
Huawei Unveils Upgraded Stellar AI Network to Power the Agentic World with Secure, Intelligent Connectivity
PRNewsWire
公開日時: Sep 18, 2026, 06:02 PM GMT+9
Sentiment Analysis
At HUAWEI CONNECT 2026, Huawei unveiled its upgraded Stellar AI Network Solution under the philosophy of secure, intelligent connectivity. With stand-out features such as efficient computing production, lossless delivery, and superfast application, this future-ready solution unleashes computing productivity while safeguarding services in the AI era.
Driven by explosive AI adoption, daily token consumption has surged 260-fold in a year. Huawei's upgraded Stellar AI Network—spanning AI Fabric, AI WAN, and AI Campus—responds to this by focusing on intelligent connectivity and security.
Stellar AI Fabric 2.0
Intelligent connectivity: To tackle challenges like surging networking costs, slow protocol conversion, and traffic retransmissions caused by link flapping, Huawei has introduced the UBG switch SF9300 series. Paired with Ascend A5/A6, it forms a full-series UBG compute-network integrated solution. Specifically, two-layer multi-plane architecture eliminates the core layer, cutting network deployment costs by 30%. The unified protocol reduces end-to-end latency from 20 μs to 11 μs. Link-Layer Retransmission (LLR) achieves zero packet loss during link flaps, reducing annual cluster service downtime by 100 hours.
Huawei also unveiled its fully in-house CloudEngine XH9300-EN series NPO switches, delivering three major breakthroughs. First, centralized light-source design lowers interconnect power consumption from 1,000 W to 600 W, saving cluster electricity costs. Second, near-packaged architecture reduces optical-electrical conversion latency by 180 ns, optimizing single-node latency by 26%. Third, pluggable snap-fit structure enables on-site replacement, shortening fault recovery time from days to hours.
Security: Huawei's Three-Layer Security Guardrail Solution defends against hashrate theft , prompt injection, and model poisoning. Specifically, Network Guardrail features an integrated cryptojacking detection model delivering a 95% detection rate with millisecond-level malicious traffic blocking. Agent Guardrail guards against prompt injection attacks with 95% accuracy at under 150 ms latency. Host Guardrail performs real-time monitoring of malicious processes while keeping CPU usage below 1%. At a real-world customer site, the solution identified 415 high-risk vulnerabilities and 322 threat events in its first seven days—leveraging AI to safeguard AI and ensure secure token production.
Stellar AI WAN
Intelligent connectivity: Cross-region computing faces issues like severe efficiency loss, high on-premises costs, and expensive private lines. To address these, Huawei's Stellar AI WAN Solution enables lossless delivery of computing power over 1,000 km: The Starnet lossless algorithm eliminates long-distance packet loss, boosting remote computing efficiency to over 95%. The XH computing-network appliance uses the layerwise model partitioning algorithm to run initial and final layers locally, transmitting only high-dimensional vectors over the network. This ensures raw data remains strictly on-premises and reduces initial branch xPU costs fourfold. Furthermore, computation-communication overlapping cuts bandwidth idle time by over 80% and reduces required bandwidth fivefold.
Security: Huawei's multi-dimensional WAN defense technology counters novel vectors like APTs and HNDL attacks. At the device layer, intrinsic security boards block intrusions in real time, trace threats within minutes at over 95% accuracy, and visualize up to 100-hop attack paths. At the link layer, built-in QKD delivers robust quantum security without extra devices, cutting construction costs by over 60% while extending transmission distance to 80 km. At the network layer, APN6-based data fencing enables network-wide control over traffic paths, ensuring the secure circulation of sensitive data.
Stellar AI Campus Netwo...
Source: PRNewsWire
個別の投資に関する推奨やアドバイスを提供することを意図しておりません。ここで述べられている意見や見解は、あくまでも各記事の個人的見解です。