Huawei Connect 2025 was more than showing off new products. Eric Xu explained from the stage that export controls and supply chain constraints have forced Huawei to build its own domestic supply chain and define its own path to market for tech infrastructure or risk obsolescence. This turned Connect announcements into statements of strategic autonomy from Western technology hegemony, for Huawei and its customers.
Geopolitical Constraints Catalyze Eastern Innovation
Huawei’s push into AI infrastructure has been shaped by US export controls that block access to Nvidia’s high-end GPUs and ASML’s EUV lithography machines. These constraints prompted Huawei to innovate across the stack with:
- Ascend AI NPUs: The Ascend 910C powers Huawei’s CloudMatrix 384 “supernode” cluster, rivaling Nvidia’s GB200 NVL72 in compute performance by stacking 384 chips with high-speed interconnects, and Huawei has a long roadmap of improvements.
- ARM-Based Kunpeng Processors: Huawei’s in-house ARM CPUs now power laptops, servers, and AI infrastructure, reducing reliance on x86 and enabling tighter integration with its software ecosystem.
- HarmonyOS: Huawei AI-native operating system powers mobile devices, laptops, even AI infrastructure components. It supports real-time inference, intelligent scheduling, and context-aware services, making it a strategic asset in their push for technological sovereignty.
- UnifiedBus: Huawei opened up technical specifications for UB 2.0. The new spec is designed to link more than 10,000 NPUs in a SuperPoD and will scale to a million NPUs in future SuperClusters. UB is meant to outscale existing NVLink and RoCE based interconnects.
An AI Future With Ascend, Kunpeng, TaiShan, and Atlas
Ascend NPU Roadmap
The Ascend NPU roadmap extends into 2028 and each chip has corresponding SuperPoD and SuperCluster configurations. One of the dazzling promises include the Atlas 960 SuperCluster expected to achieve a Zettaflop of compute power with almost a million clustered NPUs in 2027.
- Ascend 950PR, Q1 2026: (prefill/recommendation) 1 PFLOPS FP8, 2 PFLOPS FP4, 2 TB/s interconnect, 128GB memory at 1.6 TB/s
- Ascend 950DT, Q4 2026: (decode/training) 1 PFLOPS FP8, 2 PFLOPS FP4, 2 TB/s interconnect, 144GB memory at 4 TB/s
- Ascend 960, 2027: 2 PFLOPS FP8 , 4 PFLOPS FP4, 2.2TB/s interconnect, 288GB memory at 9.6 TB/s
- Ascend 970, 2028: 4 PFLOPS FP8 , 8 PFLOPS FP4, 4TB/s interconnect, 288GB memory, at 14.4TB/s
TaiShan Product Portfolio
The TaiShan 950 SuperPoD is a general-purpose compute pod designed to handle enterprise-grade workloads like application servers, database operations, and enterprise workloads with AI-inferencing support. It will be powered by Kunpeng 950 processors, available in 96-core/192-thread and 192-core/384-thread variants. Expected in Q1 2026.
Atlas SuperPoD Servers and NPU Cards
Atlas SuperPoD nodes are focused on enterprise AI adoption and will compete with Nvidia-based AI Servers from Dell, HPE, and Lenovo. The product line maps to Ascend NPU availability and includes a PCIe 5.0 card option for upgrading existing servers.
- Atlas 850 SuperPoD Server: Air-cooled design powered by eight Ascend NPUs, clustering up to 128 nodes for a total of 1024 NPUs in a single SuperPoD. Expected in Q1 2026.
- Atlas 860 SuperPoD Server: Iteration on the 850 with higher memory bandwidth, and enhanced interconnect for more demanding inference tasks. Expected in Q1 2026.
- Atlas 350 NPU Card: Powered by the Ascend 950PR, it supports standalone use or connected via UnifiedBus in 2 or 4 card configurations allowing up to 512GB of memory for large parameter models. Expected in Q2 2026.
Storage Innovation With AI, Flash, and Improved Data Resilience
Huawei’s storage portfolio reflects growing capabilities in enterprise storage, data protection, and fleet level datacenter management with DME:
- OceanStor A600: High performance AI-native storage delivering 160 GB/s and 3.6M IOPS. It supports vector retrieval and multi-level KV cache to reduce both hallucinations and latency
- OceanStor All-Flash Dorado v7: Updated platform with PCIe 5.0, scalability to 500PB and 100M IOPS, and support for 32 controllers using SmartMatrix 4.0 for improved reliability.
- OceanStor Pacific 9926: Bringing all-flash performance to the Pacific scale-out storage portfolio. 8PB in a 2U chassis using 61TB SSDs, Expected September 30, with 122TB SSDs in 2026.
- OceanProtect Appliance: Enhanced data protection solution deployed as software and now in a new appliance offering.
- OceanCyber: Data security appliance for Huawei storage platforms. Unifies security policy management across devices using DME, enhanced ransomware detection, and secure recovery orchestration for OceanStor and OceanProtect systems.
Networking Is AI-Centric Switching and DPUs
Huawei’s networking innovations are designed to support massive AI clusters:
- Xinghe Intelligent Fabric 2.0: AI-Centric networking that includes network-scale load balancing, self-healing, and built-in security.
- CloudEngine Switches: 128×800GE switches with StarryLink optical modules enable high-speed SuperPoD interconnect.
- DPUs and Network AI Agents: Network-focused compute offload for CPUs and NPUs, with AI agents designed to automate network operations and enhance security.
What It Means For Enterprise IT Buyers
Huawei’s vertically integrated approach spans chip design, operating systems, and O&M software mirroring the ecosystem advantages seen in vendors like Apple and Oracle. While components can run independently, the portfolio is optimized for joint deployment, as demonstrated on the exhibition floor through use-case focused solutions.
For AI enablement, Huawei’s open-sourced AI stack with MindSpore, CANN, and openEuler come together to form a full-stack alternative to Nvidia’s CUDA, aiming to foster a collaborative ecosystem where others like AMD and Intel have struggled.
Huawei’s conference drew interest from customers well beyond APAC, with attendees from Latin America, Africa, Europe, and Middle East together to explore smart city, grid, and enterprise IT solutions throughout the show floor. Customers cited affordability in case studies as well as reliable platforms and strong partnership from Huawei for deployment. It was clear that for many enterprises Huawei is a serious infrastructure contender. However, businesses operating across regions or subject to compliance audits in regions where Huawei technology is restricted may face regulatory challenges. In such cases, Huawei’s platform could be more of a compliance headache than a cost-saving panacea.
Next Steps
If you’re evaluating Huawei as part of an upcoming infrastructure refresh or long-term tech strategy, reach out to Forrester for tailored guidance. Our analysts can help you assess Huawei’s fit for your business, navigate regional compliance risks, and benchmark its offerings against competitors.
