Huawei Cloud Unveils ModelArts Next to Slash Enterprise AI Inference Costs

Huawei Cloud has intensified China's artificial intelligence efficiency race with the launch of ModelArts Next, a new-generation training and inference platform designed to cut enterprise AI deployment costs by an average of 20%.

Unveiled at the INSPIRE Innovator Conference, the platform introduces dynamic Model-as-a-Service (MaaS) routing capabilities. As enterprise demand shifts aggressively from large language model training to high-frequency commercial inference in 2026, Huawei's new architecture automatically allocates computational requests to the most efficient model based on real-time task characteristics. Initial deployment data shows this intelligent scheduling achieves an accuracy rate exceeding 95%.

Dynamic Routing Drives Cost Efficiency

The ModelArts Next ecosystem currently integrates over 15 State-of-the-Art (SOTA) models. Instead of relying on a single, resource-heavy flagship model for all queries, the platform's routing matrix functions as a computational traffic controller. By matching the precise processing weight required for specific corporate prompts, Huawei effectively mitigates the hardware bottlenecks and spiraling API expenses that have constrained broader AI adoption across Chinese industries.

Confidential Inference Secures Corporate Data

Beyond cost reduction, the upgraded platform deploys three additional core pillars: Reinforcement Learning (RL) services, a comprehensive model matrix framework, and confidential inference. The introduction of confidential inference directly addresses the strict data security mandates of financial institutions and state-backed enterprises. This capability allows sensitive corporate datasets to be processed and analyzed without exposing the underlying proprietary information to the broader cloud infrastructure, removing a critical hurdle for enterprise cloud migration.