single click, quickly completing model fine-tuning and inference acceleration. Additionally, the MatrixGenesis platform allows customers to access data and computing resources at a higher cost-effectiveness, significantly reducing business usage costs. During the application deployment phase, MatrixGenesis leverages the support of MatrixOne to provide users with complete data management and utilization capabilities. This includes components such as RAG, knowledge bases, and agents, which, through zero-code configuration, collectively contribute to the deployment and operation of applications, ensuring efficient utilization of data resources. Describing MatrixGenesis, Wang Long stated, "Users can achieve comprehensive optimization from data to computing power to application on this platform, ensuring the rapid construction of AI intelligent agent applications."

The cooperation between MatrixOS and VNET is the beginning of the construction of the "Chinese version of CoreWeave".

In this rapidly evolving digital era, AI technology is like an unstoppable tide sweeping through the entire business world. Many enterprises have established AIDC departments, dedicated to building and managing artificial intelligence infrastructure, providing key services such as computing power, data storage, and model training, in hopes of securing a place in this wave of technological innovation. It is widely recognized that the rise of AI technology not only brings new growth opportunities for enterprises but also enhances their core competitiveness in the fierce market competition.

As a leading third-party data center provider in China, VNET has a distinctive AIDC strategy characterized by a firm commitment to "All In AI," channeling all resources into the development of AI computing power. This transformation shifts the traditional service centered around data center cabinets to one focused on "AI computing power." This shift reflects VNET's profound insight into the future digital world — computing power will become the most critical asset in the future IT world. The collaboration between MatrixOrigin and VNET is based on their mutual recognition of the importance of computing power and electricity, as well as their shared vision to advance the development of the AI field.

MatrixOrigin, with its open-source MatrixOS software, combined with VNET's AIDC products, has jointly developed the GPU platform service, Neolink.AI, which will be launched in the third quarter of this year. This platform not only addresses enterprise challenges in AI computing power, software connectivity, and cost control but also aims to be an open-source, shared AI computing power platform, aiming to reduce inference costs and attract more users, thus transforming from a cost center to a value creation center.

VNET believes that the upper limit of AI lies in computing power and electricity, while MatrixOrigin believes that computing power and data are the bottlenecks of AI. Therefore, they coincidentally propose the idea of building a "Chinese version of CoreWeave", hoping to promote the localization of AI technology development and enhance the innovation and competitiveness of the domestic AI industry through the Neolink.AI platform. The realization of this idea will have a profound impact on China's AI field and lay a solid foundation for China's development in the global AI field.

During the interview with Wang Long, we also learned that VNET is about to acquire a batch of compliant H20 chips. The arrival of these chips will be an important milestone, marking VNET's solid step forward in computing power resources, which will be made available to everyone through Neolink.AI.

Embracing open source MatrixOS is an accelerator that empowers AI landing in thousands of industries.

In the current surging wave of AI, enterprises urgently need a ship capable of navigating the winds and waves. MatrixOS is precisely such a promising vessel. More than just an operating system, MatrixOS serves as an accelerator for enterprise AI construction. By providing efficient, stable data processing capabilities and flexible AI technical support, MatrixOS enables enterprises to easily overcome technological barriers without investing substantial resources and effort in building complex systems, thus rapidly realizing AI application construction. From the infrastructure layer to the database layer and further to the application development layer, MatrixOS offers a toolbox for enterprises across various industries in the AI era, empowering the ecosystem to create efficient and intelligent AIGC solutions, such as intelligent manufacturing in the manufacturing industry and precise risk control in the financial sector.

Looking ahead, the road of technological innovation for MatrixOS is full of infinite possibilities. MatrixOrigin will continue to explore new data and scenarios in the future. With the continuous breakthrough of computing power, MatrixOS will be able to handle larger and more complex datasets, providing solid computing power support and model services for the development of AI technology.

According to Wang Long, the exploration of MatrixOS does not stop there. It will also be closely integrated with cutting-edge technologies such as the Internet of Things and 5G to create brand new technical solutions. In an era where IoT devices are increasingly popular and 5G networks are commercialized, data is growing at an unprecedented rate. By integrating these technologies, MatrixOS will achieve more efficient data management capabilities, providing enterprises with more timely and accurate data services and AI development services, promoting enterprises across various industries to move towards a higher level of intelligence and digitization.

At the same time, as an open-source platform, the importance of ecosystem construction for MatrixOS is self-evident. It will join hands with strategic investors such as VNET to attract more enterprises and individuals to join the open-source community, jointly promoting the advancement and popularization of AI technology. MatrixOS will also host offline exchange activities and technical competitions to strengthen communication and cooperation among community members, jointly creating an open, shared, and innovative AI technology ecosystem.

MatrixOS is not only a technological innovation but also a heritage of wisdom. Let us witness and participate in this great transformation together. The story of MatrixOS has just begun, and we hope that more enterprises and individuals will pay attention to and support the development of MatrixOS, jointly promoting the upgrading and popularization of AI technology, contributing wisdom and strength to the "comprehensive AI transformation" of various industries, and embracing a more intelligent and interconnected AI future.