Kathy Gibson reports – Artificial intelligence (AI), with its ability to drive productivity and competitiveness, has become a critical geopolitical battleground.
With US sanctions preventing China from accessing some of the high-powered chips and chipsets required to run AI models, Huawei has been working to develop alternative solutions.
Diego Han, director ecosystem development at Huawei southern Africa, explains that until last year, organisations needed to invest significantly in order to run compute-hungry AI models.
“So they either had to spend a lot of money to buy infrastructure, or embrace the cloud,” he explains.
“But this year, after DeepSeek went public, people discovered they can run efficient AI models with less compute investment.”
Huawei, with its background in cloud-native technology and digital transformation, was successful in combining AI and large language models (LLMs) into its public cloud platforms.
“We come from an infrastructure background, so we provide the whole platform – the chipset, the servers, storage and switches,” he says. “Our full-stack technology is created by Huawei.”
Before 2019, Han explains that Huawei worked closely with Nvidia on chip technology. “But since then we have been forced to invest more in our own chipset. So now we have our own GPUs and NPUs, all integrated into our cloud platform.”
When DeepSeek came out, Huawei was quick to deploy it on its own platform. We were the first to provide a whole new hardware infrastructure hardware platform to DeepSeek, using more than 2 500 NPU chipsets to train the models.”
Today, the full series of DeepSeek models are available on Huawei Cloud.
Importantly, Huawei Cloud ensures not only DeepSeek’s performance, but its security as well.
“When DeepSeek went viral, everyone saw the performance benchmarks,” Han says. “What we didn’t see is that, as soon as it launched, it was massively attacked from around the world.
“We quickly realised how important the protection and security were, so in February and March we deployed hundreds of cybersecurity engineers to help DeepSeek counter these attacks.”
In the few months since DeepSeek launched, Huawei has also co-operated to integrate the model with its own portfolio of LLMs, namely CodeArts, KnowledgeArt, DataArts and ModelArts.
“We have added DeepSeek, so customers can run their whole environment on one platform,” Han explains.
For instance, CodeArts+DeepSeek adds coding efficiency and reduces the time to create code from weeks to minutes.
MetaStudio assists media creators with digital content production, processing and distribution; and ModelArts allows developers to quickly transform their model to a commercial product.
Han adds that companies everywhere are keen to embrace AI capabilities, but he cautions that they need to ensure their environment is ready.
“Customers need to do some work before they embrace the technology. If you don’t have connectivity or capable infrastructure to run and protect the data, it is very hard to employ AI.
“So you need to prepare the different layers of the infrastructure. Which is why Huawei talks about the whole data stack.”