Press Launch
Constant innovation in AI Compute Service: Unleashing highly effective compute within the clever period.
This yr, Huawei Cloud introduced its AI Compute Service powered by CloudMatrix384. The specs of the Huawei CloudMatrix supernode can be upgraded from 384 playing cards to eight,192 playing cards. The supernodes can help a hyperscale cluster operating on 500,000 to 1 million playing cards, thus offering strong AI compute, a useful useful resource within the clever period. Huawei Cloud additionally introduced progressive reminiscence storage with its Elastic Reminiscence Service (EMS), which achieved an {industry} first by increasing video RAM with reminiscence. This drastically reduces the latency of multi-round conversations on basis fashions, vastly enhancing person expertise.
With quickly growing demand for AI compute, conventional information facilities are now not ample to function environment friendly AI infrastructure. To help AI workloads, the facility consumption of a single cupboard will must be upgraded from 10 kW to 70 kW, and presumably as excessive as 200 kW, and warmth dissipation might want to shift from air cooling to full liquid cooling. Huawei Cloud has deployed totally liquid-cooled AI information facilities in China’s Guizhou, Internal Mongolia, and Anhui. These AI information facilities help 80 kW warmth dissipation per cupboard, cut back energy utilization effectiveness (PUE) to 1.1, and provide AI-enabled O&M. This implies enterprises don’t have to reconstruct conventional information facilities or construct new ones. As an alternative, they require solely a pair of optical fibers in an effort to hook up with the info heart and entry environment friendly AI compute, in addition to full-stack devoted AI cloud providers, on Huawei Cloud.
Zhang Ping’an identified that Huawei Cloud’s AI Token Service abstracts away the underlying technical complexity and instantly gives customers with the ultimate AI computing outcomes. This enables customers to make the most of the inference computing energy in essentially the most environment friendly means doable. The CloudMatrix384 supernode realizes the complete pooling of compute, reminiscence, and storage sources, decouples compute duties, storage duties, and AI professional techniques, and converts serial duties into distributed parallel duties, vastly enhancing the inference efficiency of the system. In eventualities involving inference duties with totally different latency necessities, reminiscent of on-line, nearline, and offline inference, CloudMatrix384 delivers a median inference efficiency per card that’s 3 to 4 occasions that of H20.
On the convention, Zhang Ping’an introduced the official launch of the AI Token Service powered by CloudMatrix384. The service delivers superior efficiency, service, and high quality to prospects.
As of now, the variety of prospects worldwide utilizing Huawei Cloud’s AI Compute Service has elevated from 321 in 2024 to this yr’s 1,805. Qihoo 360’s Nano Search has used the CloudMatrix384-powered token inference service to course of tens of hundreds of content material technology requests every single day.
Zhang Ping’an famous that Huawei Cloud will proceed to advance software-hardware synergy and structure innovation, in order to construct cloud providers that continually create worth for his or her prospects and all the time preserve them up-to-date with the most recent applied sciences. As well as, Huawei Cloud will make investments extra sources into AI and computing, in an effort to domesticate the fertile floor on which totally different industries’ functions can run smoother and higher. Within the AI period, Huawei Cloud is dedicated to pursuing innovation in AI computing energy and applied sciences, serving to each buyer grow to be an AI pioneer of their {industry}.
Zhang Ping’an, Huawei’s Government Director of the Board and CEO of Huawei Cloud
Tackling challenges head-on: Serving to enterprises construct their very own fashions
Huawei Cloud has been honing its Pangu Fashions by diving into industry-specific eventualities, and has labored with its prospects to sort out their most urgent challenges head-on, reimagining what is feasible in these industries. Huawei makes use of openPangu to supply finest practices for AI coaching and inference, making it simpler for builders to effectively use AI computing energy. Zhang Ping’an famous that, on the identical time, Huawei is growing the closed-source Pangu Mannequin. Huawei will frequently improve funding in Pangu Fashions, continually research {industry} eventualities to raised perceive buyer necessities, and help prospects in growing their very own industry-specific fashions, thus accelerating clever transformation throughout industries.
To higher align the incremental coaching of fashions with enterprise eventualities, Huawei Cloud integrates {industry} know-how into ModelArts, a one-stop AI growth platform. ModelArts gives data relating to the optimum information ratios for {industry} duties, the optimum configurations for incremental coaching, and a system for assessing mannequin effectiveness and accuracy.
Zhang Ping’an additionally identified that information high quality determines mannequin high quality. Huawei Cloud makes use of the DataArts information governance pipeline to mechanically extract multi-modal information semantics and construct converged data graphs that kind a unified data lake for enterprises. This gives complete, wealthy, and constant enterprise corpuses for big mannequin pre-training. As well as, Huawei Cloud gives one-stop toolchains to rapidly generate high-quality tuning datasets with chains of thought (CoT) for mannequin post-training.
China Southern Airways makes use of the Pangu Mannequin’s cross-modal prediction know-how to foretell passenger site visitors and common ticket costs over an 18-month interval. Prediction accuracy reaches 90%, considerably enhancing the airline’s working income. The Chinese language Academy of Agricultural Sciences (CAAS) has developed the Agricultural Scientific Discovery Mannequin based mostly on the Pangu basis mannequin and Pangu Scientific Computing Mannequin. This mannequin is skilled utilizing huge quantities of agricultural data and tens of billions of organic information entries, shortening the R&D cycle by greater than 50%.
Pangu Fashions have been utilized in additional than 500 eventualities throughout over 30 industries. They’ve performed a big function in fields like authorities providers, finance, manufacturing, healthcare, coal mining, metal, railways, autonomous driving, and meteorology.
Transferring beyond terminals: Enabling infinite intelligence evolution on the cloud
This yr, Huawei Cloud launched the CloudRobo Embodied AI Platform, which deploys advanced algorithms and clever logic on the cloud to appreciate extra light-weight robots. By profiting from the large computing energy and superior AI fashions on the cloud, the platform makes robotic execution extra clever. Cloud intelligence overcomes the restrictions which have been holding robots again, making them relevant to extra eventualities.
Within the growth part, CloudRobo makes use of a cloud-based, embodied AI coaching information technology platform to mechanically generate operation trajectories for various poses, textures, and lighting situations, vastly enriching information range and growing the artificial information scale by hundreds of occasions. This not solely vastly reduces information assortment prices, however successfully improves the generalization functionality of fashions. CloudRobo additionally gives cloud-based atmosphere notion, international planning, and autonomous execution capabilities, reaching a job execution accuracy exceeding 90%. With the help of CloudRobo, the Nationwide and Native Co-Constructed Humanoid Robotics Innovation Heart in Shanghai makes use of information collected from actual machines alongside artificial information to coach robots, enhancing the general sorting success charges of robots in a number of eventualities to over 90%, which is an industry-leading success fee to date.
To construct a unified, open, and safe communication channel between robots and the cloud, Huawei Cloud has launched the Robotic to Cloud (R2C) Protocol, which realizes the collaborative pondering and environment friendly execution of AI brokers. As of immediately, nationwide requirements are in progress relating to information assortment, information technology, and communication interfaces across the R2C Protocol. Throughout his keynote speech, Zhang Ping’an introduced that the primary 20 companions of the R2C Protocol have been formally onboard, together with Hualong, Flexiv, Yijiahe, and Avant Robotics. Huawei Cloud works alongside its companions to develop cloud-based built-in robotic options based mostly on the R2C Protocol in fields reminiscent of manufacturing, logistics, inspection, and healthcare.
Kunpeng Cloud Companies: Empowering {industry} innovation with software-hardware synergy and an open ecosystem
One among Huawei Cloud‘s key methods is to develop Kunpeng-powered ARM cloud providers that ship efficiency, safety, and reliability. Up to now yr, the variety of Kunpeng compute cores on Huawei Cloud has elevated from 9 million to fifteen million, a rise of 67%. As well as, the Kunpeng platform has been repeatedly improved to help compatibility with mainstream software program, and has additionally been tailored to greater than 25,000 functions. The Kunpeng platform gives strong help for Kunpeng Cloud Companies to be utilized to much more general-computing eventualities, along with transcoding, databases, net functions, and cloud telephones.
GaussDB: Constructing environment friendly, dependable information foundations based mostly on supernodes and full pooling
Based mostly on general-purpose computing supernodes, Huawei Cloud’s GaussDB databases notice the layered pooling of compute, reminiscence, and storage sources, and permit multi-read and multi-write on any node on the identical time, breaking free from the restrictions of the normal structure the place solely the first node helps information learn/write. GaussDB databases additionally help dynamic load scheduling, vastly enhancing the efficiency of concurrent transaction processing. A GaussDB cluster deployed based mostly on computing supernodes can course of 5.4 million transactions per minute, marking a 2.9-fold efficiency improve over non-supernode clusters.
All-scenario distributed cloud: Ubiquitous and absolute best compute with native entry
Huawei Cloud has constructed a distributed cloud answer masking all eventualities, together with CloudOcean, CloudSea, CloudLake, and CloudPond. This covers central areas, hotspot areas, and edge websites, bringing constant Huawei Cloud expertise to wherever prospects’ enterprise is positioned. Huawei Cloud and FAW-Volkswagen have labored collectively to efficiently implement China’s first automotive manufacturing facility based mostly on the distributed cloud answer. The answer overcomes geographical constraints and realizes complete, in-depth collaboration between the HQ and 6 factories unfold throughout 5 cities, enabling seamless information move and clever manufacturing.
Constructing an easy-to-use, efficient, and open platform for growing and operating brokers
AI brokers are remodeling into a brand new software paradigm within the AI period. With this in thoughts, Huawei Cloud has launched Versatile, an enterprise-grade agent platform. It goals to function an easy-to-use, efficient, and open platform for growing and operating AI brokers. With this platform, prospects can be higher geared up to rapidly develop AI brokers that go well with their software eventualities.
Many nodes are included within the agent growth course of, involving extremely advanced companies. Based mostly on Versatile, customers merely want to arrange and enter enterprise description paperwork and flowcharts. After easy affirmation, the agent will be generated in two steps, vastly enhancing technology effectivity. Brokers mechanically invoke quite a few applications when they’re working, inflicting extreme site visitors jitters. Versatile additionally makes use of system improvements reminiscent of high-performance sandbox and full-memory cache to scale back end-to-end response latency inside companies by greater than 40%, enabling brokers to run stably and effectively.
Steady innovation: Empowering prospects to reach international markets
As industries worldwide rush to embrace digital and clever transformation, Huawei Cloud persistently pursues technological innovation as its core driving pressure and is dedicated to offering secure, environment friendly, and safe cloud providers for international prospects.
Charles Yang, Senior Vice President of Huawei and President of Huawei Cloud International Advertising and Gross sales Service, delivered a keynote speech relating to this matter. He highlighted the truth that enterprises are going through quite a few advanced challenges within the digital intelligence period. With this in thoughts, Huawei Cloud identifies its core positioning and adheres to this positioning. Huawei Cloud gives a holistic help system that includes “one international cloud infrastructure and two engines – one with information and one with AI” for its international prospects.
The primary of those is KooVerse, a world cloud infrastructure that provides prospects elastic cloud providers. Second are information enablement providers, which effectively provide information and unleash information worth. Third is a one-stop AI growth platform that helps enterprises construct AI-native functions and innovate with higher ease. Lastly, Huawei Cloud brings collectively the perfect practices in international digital and clever transformation, and gives them by means of cloud providers to assist extra prospects succeed.
Charles Yang, Senior Vice President of Huawei and President of Huawei Cloud International Advertising and Gross sales Service
DeFacto, a prime quick trend retailer in Türkiye, faces a site visitors surge of greater than 10 occasions throughout every promotional season, and its conventional IDC structure struggled to deal with this surge. DeFacto due to this fact turned to Huawei Cloud’s container service, which may mechanically scale up 4,000 pods inside 30 seconds, permitting the corporate to simply address site visitors spikes.
Based mostly on Huawei Cloud’s DWS, Neogrid, Brazil’s largest information intelligence and know-how firm, migrated and built-in its information that was scattered throughout a number of platforms, enhancing information integration effectivity by 40% and information evaluation effectivity by 50%. Right this moment, this offers Neogrid’s prospects the power to acquire the day’s information earlier than the top of labor and due to this fact make enterprise choices earlier.
Based mostly on the highly effective generalization functionality of Huawei Cloud’s Pangu Prediction Mannequin on industrial time-series information, Wanhua Chemical, a world chief within the chemical {industry}, has precisely captured information on greater than 2,000 key gadgets, permitting them to rework their reactive upkeep mannequin into predictive upkeep. Mannequin prediction accuracy has improved from 70% to 90%, and anomaly identification effectivity has elevated by 10%.
Huawei labored with companions reminiscent of China Unicom to construct “One Cloud, One Community, One Platform” for Changan Vehicle, driving versatile manufacturing with information. Changan Vehicle makes use of 5G and IoT to attach 12,000 gadgets, permitting it to interconnect information from totally different domains. Based mostly on the great information providers of Huawei Cloud, Changan has constructed a unified information platform. This platform breaks aside the info silos, permits environment friendly manufacturing operations, and shortens order supply lead time from 21 days to fifteen days.
Along with offering end-to-end service capabilities that span the cloud basis, know-how enablement toolchains, and experiences, Huawei Cloud gives tailor-made consulting {and professional} providers, in addition to automation instruments and platforms for enterprises’ digital and clever transformation.
Charles Yang expressed that digital and clever transformation can by no means be achieved by one participant in a single transfer. Slightly, it requires shut, steady collaboration all through your entire journey. With its applied sciences, experiences, and providers, Huawei Cloud is able to work with international enterprises at each step of their transformation journeys.
HUAWEI CONNECT 2025 runs from September 18 to twenty on the Shanghai World Expo Exhibition & Conference Heart and Shanghai Expo Heart. This yr, the occasion explores the theme of “All Intelligence” and delved into AI throughout three dimensions: technique, know-how, and ecosystems. This concerned an in-depth take a look at Huawei’s newest strategic initiatives, and the disclosing of a collection of all-new digital and clever infrastructure merchandise, scenario-specific options for industries, and growth instruments.
Along with keynote speeches given on the occasion, Huawei Cloud introduced a wide range of agendas, reminiscent of summit boards and roundtables. Huawei Cloud additionally works with prospects and companions to exhibit a big selection of progressive applied sciences and practices in fields reminiscent of cloud infrastructure, massive fashions, databases, AI brokers, and embodied AI, demonstrating how know-how can facilitate the digital and clever transformation of industries.
