Huawei has launched a new full-stack data infrastructure offering for AI data centers.

Revealed during Huawei's Innovative Data Infrastructure Forum 2026, the launch includes the OceanStor Pacific Scale-Out Storage offering, which provides 11PB of storage capacity across a 2U space.

According to Huawei, this high-density storage offering can help reduce the total cost of ownership for customers.

In addition, for large-scale inference clusters, Huawei is offering a Context Memory Storage (CMS) which supports heterogeneous computing power, and can expand into a petabyte-scale shared KV cache pool, reducing the time to first token by up to 90 percent.

Speaking at the IDI Forum, Huawei's president of the Huawei Data Storage product line, Yuan Yuan, said: "AI is unlocking new opportunities for the IT industry. The next chapter of AI is data. Committed to technological innovation in data storage, Huawei will accumulate the experience of industrial AI adoption, and work closely with the entire industry to help customers accelerate their journey into the intelligent era."

In addition to the hardware launch, Huawei has announced its ModelEngine, which enables zero-code adaptation to new models and fast model deployment, along with compute resource partitioning and intelligence scheduling.

While Huawei uses the same technology in its Huawei Cloud, Yuan was keen to emphasize the importance of purchasing hardware to build AI infrastructure yourself. He said: "If you are really serious about bringing AI to your enterprise, you need to think about building the AI stack on-premises, to keep your privacy, complication modulation, and to generate the unique value by yourself. You don't have to just use public AI services; you need a private stack."

Yuan later told DCD: "As the hardware manufacturer, we hold a very optimistic view of the on-premise construction. Of course, many customers need to adopt cloud services to consume AI for training services, governance services, and leverage, using public cloud services, but there are still many customers because of their regulation requirements, privacy that, for them, it's not easy to put data on the public cloud. They need to build some on-premise infrastructure systems, and to create their unique value," for example, healthcare, governments, and manufacturers.

"Gradually, I think many bigger enterprise customers will adopt on-premise construction models even at an early stage," he added.

On the current memory chip shortage, Yuan noted that, for the most part, this actually benefits companies like Huawei. "Yes, there are global issues with the upward trend of memory and storage media, but from my point of view, this is good. We are storage vendors, and storage can turn into a lot of value for us." He noted that they follow the 'multiple vendor strategy' and choose different vendors globally to serve for different assets, and: "We have global collaboration with major providers, so we don't have any supply chain problems."

Currently, Yuan and Huawei are expecting the memory chip shortage to continue until at least 2028.

This week, Huawei also revealed its Grid Interactive AIDC strategy, which aims to deliver reliable, energy-efficient, fast-delivered, and grid-friendly AIDC solution that maximizes tokens per watt.