AI infrastructure has hit a new hard limit and it’s not compute, it’s data. As organizations scale training and inference, the bottleneck has increasingly become the ability to find, govern, and deliver the right data to the right GPUs fast enough – across sites, clouds, and jurisdictions.
That’s why TGC Square, the overseas investment arm referenced in SK Square’s announcement this week, invested in Hammerspace: to back a platform purpose-built to eliminate data fragmentation and data-path friction, all while making sovereignty enforceable in the real world.
In the AI era, performance isn’t limited by how many GPUs you can buy; it’s limited by whether data can reach those GPUs fast enough. That’s why SK Square invested in Hammerspace: to back a data orchestrator that can logically unify distributed data and then move the right data to available GPUs without interrupting access. In a world where datasets span sites, clouds, and jurisdictions, orchestration is how you turn fragmented storage into an AI-ready data plane – globally and in sovereign environments.
AI Needs Distributed Data
AI pipelines don’t stay neatly inside one storage system, data center, or geographic location. Data is created in one place, enriched in another, and consumed wherever GPU capacity exists. The common “fix” is to duplicate datasets into new AI silos per region or per cluster.
That approach creates a familiar failure mode:
- More copies → more drift
- More silos → more policy gaps
- More manual governance → more operational risk
- More storage sprawl → more cost and slower AI cycles
Global namespace + orchestration is what makes Sovereign AI real: one consistent view of data everywhere, with policy-driven control over where each file can live, move, and be computed on, so data stays where it must, access is provable, and AI runs at full speed.
The Basis for the Investment
Hammerspace addresses the modern AI constraint with data orchestration within a global namespace that turns fragmented data sets into a unified data estate – across distributed environments while staying within sovereign boundaries. Our unique data platform can:
- Orchestrate data in place by indexing and leveraging file metadata, so teams can use distributed datasets without disruptive migrations or creating new storage silos.
- Orchestrate access through a global namespace so users and applications see one consistent view of data across on-premesis, multi-site, and cloud environments.
- Orchestrate policy-driven outcomes so data movement, placement, performance, durability, and compliance behaviors are automatically enforced, and continuously re-evaluated as infrastructure and requirements change.
AI Data Access: Controlled Participation, Not Copied Isolation
AI only delivers value when the right data can be found, accessed, and delivered to GPUs quickly. This is more complex, requires more humans, and is much slower when that data is spread across sites, clouds, and storage systems. The instinctive response is to copy everything into a dedicated “AI zone” or per-cluster silo. That creates delays, duplicate datasets, and governance drift.
Hammerspace takes a different approach: one global namespace for access, paired with policy-driven orchestration that determines what data can be used where, by whom, and under what conditions—at file granularity. Teams get the speed and simplicity of local access, without forcing new silos or breaking the guardrails that matter in regulated and sovereign environments.
Proven in Demanding Environments
Hammerspace has been adopted in high-scale, high-performance environments, including top-tier customers referenced in the announcement such as Meta and Los Alamos National Laboratory, where data bottlenecks pose an existential threat to productivity and compute ROI.
And it’s driven by deep systems expertise: David Flynn previously founded Fusion-io, which was acquired by SanDisk — experience that shows up in a platform built to remove I/O friction instead of adding new layers of overhead.
The Flynn Factor
Hammerspace was founded and is led by Flynn, a first mover who sees what infrastructure must become next, and builds it before the market even has the language for it. Flynn invented the PCIe flash model at Fusion-io (the NVMe precursor) and sold it to SanDisk. Hammerspace is the next act: a global namespace data orchestrator engineered to remove I/O friction and feed GPUs at full speed—without creating new silos or breaking sovereignty.
Additional Resources
- Press Release: Corporate Value of the Global AI and Semiconductor Companies Invested by SK Square Increased Up to Sevenfold in Two Years
- Data Unchained Podcast: The Infrastructure Revolution for AI Factories w/ David Flynn
- Analyst Research Brief (HyperFRAME Research): Hammerspace: Unifying Data for Enterprise AI Success
- Hammerspace Data Orchestration Services
- Hammerspace Global Namespace
