Rain shifts its cloud to Huawei

Huawei has shifted business systems for billing, data analytics, storage, application connections and artificial intelligence onto Huawei platforms. Moving its cloud-based operations has extended the companies’ relationship beyond mobile network technology.

The South African mobile network operator began the programme in April 2024 to move systems supporting network usage, customer billing and other services into a public cloud environment.

“We decided to invest in our partner, Huawei, to assist us with moving [those systems] into a public cloud,” said Leon Nortje, Rain SA principal and senior architect, at Huawei SA Connect 2026 in Johannesburg last week.

The premier technology event focused on how large-scale AI adoption depends on strong foundations across networks, data and industry digitalisation. Huawei Cloud currently operates from three availability zones in SA and serves more than 300 customers.

Will Meng, Huawei SA CEO, said: “Large-scale deployment of AI cannot be built without solid digital infrastructure. As proven by global practices, developing economies seeking to advance intelligence must gradually shore up foundations including networks, data and industry digitalisation.”

Rain’s project provides a practical example of that approach. The operator had to redesign parts of the underlying architecture rather than simply transfer existing workloads onto another platform. Rain first focused on billing and charging systems. The operator replaced the billing platform in 2025 with a system based on TM Forum standards, which provide common frameworks and interfaces for telecommunications systems.

Rain also moved onto Huawei’s cloud-based online charging system (OCS), which calculates usage and customer charges while mobile services are being consumed.

Nortje said: “What that achieved and opened up for us is [that], with a cloud OCS, we could now scale our company in multiple ways.”

The architecture gives Rain room to support more customers without rebuilding the wider billing environment. The same cloud system could support operations in other countries, although international expansion is not a current priority. Rain then began upgrading the mobile network to take advantage of the additional capacity. The programme widened in September 2025, when the company began reducing its dependence on Amazon Web Services (AWS) for data analytics.

Much of the data warehousing moved onto Huawei Cloud GaussDB(DWS), a cloud data warehouse designed to store and analyse large volumes of business data. Rain transferred files and other large datasets onto Huawei’s Object Storage Service, which stores files and datasets as objects in the cloud. The migration led Rain to reconsider how the company classified and stored information.

“We found that a lot of our unstructured data was actually structured.”

Some structured information had been placed in object storage and then treated as unstructured data. Separating the two categories produced a substantial improvement in processing times.

“With the optimisation of keeping structured data structured, we were able to take a three-hour query down to two minutes.”

Rain had used Amazon Athena to query information stored in AWS. Nortje said Athena could return some results within 20-30 seconds, but the service could temporarily allocate about 170 servers to complete the task. Rain now runs the migrated data analytics workload across five servers. The company recorded a cost reduction, although greater operational control remained the central reason for the migration.

Rain moves API traffic to Huawei

Another stage followed in May 2026, when Rain replaced Google Apigee with Huawei Cloud API Gateway (APIG). Both platforms manage traffic and connections between applications and internal services.

Rain’s Apigee environment included 128 proxies and numerous connections across the business. The operator needed support for HTTP/2, which improves how web traffic is transferred, and WebSockets, which allow chatbots to maintain a continuous two-way connection with users.

“We trialled Huawei’s APIG gateway, which handles WebSockets [and] HTTP/2,” said Nortje. “Everything worked well, and we switched over.”

The migration began after Nortje approached Rain CEO Conrad Leigh and proposed removing Apigee from the business. Leigh agreed, but challenged Nortje to complete the project without a wider migration team.

The architect turned to Claude Code to study Huawei’s software development kit (SDK), inspect the existing Apigee proxies and map the connections that needed to move.

“I said to the AI: ‘There’s the SDK. Learn what it does. Here’s Apigee. Pull everything that you can from [Apigee], all the proxies, everything, [and] tie [the proxies] into the back end.’”

Claude Code helped Nortje identify the required changes and prepare a migration plan over a weekend. He began testing the proposed solution on Monday morning and had completed half of the migration by Wednesday. Rain completed the remaining work over the following two to three weeks. Developers generally needed only to redirect their applications towards the new proxy locations.

Nortje said Claude Code generated much of the required code from detailed prompts while helping him analyse the 128 proxies, map their back-end connections and prepare the migration. The AI coding assistant allowed one architect to carry out work that could previously have required a broader team of engineers.

Rain tests interchangeable AI models

Rain has started testing a models-as-a-service architecture on Huawei’s cloud platform. The approach allows the operator to replace one AI model with another without redesigning the surrounding application or infrastructure. Rain has tested Chinese AI models including DeepSeek, Alibaba’s Qwen and Zhipu AI’s GLM across different workloads.

“We can switch out the models that we need without changing too much of our infrastructure.”

The company has built a control harness around the AI environment. The harness limits the tasks available to a model and defines how the output should be evaluated.

“One of the big things about AI [is that], if you let [AI] loose, you get burned,” said Nortje. “What we’ve decided to do is build a harness around our AI.”

Another AI system tests the harness and compares the results produced by different models. Engineers can then determine whether a change in output comes from the selected model or the surrounding application. Nortje said models available through Huawei’s platform cost about one-tenth as much as comparable services from Anthropic and OpenAI for Rain’s workloads.

According to Huawei deputy general manager Kui Zheng, wider AI adoption would depend on collaboration between technology companies, partners and customers. He said: “The future of AI won’t be built by one company, even Huawei. We couldn’t do that. It will be built by an open ecosystem of partners, developers and customers creating value together.”

Huawei said fewer than 35% of SA enterprises have identified where AI can unlock value in their data. Rain’s programme shows one route from experimentation to operational use, with the operator first rebuilding the systems that support data, applications and billing before expanding into AI models.

The approach has allowed Rain to test different models without redesigning the wider environment, while the same Huawei-backed architecture supports the company’s existing operational systems.

* Jason Bannier is a data analyst at World Wide Worx and deputy editor of Gadget.co.za. Follow him on Bluesky at @jas2bann.