"It is the best of times, and it is the worst of times." On April 11th, Zhang Xiuzheng, Vice President of Huawei and Minister of China Cloud Business Department, presented such a view at the Huawei Cloud Product Technology Release Conference. Entering 2025, global situations are changing rapidly, bringing anxiety and unease. However, simultaneously, since the beginning of the year, AI applications have continued to explode in China, allowing more people to see China's strength in seizing the high ground of artificial intelligence.
On that day, Huawei Cloud released its newly upgraded Ascend AI cloud services, software development production line CodeArts, digital content production line MetaStudio, cloud-native security system, data governance production line DataArts, GaussDB database, and other products and solutions for partners and customers.
Zhang Xiuzheng, Vice President of Huawei and Minister of China Cloud Business Department, particularly pointed out regarding this product technology conference that Huawei Cloud will continue to make strong investments in building equal diversified computing power infrastructure, supporting hundreds of models and thousands of variations of Ascend AI cloud services, reshaping applications with core software and toolchains, and being the "black soil" of the intelligent world in four major aspects. This aims to solve complex challenges such as computing power, illusions, data, and security when implementing AI, accelerating the intelligent upgrade of thousands of industries.
Vice President of Huawei and Minister of China Cloud Business Department, Zhang Xiuzheng
From "Stacking Chips" to "Assembling Architectures"
Computing power is the core resource in the AI era. Zhang Xiuzheng pointed out: "We should exchange space for computing power, bandwidth for computing power, and energy for computing power. We should consider computing power, storage power, transportation power, and electrical power as a composite element. This systematic thinking is particularly evident in the newly released CloudMatrix 384 Super Node architecture."
Different from previous industry solutions, the CloudMatrix 384 Super Node technology leaps out of single-point technological constraints to achieve systemic and engineering innovation, transitioning from "stacking chips" to "assembling architectures," taking a leap forward in breaking through AI computing power blockades, providing opportunities to solve many AI application implementation challenges.
Zhang Xiuzheng emphasized: "With breakthroughs in architectural innovation, Ascend AI cloud services have achieved parity with leading levels in the industry in terms of inference efficiency and model accuracy, and have surpassed the industry in some fields like autonomous driving. This may completely end everyone's computing power anxiety."
On the same day, Zhang Xiuzheng stated that Huawei Cloud's CloudMatrix384 Super Node will go online with tens of thousands of scales this year. In the future, CloudMatrix super nodes can build large clusters exceeding ten thousand chips to provide computing power.
It is understood that SiliconFlow has jointly launched DeepSeek-R1 with Huawei Cloud based on the CloudMatrix 384 Super Node Ascend cloud service and the high-performance inference framework SiliconLLM. After validation by mainstream test sets and large-scale online blind tests, the model precision of DeepSeek-R1 deployed on Ascend computing power remains consistent with the official version of DeepSeek.
Full-line Product Intelligence Upgrade
Currently, large models are evolving toward multimodal, long sequences, MoE, and deep thinking, with model sizes reaching trillions or even tens of trillions of parameters, causing a surge in computing power, network, and storage requirements, driving the reconstruction of AI infrastructure systems. You Peng, Minister of Huawei Cloud EI Service Product Department, pointed out that to address the "three highs and one low" challenges of the AI large model era, Huawei Cloud has reconstructed AI infrastructure based on full-stack systematic innovation capabilities, comprehensively upgrading Ascend AI cloud services focusing on four aspects: computing power base, resource scheduling, inference services, and cluster reliability.
In terms of computing power base, Huawei Cloud officially launched the new generation of Ascend AI cloud services based on the CloudMatrix 384 Super Node cluster with the new high-speed bus architecture. It provides an ultra-large-scale, high-performance, stable, and reliable AI computing power cluster to meet the complex training and inference needs of large models. This also makes Huawei Cloud the only cloud provider in the industry using peer architecture super node technology to provide computing power services.
Large models generate massive data demands, but data management faces multiple challenges such as managing large-scale databases and ensuring data security. To address this, Huawei Cloud has comprehensively upgraded GaussDB database to meet the data management requirements in the AI era. Li Yuzhang, Minister of Huawei Cloud Database Services Product Department, introduced that the native distributed performance of GaussDB has been improved by 50%, providing PB-level data management capabilities, and online expansion speed has increased by 10 times; Huawei Cloud has launched GaussRecorder to achieve database anti-ransomware and zero data loss; it has launched an AI-enabled database intelligent assistant to make development and maintenance more convenient.
In terms of data and large model application security issues, Huawei Cloud has built a complete cloud-native security system, including one center and seven layers of defense lines, achieving "no system collapse during attacks, no data loss, and regulatory compliance." Zuo Wenshu, Minister of Huawei Cloud Security Service Product Department, stated that Huawei Cloud empowers the cloud-native security system with AI, releasing four major security solutions, including the industry's first E2E large model security solution, an all-in-one data security solution, a full-scenario security operation solution, and an all-in-one edge security solution, making digital intelligence trustworthy.
To create an intelligent-driven, self-innovative, and secure software toolchain, Xu Feng, Minister of Huawei Cloud PaaS Service Product Department, showcased a series of CodeArts intelligent upgrade experiences covering code intelligent generation and explanation, code intelligent inspection, code problem intelligent repair, test point intelligent generation, and intelligent Q&A scenarios, improving end-to-end R&D efficiency by 30%.
Focusing on the intelligence of the media industry, Huawei Cloud's MetaStudio digital content production line has also upgraded intelligent interactive digital humans, deeply integrating digital humans and AI large models. On one hand, it continuously improves the model training capabilities of digital humans and integrates smarter large model brains based on voice recognition and interaction capabilities to achieve multimodal, human-like, and highly precise intelligent interactions; on the other hand, it adheres to the integrated strategy and can be quickly integrated with just one SDK and five lines of code, and currently Huawei Cloud has delivered over 10,000 digital humans in collaboration with partners.
For the acceleration of terminals moving to the cloud in the AI era, Huawei Cloud has developed a strategic product called CloudDevice. Hu Jianjun, Minister of Huawei Cloud aPaaS Service Product Department, introduced that CloudDevice achieves competitive computing power services through infinite computing power + storage power + electrical power + transportation power on the cloud side, featuring ultra-low latency of 60ms end-to-cloud, accessible AI entry points, seamless screen transitions, and opening up more innovative paths for intelligent applications under AI empowerment.
Assisting DeepSeek in Landing Industry Scenarios
Huawei Cloud continues to improve Ascend AI cloud services, adapting to hundreds of models and thousands of variations, and being the "black soil" of AI applications. Currently, Ascend AI cloud services have fully adapted to the main 160-plus large models in the industry, serving more than six hundred innovative pioneer enterprises. Among them, the DeepSeek series models have also been launched on Ascend AI cloud services, and with full-stack optimization adaptation based on Ascend AI cloud services, they can achieve effects comparable to those of global high-end GPU deployments.
Based on the accumulation of experience in over 30 industry scenarios and over 1000 scene models, Huawei Cloud's Ascend AI cloud services provide a full-chain DeepSeek professional service system, covering six major service scenarios and over 30 service capabilities, including consulting, deployment and implementation, scenario development fine-tuning, distillation model development, large model development incremental pre-training, and training. These capabilities are embedded into three engineering tools: data, models, and engineering. While providing powerful, elastic, and sufficient computing power, it also offers developers efficient, flexible, and stable AI inference computing power services and meets business commercial deployment needs, helping enterprises achieve cost reduction and efficiency improvement, and accelerating the rapid deployment and landing of intelligent businesses.
Deploying DeepSeek on Ascend AI cloud services has multiple advantages, including full-stack domestic computing power, local deployment solutions from cloud to edge, rich application scenario solutions, powerful cloud services, and full-process professional services. It has become the preferred choice for enterprises to accelerate DeepSeek application innovations. By March 2025, over 40 companies have deployed DeepSeek on Ascend AI cloud services and applied it deeply in their business scenarios.
Wang Wei, COO of Sina Weibo, CEO of Sina Mobile, and Chairman of the Sina Weibo Technology Committee, stated at the conference that products such as the Weibo Intelligent Search Deep Thinking Edition and Comment Robert run on the computing power foundation of Huawei Cloud Ascend AI cloud services. Meanwhile, Huawei Cloud provides powerful and rapid elastic scaling for Sina Weibo, offering computing power support during Weibo's trending topic traffic peaks; it also provides hardware-software synergy capabilities, increasing the utilization rate of Sina's clusters by 40%.
To enable small and medium-sized enterprises to use large models effectively, Huawei Cloud has launched the Flexus enterprise large model application solution, which significantly enhances the reliability and user experience of large model applications based on Flexus cloud services with performance doubling and experience upgrading. Through Solution as Code, it enables one-click deployment of scenario-based solutions, greatly reducing the threshold for SMEs to apply large models and driving business growth.
"Opportunities are more important than being well-prepared. Huawei Cloud insists on building a solid computing power base for China, providing a second choice for the world. We persist in solving problems and doing difficult things, hoping to work together with all ecosystem partners and customers to ride the wave of the great era and seize the big opportunities of the intelligent era!" Zhang Xiuzheng said. In the future, Huawei Cloud will continue to open up its full-stack AI capabilities, providing more affordable computing resources, more agile development tools, and a safer service system to create fertile ground for AI innovation in thousands of industries.
This article is an exclusive article of Guancha Observer, unauthorized reproduction is prohibited.
Source: https://www.toutiao.com/article/7492320050574213644/
Disclaimer: The views expressed in this article are solely those of the author and welcome your opinions to be expressed through the 'thumbs up/thumbs down' buttons below.