AI Computing Power Demand Will Exceed $1.3 Trillion! Huawei Doubles Down on "Enterprise Lobster" Involving Multiple Listed Companies

“The future growth will definitely be in the computing sector.” Huawei China’s Enterprise Business President Wu Hui said in an interview with “Science and Technology Innovation Board Daily” and other media outlets, “AI is accelerating into core enterprise production systems. The ICT industry is expected to have an annual compound growth rate of 25.9% over the next five years, entering a rare period of significant growth.”

Huawei China’s Vice President of Enterprise Business Guo Zhenxing further pointed out that as intelligentization accelerates, the demand for AI computing infrastructure is expected to exceed $1.3 trillion by 2026, a 41% year-over-year increase.

“Computing power is no longer just for model training but also supports massive intelligent decision-making happening anytime and anywhere. Scenarios such as autonomous driving, AI for Smart Services (AI4S), financial risk control, industrial quality inspection, smart wearables, and AI PCs are driving demand for a full spectrum of computing power from ultra-large scale to edge. Intelligent agents have become the core driving force for comprehensive industry intelligence, and multi-agent collaboration is replacing single-point intelligence, becoming the optimal solution for handling complex enterprise tasks.”

According to the “Science and Technology Innovation Board Daily,” Huawei is collaborating with multiple listed companies to promote the implementation of “industry + AI” lighthouse solutions. In terms of computing power and applications, Huawei will open up series hardware such as modules, standard cards, servers, and supernodes to partners this year, and plans to launch a series of scenario-based “Enterprise Lobster” products.

Huawei to Launch Series of “Enterprise Lobster”

AI Agents are moving from concept to large-scale deployment. OpenClaw, with its strong autonomous execution capabilities, has become a globally renowned open-source project, rapidly iterating and shifting AI from “dialogue” to “autonomous task completion.”

Guo Zhenxing believes that practical Agents like OpenClaw are profoundly changing how individuals and enterprises work, ushering in a new era of AI “hands-on” capabilities. Meanwhile, AI intelligent agents are accelerating application innovation, causing token consumption to grow exponentially. Currently, the global daily token consumption exceeds 360 trillion. IDC predicts it will grow 300 million times over the next five years.

Huawei has internally developed over 600 intelligent applications and more than 19,000 digital employees, and is co-building industry intelligent agents with partners in automotive, healthcare, finance, transportation, and other sectors.

At the conference, Huawei Cloud’s intelligent agent development platform AgentArts Enterprise Edition announced it will begin public testing in April, and the open-source version openJiuwen Enhanced Edition will be officially open-sourced in May.

It is also revealed that Huawei Cloud will launch a series of scenario-based “Enterprise Lobster” products based on AgentArts, covering office assistants, code generation, marketing creativity, intelligent customer service, and more. Additionally, DataArts data intelligent agents will be launched soon.

Huawei Cloud has also introduced a computing power operating system—FlexNPU, a flexible intelligent computing platform that improves inference pool efficiency through AI Infra layer scheduling technology, providing greater token throughput with the same AI hardware investment. Furthermore, Huawei Cloud will release a next-generation 3D data center this year, breaking traditional coupling architectures of cooling, power supply, and cabinets, enabling cross-floor distributed deployment.

At the Wuhu base, this allows a single building to host 100,000 AI acceleration cards, doubling computing density. This data center will better support Huawei’s next-generation supernodes (comprising 8,192 chips), meeting enterprise demands for large-scale computing power.

For enterprises with sensitive data security needs, local deployment of “Lobster” has become a mainstream choice. Huawei Vice President and President of ICT Product Portfolio Management and Solutions Ma Haixu introduced that over the past month, more than ten partners have launched Tengyun-based Claw integrated machines, supporting over 100 clients to develop Agent applications based on openClaw in the first instance.

At the exhibition, several companies showcased local “Lobster farming” solutions based on Ascend chips. Huawei Kunzhengyu’s inference server HuaKun AT3500G3 supports billion-parameter inference and can host 50 lobsters, providing over 1.8 billion tokens per day.

The Softcom Huawei CLaw workstation all-in-one uses Huawei Ascend Atlas 300I A2 cards, capable of hosting “ten lobsters” simultaneously.

The Hongxin Ascend OpenClaw all-in-one is equipped with Kunpeng 920 modules and Ascend computing power.

Series of Modular Hardware, Servers, and Supernodes

Guo Zhenxing revealed that by 2026, Huawei will open up series hardware such as modules, standard cards, servers, and supernodes, working with partners to build a comprehensive scenario-based computing power foundation.

In large liquid-cooled data center scenarios, Huawei will open the Lingqu protocol and architecture, collaborating with partners to develop Lingqu computing, interconnected devices, and related components, jointly constructing supernodes to support large model training and center inference.

In enterprise-level air-cooled data centers, Huawei will launch Atlas 850E servers, 350 standard cards, and new Kunpeng 920 motherboards, providing reference designs to help partners develop large model integrated machines and adapt to various industry scenario applications.

In edge computing scenarios, Huawei will provide components, modules, application acceleration SDKs, and more, supporting partners in creating diverse edge intelligent devices.

The “Science and Technology Innovation Board Daily” reporter observed on-site that based on Huawei Ascend’s edge computing solutions, listed company Keso Technology has built a full-scale AI large model unmanned cluster system supporting heterogeneous collaboration across water, land, air, and space.

During the conference, seven core partners—Kunlun, Huawei Kunzhengyu, Shen Zhou Kun Tai, Yangtze River Computing, Baode, Softcom Huawei, and Baixin—launched complete machine products based on Ascend Atlas 350. As the first hardware product in the Ascend 950 generation, Ascend Atlas 350 uses the new Ascend 950PR processor, with upgrades in computing power, memory access, and microarchitecture.

Ascend also partnered with iFlytek, ChinaSoft International Education, Runda Health, Huitian Ruosheng, Fanwei, and Shenxinfu to release AI application scenario solutions, focusing on healthcare, finance, education, legal, and other industry scenarios.

Huawei’s Executive Director and ICT BG CEO Yang Chaobin stated that in lightweight AI deployment scenarios emphasizing rapid deployment and effectiveness, Huawei has fully upgraded Ascend all-in-one machines, with tiered and graded configurations to meet different model needs. Currently, applications from over 350 partners have been implemented in key areas such as education, healthcare, finance, and government.

Huawei also announced eight flagship “industry + AI” solutions for 2026 and launched a pioneering initiative for “industry + AI” all-in-one solutions targeting lower-tier markets.

Guo Zhenxing introduced that Huawei is jointly developing “government services and urban intelligence hubs” with Nanwei Software, iFlytek, and others, successfully deploying in Guangzhou, Shenzhen Futian, and other locations; with YunDing Technology, XCMG Group, and others, launching “Mining and Industrial Automation” and “Mine Unmanned Driving” solutions, saving over 100 million yuan annually with autonomous vehicle fleets; and collaborating with Runda and Huimei on “Medical Technology Digital Intelligence” solutions, which have been implemented in Southern Hospital, West China Hospital, and other top-tier hospitals.

Embodied Intelligence Zone to be Launched Soon

Huawei Cloud announced that this year it will focus on building industry AI “dream factories” in fields like smart healthcare and embodied intelligence. Among them, the Industry AI Dream Factory—Embodied Intelligence Zone—is expected to be released in April, covering more than 30 embodied models, becoming a hub for cerebellum and brain scenario development platforms.

The “Science and Technology Innovation Board Daily” learned that Huawei has partnered with Shanghai, Shenzhen, and other regions to participate in local robot training ground construction. Relying on Huawei Cloud’s embodied intelligence virtual training ground, a complete closed-loop can be built, including environment simulation, data augmentation, model training, and validation. Additionally, deploying robot core models based on Ascend platform will help realize “brain” and “cerebellum” capabilities.

At the Daohe Tongtai booth, the reporter saw Huawei collaborating with the company to build an integrated aerial-ground solution. Based on the intelligent agent framework, it enables task orchestration and supports routine, all-weather autonomous drone patrols. Using embodied large models and business large models, the system can autonomously plan paths and make decisions, supporting cluster-based autonomous inspections. When anomalies are detected during aerial patrols, nearby ground robots are automatically dispatched for detailed inspection.

Softcom Tianqi also showcased embodied intelligence data collection solutions.

It is reported that Hangzhou Pilot Base plans to establish a national-level embodied intelligence scenario application comprehensive test site, with Softcom Tianqi participating in the scene verification center construction. Focusing on “manufacturing consumption” and “service consumption,” they are creating intelligent operation lines and health and wellness companion scenarios, using self-developed robots for scene data collection and training, accelerating the transition from laboratory to real-world applications.

(Article source: Cailian Press)