The annual GLOMO Awards ceremony, one of the most prestigious awards in the mobile technology world, was held at MWC26 in Barcelona. This year, the South Korean company SK Telecom won in the «Best Cloud Solution» category with its GPU cluster named «Haein.» This marks the company's third consecutive win in this category, each time with a different product.
What Is «Haein» and Why Is It Needed?
Simply put, «Haein» is a very powerful computing center specifically built for artificial intelligence tasks. Inside, it houses over 1,000 NVIDIA B200 graphics accelerators (these are Blackwell series chips, the latest generation at the time of launch) combined into a single cluster.
Why so many GPUs? Because training and running large AI models involve colossal computational loads. Training a single large language model can take weeks and require the parallel operation of hundreds of accelerators simultaneously. «Haein» was created precisely for such tasks.
But raw computing power alone doesn't make a product. SK Telecom built a service around the cluster called GPUaaS (GPU as a Service), allowing companies and developers to rent the required amount of computing resources and pay only for what they actually use – more like a subscription than purchasing their own hardware.
A Name with Meaning
«Haein» is not just a technical designation. The name refers to the Haeinsa Buddhist temple in South Korea, which houses the Tripitaka Koreana – a famous collection of over 80,000 wooden printing blocks with Buddhist texts created back in the 13th century. It is one of the main symbols of Korean cultural heritage.
Choosing such a name for an AI infrastructure was a deliberate gesture: the team wanted to emphasize that «Haein» should become a repository of knowledge for the digital age and the foundation for a sovereign Korean AI – meaning the country's technological independence in the field of artificial intelligence.
What's Inside Besides the Hardware
Building a 1,000-GPU cluster is a non-trivial task. But it's even more challenging to make it possible to flexibly divide this cluster among different users and tasks without losing efficiency.
To achieve this, SK Telecom developed its own virtualization system that allows the cluster to be literally «sliced» for specific needs: one client gets resources for model training, another for running it in production, and a third for experiments. This ensures that the overall hardware utilization remains at its maximum, with nothing sitting idle.
Additionally, the company created a management platform for developers – a unified tool for tracking resource usage, launching tasks, and overseeing the entire workflow with AI models. It's something like a «personal dashboard» for teams working with large-scale computations.
The Government Is Also Involved
«Haein» received government recognition even before the GLOMO award: the cluster was selected by South Korea's Ministry of Science and ICT to participate in an AI computing support program. As part of this initiative, the infrastructure is used to develop proprietary foundational AI models – the basic systems upon which applied solutions are then built.
This is a crucial point: a government contract means we are talking not just about a commercial cloud for corporations, but about an infrastructure of national significance.
Three Years in a Row – Coincidence or a Pattern?
For SK Telecom, this is the third consecutive win in the «Best Cloud Solution» category at GLOMO. It's telling that each time the company won with fundamentally different products: first with a cloud load analysis system, then with GPU virtualization technology, and now with a full-fledged cluster.
This speaks not so much to luck as to a consistent strategy: the company is methodically building a technology stack – from resource management to physical infrastructure – and each layer proves mature enough to contend for industry recognition.
Why This Matters Beyond Korea
The story of «Haein» is a good example of a broader trend. Large AI models require immense computing resources, and not every company or country can afford its own data center with thousands of GPUs. Cloud services like GPUaaS lower the barrier to entry: instead of capital expenditures on hardware, there's a subscription; instead of your own infrastructure, you rent someone else's, but a reliable one.
However, the question of sovereignty remains. Relying on American cloud giants is one thing. Having your own national alternative, certified by the government and tailored to local needs, is quite another. This is precisely the direction SK Telecom is moving in, and «Haein» is the most visible result of this movement so far.
How soon similar models will appear in other countries is an open question. But it seems quite obvious that Korea's experience will be studied closely.
