SK Square's recent investment in the U.S. data-orchestration company Hammerspace signals a significant expansion of its AI investment thesis.
This move extends the group's focus beyond just semiconductor chips to tackle what's known as the 'AI infrastructure bottleneck' at the data layer. Simply put, even the most powerful AI systems can't work at full capacity if they can't get the data they need fast enough. This is a huge and growing problem in the industry, where expensive GPUs often run at less than 50% of their potential because they are stuck waiting for data from storage systems. Hammerspace develops technology that directly addresses this by creating a unified, high-speed data environment, ensuring that GPUs are constantly 'fed' with data to maximize their performance.
The logic behind this investment becomes clear when we look at the broader market trends. First, major cloud service providers, or 'hyperscalers' like Meta and Microsoft, are investing hundreds of billions of dollars in AI infrastructure. They are buying up GPUs and high-bandwidth memory (HBM) as fast as they can be produced. This massive spending spree is creating enormous demand for every component in the AI ecosystem.
Second, this rapid expansion has exposed a critical weakness: the data pipeline. SK hynix, an SK affiliate, has reported that its advanced HBM is sold out through 2026. This highlights the scarcity and high value of these components. It makes no sense to have these precious, high-performance resources sitting idle. Therefore, improving GPU utilization—the percentage of time a GPU is actively processing—becomes incredibly valuable. A small improvement in utilization can translate into massive ROI on a multi-billion dollar AI cluster.
This is precisely where SK Square's investment fits in. By backing Hammerspace through its overseas investment arm, TGC Square, the company is placing a strategic bet on a 'picks-and-shovels' solution for the AI gold rush. Instead of just making the chips, they are investing in the critical technology needed to make those chips work efficiently. This complements SK hynix's position in the memory market and positions the entire SK group to capture more value from the ongoing AI revolution.
- Glossary
- Data Orchestration: The automated management and coordination of data across different storage systems and locations to ensure it is available where and when it's needed for applications, like AI models.
- GPU Utilization: A metric that measures how much of a Graphics Processing Unit's (GPU) potential processing power is being used over a period of time. Low utilization in AI often points to data bottlenecks.
- Hyperscaler: A large-scale cloud service provider (e.g., Amazon Web Services, Microsoft Azure, Google Cloud) that operates massive data centers and provides computing resources to a global customer base.