The rapid growth in demand for large model training and inference is redefining the importance of computing infrastructure.
In the past, when enterprises built computing centers, the focus was more on server quantity, single-machine performance, and computing resource supply. However, as AI applications move from model training to industrial implementation, the problems that computing infrastructure needs to solve also include whether computing power, storage, networking, energy, scheduling, models, and applications can be organized into a highly efficient and collaborative system.
This also means that competition in AI servers is shifting from point-to-point hardware comparisons to contests of system capability and ecosystem synergy.
Amid this change, hardware manufacturers represented by Lenovo Group Ltd. (OTC: LNVGY) are beginning to redefine their position within AI infrastructure.
Recently, the "Together, Ask the Sky Again: Lenovo Wentian Brand Renewal & Computing Ecosystem Conference" was held in Beijing. At the event, Lenovo Wentian officially announced its brand renewal strategy.
This marks an upgrade for Lenovo Wentian from a localized server brand to a computing infrastructure brand for the AI era.
Chen Zhenkuan, Vice President of Lenovo Group and General Manager of the China Infrastructure Business Group, stated that AI is currently evolving from a tool application to a production factor, and computing power is correspondingly upgrading from resource supply to a system capability for Token production. The entire industry is entering a new stage where competition is shifting from "capability competition" to "production paradigm competition."
The brand upgrade also corresponds to clearer business objectives. Chen Zhenkuan set a target for the China Infrastructure Business Group to achieve 100 billion yuan in revenue by 2027, aiming to become number one in China's server market.
Specifically, Lenovo Wentian hopes to use AI factories, hyper-intelligent converged computing power, and a full-stack product system to bridge the complete chain from computing power and data to models and applications, transforming originally scattered and complex AI capabilities into more standardized, scalable production capacity.
At this conference, Lenovo notably launched the Wanquan Heterogeneous Intelligent Computing Platform V5.0 and the HyperNode solution.
The Wanquan Heterogeneous Intelligent Computing Platform V5.0 is primarily targeted at large model training and inference scenarios, attempting to improve the efficiency of computing clusters through hardware-software synergy.
It is understood that this platform has achieved two core breakthroughs. The first is cluster training/inference acceleration technology, which enhances large model training and inference performance and improves cluster resource utilization through technologies like the layered decoupled PD separation architecture and KV Cache shared cache optimization. The second is chip-model compilation optimization technology, which adapts to the diverse computing chip ecosystem and improves the efficiency of the entire training and inference process through computation graph adaptive matching and operator auto-generation tailored to different models.
This also reflects a significant change in the current AI computing industry: competition in computing power is shifting from "resource supply" to "efficiency competition." A company's competitiveness will increasingly depend on the efficiency of producing, acquiring, and utilizing Tokens.
Comments