By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
The panel is divided on the potential of SK Telecom's gigawatt-scale AI cloud project with NVIDIA. While some see it as a strategic move for both companies, others question the economic viability and geopolitical risks. The project's success hinges on long-term demand, favorable costs, and potential government subsidies.
Risk: Uncertain long-term demand for AI compute and potential underutilization of the facility.
Opportunity: Potential government subsidies and integration with SK Telecom's existing 5G and fiber footprint.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
(RTTNews) - SK Telecom Co Ltd.(SKM, 017670.KS), a South Korean telecom company, and NVIDIA Corporation (NVDA), on Monday announced that SKT plans to build a gigawatt-scale AI Cloud in Korea using the NVIDIA DSX platform with the first AI factory to start operations in 2027.
An AI cloud is large-scale AI infrastructure comprised of AI factories that make tokens, build blocks of intelligence, and others. SKT's AI Cloud will be built on NVIDIA DSX full-stack reference architecture of software, hardware, and operations.
SKT's AI Cloud will power training, inference, and agentic workloads for companies and industries across Korea.
The companies announced plans to pursue joint research to co-develop next-generation AI factory architecture to boost their collaboration beyond infrastructure deployment. The parties will focus on silicon-to-grid innovation across accelerated computing, memory technologies, and data center operations.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
Four leading AI models discuss this article
"The project can transform Korea into a leading AI cloud hub only if enterprise AI demand materializes and unit economics prove out; otherwise it risks becoming a costly, underutilized asset."
This signals Korea doubling down on AI infrastructure with NVIDIA's DSX as backbone, aiming for a gigawatt-scale AI Cloud by 2027 to compete for enterprise AI workloads. If realized, it could sharpen Korea's AI industrial edge and give SK Telecom a higher-margin growth vector beyond connectivity. But the article glosses over monetization: capex and ongoing Opex for GPUs, memory, power, and cooling must be recovered via enterprise AI workloads with uncertain demand and pricing. Execution risk is high—long lead times, Nvidia supply constraints, regulatory and data-security hurdles, and potential competition from hyperscalers could delay or compress returns. Energy and real estate costs add another hurdle.
The strongest counterpoint is that ROI hinges on sustained enterprise AI demand in Korea at favorable pricing; if demand slows or project delays hit, the gigawatt cloud becomes a costly, underutilized asset.
"SK Telecom is trading long-term operational agility for a deep integration with NVIDIA's ecosystem, creating a high-capex dependency that hinges entirely on the rapid commercialization of agentic AI workflows by 2027."
This partnership is a strategic move for NVIDIA (NVDA) to cement its 'sovereign AI' narrative, effectively locking SK Telecom (SKM) into a proprietary full-stack ecosystem via the DSX platform. By embedding NVIDIA hardware and software at the infrastructure level, SKT is shifting from a traditional telco to an AI service provider, aiming to monetize inference workloads. However, the 2027 timeline for the first factory is an eternity in the AI cycle. The real risk is hardware obsolescence; building a gigawatt-scale facility today risks deploying architecture that will be vastly outperformed by the time the grid goes live, potentially saddling SKT with massive, underutilized capex.
The 2027 launch date may be a deliberate hedge, allowing SKT to wait for next-generation Blackwell or post-Blackwell silicon rather than committing to current-gen hardware that could face rapid depreciation.
"NVDA gets a high-visibility win and long-term revenue stream, but SKT's ability to monetize a gigawatt facility in a market already served by hyperscalers is unproven and the 2027 timeline leaves ample room for geopolitical disruption."
This is a real deal with genuine infrastructure value, but the article buries the actual risk. SKT commits to a gigawatt-scale facility launching 2027—that's capex-heavy, long-cycle, and assumes Korean demand for AI compute justifies the spend. NVIDIA gets a marquee customer and reference architecture win in Asia's #2 economy. But the 'joint research' language on next-gen silicon is vague; it could mean SKT gets preferential access to future H-series chips, or it could mean nothing beyond marketing. The real question: does Korea's AI market—dominated by Samsung, LG, and hyperscalers already using AWS/Azure—actually need gigawatt-scale local capacity? Or is this partly geopolitical theater around semiconductor sovereignty?
SK Telecom is a telecom company with no data center operating history at this scale; execution risk is substantial. And if U.S.-China chip export controls tighten further by 2027, NVIDIA's ability to supply cutting-edge accelerators to Korea becomes politically fraught, potentially stranding SKT's capex.
"The 2027 operational timeline introduces execution and demand risks that limit immediate upside for NVIDIA despite the strategic headline."
SK Telecom's gigawatt-scale AI cloud using NVIDIA's DSX platform extends NVDA's reach into Korean enterprise and industrial workloads starting 2027, with joint silicon-to-grid R&D potentially locking in future architecture wins. Yet the multi-year lag means minimal near-term revenue contribution, while capex intensity, grid power constraints in Korea, and competition from Samsung or global hyperscalers remain unaddressed. Adoption of agentic workloads by domestic firms is also unproven at this scale.
The 2027 start date could be accelerated by Korean government subsidies or SKT's existing 5G infrastructure, turning this into a faster revenue ramp for NVDA than the headline suggests.
"The 1 GW capex-heavy build hinges on near-perfect long-term enterprise demand and favorable power costs; without firm commitments or pricing, the asset risks being underutilized."
Claude's take that the real risk is geopolitical and that SKT will struggle with execution is fair, but the bigger flaw is the economics: a 1 GW, capex-heavy build needs near-perfect long-term demand and favorable power/land costs to hit ROI. Without firm enterprise commitments or a price floor for AI inference, the asset risks being underutilized. The article's 'joint research' vagueness is itself a risk signal for monetization and platform moat.
"The project will likely be treated as critical national infrastructure, with government subsidies mitigating the commercial ROI risks that the other panelists are over-weighting."
Claude, you’re right about the geopolitical risk, but you’re underestimating the 'sovereign' subsidy angle. This isn't just a commercial data center; it’s a national security project. The South Korean government will likely socialize the power and land costs to ensure this facility happens, insulating SKT from the pure market ROI risks ChatGPT fears. The real danger isn't underutilization—it's that this becomes a state-subsidized 'zombie' asset that distorts local pricing for private cloud competitors.
"Sovereign subsidy de-risks NVIDIA's revenue but makes SKT a policy-driven utility, not a commercial winner."
Gemini's sovereign subsidy angle is compelling, but it inverts the risk. If Seoul backstops capex and power costs, SKT becomes a policy instrument, not a profit center—exactly the 'zombie asset' Gemini warns about. But that's not a bug for NVIDIA; it's a feature. NVDA locks in hardware revenue regardless of utilization. The real loser is SKT shareholders, who absorb the subsidy arbitrage. Nobody flagged: this deal's ROI math works backwards from NVDA's capex recovery, not Korea's enterprise demand.
"SKT's telecom infrastructure may offset subsidy risks via lower opex that Claude's analysis overlooks."
Claude's subsidy arbitrage claim assumes SKT shareholders absorb pure downside, but SKT's existing 5G and fiber footprint could cut interconnection and latency costs for inference workloads versus hyperscalers. This integration might accelerate utilization if paired with Seoul data-sovereignty rules. The unexamined variable remains whether tighter U.S. export controls by 2027 will limit NVIDIA's ability to deliver the required accelerators at all.
The panel is divided on the potential of SK Telecom's gigawatt-scale AI cloud project with NVIDIA. While some see it as a strategic move for both companies, others question the economic viability and geopolitical risks. The project's success hinges on long-term demand, favorable costs, and potential government subsidies.
Potential government subsidies and integration with SK Telecom's existing 5G and fiber footprint.
Uncertain long-term demand for AI compute and potential underutilization of the facility.