At the 2026 Sustainable Global Leaders Conference held in Shanghai from October 20 to 22, Chen Weiliang, founder, chairman and general manager of MetaX, delivered a keynote speech titled "Building a Solid Compute Foundation, Developing New Quality Productive Forces — Reflections and Practices of MetaX."
Chen pointed out that new quality productive forces are advanced productive forces led by technological innovation, characterized by high technology, high efficiency and high quality. They emphasize driving total factor productivity through technological breakthroughs, optimized allocation of production factors and industrial transformation and upgrading. Simply put, developing new quality productive forces is not just about increasing investment and expanding scale, but more importantly about relying on innovation to improve production efficiency and development quality.
In his view, computing power itself is an important manifestation of new quality productive forces, because computing power is a major achievement of technological innovation, and AI technology supported by computing power is also a critical foundation for other industries to develop new quality productive forces. Therefore, measuring the value of computing power requires not only looking at how strong the computing capability is, but also whether it can be converted into actual production efficiency and innovation outcomes. Achieving this goal requires coordination among chips, software, interconnect and computing systems.
Chen cited the Xijing S600 super node as an example, saying it is a practice by MetaX in system-level computing power, reflecting the idea of improving computing supply capacity and usage efficiency through system coordination.
Chen introduced that in the field of artificial intelligence, Jensen Huang proposed the "five-layer cake" model. The bottom layer is Energy, the foundation of all industrial production. Above it is Chips — GPU and other AI chips are the main representatives and the "engine" that converts energy into the computing power AI needs. MetaX works at this layer. The third layer is Infrastructure, also known as AI Infra, the software and hardware systems of intelligent computing centers and the physical carrier of AI factories. The fourth layer is Models — large models are the brain of AI and the core of intelligence. The top layer is Applications, where AI creates actual economic value.
He noted that from this structure, application demand transmits to models, infrastructure, chips and energy. To let computing capability continuously support industrial innovation, all these links need coordinated development, with overall consideration of resource supply and operating conditions. As AI moves toward large-scale application, energy supply has become a real constraint that computing infrastructure must face. According to a report released by the International Energy Agency this year, under the baseline scenario, global data center electricity consumption is expected to grow from 485 TWh in 2025 to 950 TWh in 2030.
Chen said that new quality productive forces emphasize high efficiency, which requires improving the effective output per unit of energy input through coordination among GPU, software, interconnect and scheduling, while meeting task quality and service requirements. This relates both to resource use efficiency and directly to long-term usage costs. Facing these energy and operating conditions, to make computing power sustainably support the development of new quality productive forces, it is necessary to focus on GPU performance while also calculating the long-term economic account of TCO. TCO mainly includes hardware procurement costs, energy costs of intelligent computing center operations, application migration and adaptation costs, and operations and maintenance management costs. Migration and maintenance also involve personnel training, stability verification and other investments, which need to be evaluated from a long-term usage perspective. Improving the utilization efficiency of computing resources is also an important path to reducing long-term operating costs.
Therefore, reducing long-term usage costs requires focus on several aspects: first, the generality of computing power, which supports diverse applications across thousands of industries; second, the ease of use of computing power, since only an open and compatible ecosystem can support rapid iteration and development of models and applications; third, self-reliance in core technologies and independent controllability of core supply chains.
He pointed out that high-quality computing infrastructure needs to convert equipment investment into continuously usable computing capability, so attention must also be paid to resource utilization efficiency and long-term stable operation. Chen proposed that AI computing demand has three major characteristics. The first is the need for generality. AI algorithm iteration has gone through three main stages, and the iteration cycle has compressed from "year-level" to "month-level." Only general-purpose computing power can support such rapid evolution. This is also the core reason why GPU has become the main source of AI computing power. Generality also helps improve the reuse capability and utilization rate of computing platforms, allowing computing resources to better serve industrial innovation.
The second is ease of use. In AI, many people discuss the CUDA ecosystem. The core advantage of the CUDA ecosystem is ease of use. Why is the ecosystem so important? First, developer barriers: after twenty years of accumulation, millions of developers worldwide deeply rely on this computing ecosystem. Second, hidden migration costs: in code rewriting, personnel training and stability verification, hidden costs sometimes exceed hardware costs. Third, network effects: frameworks, models, applications and customers are locked in layer by layer, and the stronger the ecosystem, the stronger it becomes, forming the commonly described "winner takes all." Therefore, lowering the threshold for learning and adaptation can also let more research institutions, enterprises and developers participate in industrial innovation and promote the combination of computing technology and industry knowledge.
The third characteristic is stability and reliability. It goes without saying that without computing security, there is no AI technology, and even less digital economy security. Self-reliant core technologies and independently controllable core supply chains are the guarantee for a stable and reliable computing foundation and the fundamental guarantee for the healthy development of China's AI industry and the digital-intelligent transformation of thousands of industries.
So how does MetaX do it? Chen introduced that around generality, ease of use, stability and reliability, MetaX's practices in GPU chips, software and computing platforms aim to build a world-class high-performance GPU chip and computing platform, providing the industry with domestic computing support that is "general-purpose, easy to use, stable and reliable." In terms of generality, MetaX adopts a GPU architecture, as GPU processors are the most convenient for programming. In terms of ease of use, MetaX's MXMACA software stack is compatible with CUDA, allowing customers to migrate at zero cost. In terms of stability and reliability, it has achieved full self-development from GPU IP cores and instruction sets to chip architecture, with a fully domestic supply chain closed loop.
Based on this system, MetaX has built five major product series. The Xiyun C series targets AI training and inference, the Xisi N series targets AI inference, the Xicai G series targets graphics rendering, the Xisuo X series targets scientific computing and scientific intelligence, and the Xijing S series targets intelligent computing centers and training-inference clusters. "Through this product layout, combined with supporting software tools and system support, we hope to provide corresponding computing choices for different task loads and better serve industrial innovation," Chen said.
In addition, to bring computing technology into industry applications, computing platforms need to be combined with model tools, industry knowledge and business processes. This also requires continuous collaboration between computing companies and industry partners. Around this process, MetaX has formulated a "1+6+X" ecosystem strategy. "1" is the digital computing foundation, building a general-purpose, easy-to-use, stable and reliable intelligent computing platform with MetaX general-purpose GPU. "6" is industry empowerment, empowering finance, healthcare, energy, education and research, transportation, and large cultural entertainment industries through the computing foundation. MetaX hopes to work with partners to combine computing capability, models and industry knowledge, support tasks such as information processing, auxiliary analysis, forecasting and scheduling, and content generation around specific business processes and R&D needs, promote innovation in R&D and service methods, and drive digital-intelligent upgrading of industries. "X" represents frontier directions, from embodied intelligence to space computing, where MetaX continues to explore new application directions with partners.
"Different industries have their own characteristics in development foundations and application needs. We hope to work with partners starting from specific tasks and specific scenarios, combine technology with scenarios, and explore application methods suitable for their respective business processes," Chen said. Through this layout, MetaX hopes to make computing power better serve real industrial needs, test R&D efficiency, service quality and resource use effects in applications, and transform computing technology innovation into capabilities that industries can continuously use, providing support for cultivating new quality productive forces and promoting high-quality development.
In Chen's view, new quality productive forces are themselves green productive forces. For a computing company like MetaX, requirements for sustainable development need to be integrated into the entire process of technological innovation and industrial application. In terms of environment, MetaX focuses on the resource use efficiency of computing itself and explores green innovation in energy and materials with partners. In terms of society, MetaX hopes to lower the threshold for learning, adaptation and R&D participation through open and compatible software tools, so that more research institutions and developers can make good use of computing resources and accumulate professional capabilities through collaboration. In terms of governance, MetaX has released an ESG management system. In the future, it will rely on this framework to implement sustainable development requirements into daily operations.
"The value of computing power must be tested in industrial applications and amplified through open collaboration," Chen said. He looks forward to continuing to join hands with industry partners, research institutions and developers to support industrial innovation with efficient computing power and turn technological progress into practical results of high-quality development.
All conference transcripts are compiled from on-site notes and have not been reviewed by the speakers. Publication is for the purpose of conveying more information and does not mean endorsement of the views or confirmation of the descriptions.
Comments