Economic Observer reporter Zhang Ying

Starting from the end of 2025, Huawei Pharmaceutical Corps will appear on major pharmaceutical industry forums. Does Huawei also need to make medicine? This is a question that many medical professionals have.

We don't make drugs. "On October 24, 2026, at the 11th Pharmaceutical Innovation and Investment Conference, Fan Jie, President of Huawei Pharmaceutical Corps, responded clearly.

He introduced that Huawei's pharmaceutical team is positioned as an AI enabler in the pharmaceutical industry, rather than competing with others to make drugs. Huawei comes from the field of computer communication, not biology or chemistry, so it focuses on leveraging its strengths.

Fan Jie believes that compared to other companies, Huawei's participation in AI pharmaceuticals has three major advantages:

Firstly, Huawei possesses computing power and storage capabilities, especially in providing advanced computing power that many enterprises do not have;

The second is the clustering ability of computing power, which is related to how to fully utilize computing power and how to use large computing power to increase data and obtain better results. Training a model requires at least 64 cards, often even hundreds or thousands of cards. But hundreds of cards cannot be used individually, they need to be strung together to form a cluster. Huawei has unique advantages in clustering and supernodes. Currently, Huawei has a super node solution with 1024 cards, which is the largest in the world;

The third is ecological advantage. Huawei has collaborated with companies such as Shenshi Technology, Wangshi Intelligence, and Molecular Heart to form a comprehensive supporting plan. Huawei is responsible for the coordination and adaptation of computing power, storage power, and network power, while other companies are responsible for model development.

Fan Jie introduced that in terms of cooperation with medical enterprises, Huawei has partnered with two companies to build a hundred card cluster, which is currently rare in the medical industry with an investment scale of nearly 100 million yuan.

At the same time, Huawei also empowers research institutions such as Shanghai Institute of Materia Medica and Changping Laboratory. Fan Jie introduced that Changping Laboratory has built an AI protein structure prediction model, TorchFold, based on Ascend computing power. Huawei provides adaptation and model fine-tuning services, and actual testing has found that it can align with the accuracy of AlphaFold.

He told the Economic Observer that the biggest challenge of the project is to solve the problem of model adaptation. There is a key layer between the model and GPU/NPU, which is the operator. The model calls the computing card and various modules through operators, and whether the operators are adapted directly affects the computing performance, which can lead to a performance gap of up to ten times. Previously, it took three to four months to do operator adaptation, but with the improvement of toolchains and AI programming, it has now been shortened to one or two days, greatly reducing the difficulty of model adaptation.