Huawei's ACT pathway powers scalable AI adoption across industries

Company unveils pathway for intelligent transformation

Sponsored Feature The transformative benefits of AI across industries today have turned enterprise decision-makers' priorities toward ensuring that investments generate real business value, harness enterprise data to bolster competitive advantages, and extend use cases beyond pilot projects to large-scale applications.

At the Huawei Connect 2025 event in Shanghai, the company unveiled a pathway for intelligent transformation that is aimed at enabling enterprises to adopt large-scale AI. The plan is based on five key findings derived from Huawei's project experience with customers. The "ACT" pathway entails:

• Assessing high-value scenarios: Choosing the right scenarios for implementation is critical to successful transformation beyond efficiency. Integrating AI with core production scenarios enable processes that deliver more intelligent products and services.
• Calibrating AI models using vertical data: To harness the power of AI, enterprises need to train and fine-tune models on massive amounts of high-quality proprietary data and customize the models for their industry-specific needs.
• Transforming business operations with scaled AI agents: Demand for large-scale inference continues to grow alongside AI agents. Human-AI collaboration will become a new organizational paradigm as AI systems work closely with employees. To ensure secure, sustainable and trustworthy AI, systematic governance and risk management mitigate risks introduced by AI agents, such as out-of-control autonomy and lack of traceability.

The findings that underpin the ACT pathway are drawn from vertical-industry success stories that showcased how AI has dramatically boosted industrial intelligence.

Model accuracy, efficient power

China Southern Power Grid used Ascend computing platform and MindSpore AI framework to develop a large model called MegaWatt for the power sector. The model efficiently trains and runs large-scale AI models, specifically those with a Mixture-of-Experts (MoE) Expert Parallelism (EP) cluster. Huawei MindSpore – an open-source deep learning framework that supports unified programming for AI software and hardware – enables easy development, efficient execution, and flexible deployment across cloud, edge and device environments.

China Southern Power Grid leverages vast amounts of data that it manages to identify defects in power transmission lines. Huawei used end-to-end data governance to clean, process, label and optimize the data for training. Further, Huawei's optimized operators were used to speed up training and improve model accuracy.

Computational load is distributed across multiple Ascend AI chips, alleviating bottlenecks common in traditional MoE systems. Combining computer vision and natural language processing, the MegaWatt system has helped the company to improve defect and risk identification efficiency during power line inspections by five times and to increase image recognition accuracy to over 90%. This case study shows that enterprises must train custom models to solve their specific business challenges and create lasting advantages.

Boosting health diagnosis, treatment

In healthcare, medical records are a critical part of the patient treatment process. Huawei worked with Runda Medical to improve medical record keeping by developing an AI medical record appliance solution that tapped on Ascend inference servers.

The solution recognizes doctor-patient conversations and then summarizes the patient's complaints, interprets the doctor's diagnoses, and generates accurate records that meet hospital standards. Combining open-source models for general tasks with industry-specific models for clinical understanding, the solution has helped West China Hospital to raise consultation efficiency with medical record generation time of about one second.

By coordinating with a pre-diagnosis AI agent and a medical record quality control AI agent to ensure efficiency and quality, the system allows doctors to finalize records with no more than four edits and send them to the hospital information system. With improved diagnoses and treatment efficiency, doctors and nurses have more time for interaction and communication with patients.

Implementing the ACT pathway

Driven by these findings, the ACT pathway begins with enterprises using Huawei's AI Scenario Assessment Framework to evaluate business value, scenario maturity and business-technology integration. The framework can be used to identify and implement more than 1,000 core AI production scenarios.

With the right scenarios identified, enterprises can begin calibration by building and training industry models with vertical data. This begins with data governance. Huawei provides a full toolchain to help enterprises transform raw data into knowledge and knowledge into models. For instance, Huawei offers MRS, a unified lakehouse platform. It allows enterprises to stream raw data into a data lake, then convert them into structured data warehouse assets or ready-to-use data. AI applications are secured by Huawei's security protection system that works on cloud, network, edge and device as well as models and applications.

The third step of the ACT pathway is to transform business operations through fast deployment of AI agents. Since enterprise processes entail many scenarios and heavy workloads, Huawei's one-stop Versatile platform automatically generates AI agents as well as workflows with more than 100 steps. Augmenting industries' move toward human-AI collaboration, an AI talent enablement program helps business professionals to effectively develop, deploy and operate AI agents.

AI-oriented ICT Infrastructure

The three steps of the ACT pathway require an AI-oriented ICT infrastructure that covers the entire process from data preparation and movement to model training, inference and development.

In data preparation and AI storage, Huawei's Unified Cache Manager plugin enables large models to move from limited, session-based memory to long-term memory. AI agents can retain and recall information over time, making their interactions more personalized and effective based on a persistent record of past interactions, preferences and long-term goals.

By enabling efficient retrieval of information from its long-term memory, AI agents can generate a response by up to 90% faster. Enterprises do not need to constantly re-run complex computations, which translate to faster services and lower inference costs.

Additionally, intelligent cluster computing centers require large-scale interconnects, efficient data flows, and stable training capacity. To this end, Huawei's 800GE high-speed networking solution supports clusters four times larger than the industry standard. A Network Scale Load Balance algorithm helps to increase network utilization from the industry's 80% average to 98% and training and inference efficiency by 10%.

For model training and inference, Huawei's SuperPoD offers a high-performance solution. The solution can support training and inference for trillion- and 10-trillion-parameter MoE models, ultra-long sequences, and multi-modality. With Huawei's SuperPoDs, training is three times more efficient than traditional solutions. Their inference performance is also four times greater than industry standard.

Ecosystem-wide Collaboration: Open source and open systems

The demands of the AI era will require future infrastructure to remove the unpredictability of traditional data center engineering and be AI-ready with productized and prefabricated industry-specific solutions.

In Huawei's quest for industrial intelligence, the company's partners play a pivotal role. Strategically, the Huawei plus Partners ecosystem has been continually strengthened with three key initiatives.

1. The use of open source. Alongside open systems like CANN and Mind series toolkits, the company also supports mainstream frameworks, which gives its partners greater freedom to optimize and innovate on Ascend computing.
2. Empowering partners with platforms and tools. These include a full suite of frameworks such as its training and inference framework, agent framework, and DataArts. The tools make it quicker and easier for partners to build intelligent applications.
3. Rapid replication through industry expertise. Huawei has worked closely with partners on joint development and marketing for more than 200 industry solutions, helping them speed up replication and delivery.

To date, this ecosystem includes over 6,300 Kunpeng partners, 2,700 Ascend partners, 70 consulting firms, and 750 ISVs.

The three initiatives have led to the development of cutting-edge industry solutions such as the City AI Center and Foundation Model Solution, Intelligent Computing Labs Solution, Medical Technology Digital and Intelligence 2.0 Solution, Banking AI and Foundation Model Solution, Intelligent Manufacturing R&D Solution, SMART Logistics & Warehousing Solution, Intelligent Distribution Solution, Intelligent Exploration and Development Solution for Oil and Gas, and Steel Blast Furnace Temperature Prediction Solution.

Sponsored by Huawei.