Decades in the Making: AI Clusters Demand Specific Knowledge, Still Residing in U.S. Firms

SK Telecom (SKT) has launched Haein, a sovereign Artificial Intelligence (AI) data center providing GPU-as-a-Service (GPUaaS), a cluster featuring over 1,000 NVIDIA B200 Graphics Processing Units (GPUs). The Haein cluster represents a significant step forward from the previous H100-based service and is poised to be a cornerstone of the Sout Korea’s AI development to boost the country’s domestic and global AI competitiveness.

It is built on a US$200 million strategic investment from SKT in Penguin Solutions, the U.S.-based AI and High-Performance Computing (HPC) and AI server vendor. The collaboration also includes SK hynix, a semiconductor and memory vendor part of the SK Group conglomerate. The agreement, initially announced at CES 2025, aims to build on the South Korean deployment and accelerate the development and deployment of comprehensive AI data center solutions in other geographies. SKT will contribute its AI infrastructure management software, including the Petasus AI Cloud virtualization solution and the AI Cloud Manager Artificial Intelligence Operations (AIOps) platform. SK hynix will collaborate with Penguin's SMART Modular Technologies arm to develop memory solutions tailored for the demanding environments of accelerated AI computing.

Similar collaborations include:

• Dell’s collaboration with Atos in Europe to develop AI solutions that align with the European Union’s (EU) data sovereignty regulations, and in Australia, Dell partnered with Macquarie Data Centres.
• Hewlett Packard Enterprise (HPE) has partnered with Khazna Data Centers to launch a managed data center hosting service featuring direct liquid cooling for AI, supporting the United Arab Emirates’ (UAE) national AI strategy. In Asia, HPE is collaborating with Japanese telco provider KDDI to build AI infrastructure at a data center in Osaka.
• Supermicro is collaborating with Datasection, Foxconn, KDDI, and Sharp in Japan, to deploy a large-scale AI data center. It also signed a US$20 billion agreement with DataVolt, a Saudi Arabian data center builder.

For Penguin Solutions, the partnership provides a substantial injection of capital and a strategic gateway into the competitive Asian market, backed by a powerful local conglomerate. For SK Group, it accelerates its ambition to become a global AI leader by combining its semiconductor and infrastructure strengths with Penguin's proven system integration and ongoing management and high-performance Service-Level Agreement (SLA) guarantees. The launch of the Haein cluster, built on NVIDIA's cutting-edge Blackwell GPUs, establishes a formidable sovereign AI capability for South Korea, a strategic national asset for security and economic competitiveness.

The collaboration encapsulates several market-shaping trends in the global AI infrastructure race. While highly impactful for the companies involved, its true significance is magnified when viewed alongside a wave of similar deals being struck by other U.S. technology leaders like NVIDIA, Dell Technologies, HPE, and Supermicro. It is also part of a wider proliferation of sovereign AI initiatives as advanced economies in all regions of the world are aggressively investing in domestic AI data centers driven by the need to ensure data privacy, maintain national security, and develop AI models that accurately reflect local languages and cultures.

While the goal is sovereignty, the reality is a deep reliance on U.S. technology. Deals like Dell's partnership with Macquarie Data Centres in Australia, HPE's collaboration with Khazna Data Centers in the UAE, and Supermicro's US$20 billion agreement with Saudi Arabia's DataVolt all highlight that American firms are the primary architects and enablers of such national AI ambitions. The "AI factory" model becomes the standard as the market is rapidly shifting away from procuring individual hardware components and toward demanding fully integrated, often highly customized “AI factories.” This is represented by the Penguin-SK alliance, which combines hardware, networking, specialized memory from SK hynix, and sophisticated management software (Penguin's Scyld ClusterWare and SKT's Petasus AI Cloud), and is enabled by Penguin’s AI and HPC infrastructure optimization service. This trend is mirrored across the industry by other U.S. AI server vendors, albeit to varying degrees of “hands-on” involvement.

The developments announced by Penguin Solutions and SK Group offer insights and opportunities for other specialist AI server vendors and integrators.

• Full-Stack Solution Model: The collaboration highlights a market shift toward comprehensive, full-stack AI solutions. Customers are looking for partners that can manage the complexity of designing, building, deploying, optimizing, and managing entire AI clusters. Vendors should accelerate efforts to offer broader professional services by offering integrated solutions that include cluster management software, managed services, and expertise in optimizing for specific AI workloads. Penguin Solutions’ AIM service is a good example as it includes customization and optimization, leveraging experience and data from billions of hours of GPU runtime.
• Forge Strategic Ecosystem Partnerships: The trilateral agreement between a system integrator, a telco/cloud service provider, and a component manufacturer is a powerful model. Specialist vendors should actively seek partnerships with companies that have complementary capabilities. This could include collaborations with data center operators, cloud service providers, and key component suppliers to create more holistic and competitive offerings that also address the importance of storage and memory (e.g., CXL) in large, complex AI clusters. Examples include Penguin Solutions’ memory solutions business unit, as well as Innodisk, Micron, and Samsung.
• Target the Sovereign AI Wave: The global trend toward sovereign AI presents a significant, lasting market opportunity. Nations are increasingly investing in domestic AI infrastructure to ensure data privacy, security, and economic, as well as technological independence. Vendors and integrators should tailor their marketing and solutions to address the specific needs of government-led and national AI initiatives, emphasizing security, control, and support for (as well as collaborations with) local ecosystems. This includes software platforms like SKT's Petasus AI Cloud, and could include players in other jurisdictions such as Dataiku, SAP, and Scaleway.
• Develop Deep Expertise in Leading-Edge Technologies: The fast adoption of NVIDIA's Blackwell GPUs by SKT underscores the importance of staying in the lead from a technological perspective. Vendors must invest in early access programs and develop deep technical expertise in the latest GPU architectures, networking fabrics, and cooling technologies. This allows them to act as trusted advisors for clients looking to build competitive, state-of-the-art AI infrastructure that will remain “current” for a number of years. This should include AMD’s offering, such as MI355X and the upcoming MI400X platforms, which are gaining traction with neoclouds and hyperscale clouds.