Huawei Verifies AI Inference Acceleration Solution...Token Throughput Improved by Up to 372%
Unveiled at MWC Shanghai 2026
Huawei has become the first company in China’s telecommunications industry to successfully increase token throughput for large language AI inference by up to 372% on a commercial network.
On the 29th, Huawei announced that, together with China Mobile Hubei, it unveiled these verification results at 'MWC Shanghai 2026,' held in Shanghai, China, on the 24th.
This solution is based on Huawei’s OceanStor A800 storage, Ascend A3 SuperPod, and Unified Cache Manager (UCM). Huawei explained that it has established a core technological foundation for efficiently deploying AI computing services.
As AI applications shift toward AI assistants, there has been an increase in long-context scenarios such as code generation and multi-turn conversations, which require handling lengthy input. The limited capacity of existing memory has been identified as a bottleneck in data processing. To address this, Huawei introduced UCM, which utilizes external high-performance storage, and implemented a petabyte (PB)-level KV cache.
The verification was conducted on China Mobile Hubei’s commercial network, simulating long-form input ranging from 8K to 190K tokens for major models such as Minimax M2.5 and GLM 5.1.
As a result, the GLM 5.1 model saw the time to first token (TTFT) reduced by up to 93%, while tokens per second (TPS) increased by up to 372%. For the Minimax M2.5 model, TPS increased by up to 78% in a 128K environment.
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
Michael Chu, President of Global Data Storage Marketing and Solution Sales at Huawei, stated, “The AI inference acceleration solution not only significantly reduces TTFT but also helps lower token costs, enabling telecom operators to build efficient and eco-friendly AI computing infrastructure.”
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.