SK Telecom rolls out Nvidia Blackwell GPUs for sovereign AI cloud
SK Telecom (SKT) launched “Haein,” a sovereign AI infrastructure platform that integrates 1,000+ Nvidia Blackwell GPUs for AI training and inference. The system went live on Aug. 5, 2025 and provides GPU-as-a-Service, selected to support South Korea’s national AI foundation model program.
Haein runs on an “AI Operating System” from VAST Data, with Supermicro contributing to hardware optimization. The VAST Data partnership (announced Aug. 14, 2025) targets better virtualization and data-pipeline performance for multi-tenant workloads.
SKT is scaling beyond Haein’s initial 1,000 GPUs. It is part of SK Group’s “AI factory” plan to deploy 50,000+ Nvidia GPUs, building on a prior Nvidia H100 cluster. As of June 2026, SKT and Nvidia discussed scaling toward a “gigawatt-scale AI Cloud” using Nvidia DSX, with the first AI factory expected in 2027.
Investors note the GPUaaS model’s overlap with the decentralized compute narrative, where projects like Render, Akash, and io.net have pursued GPU marketplaces. While SKT is building a centralized offering, the government-backed “sovereign AI” angle could strengthen the demand outlook for AI compute infrastructure, including Nvidia Blackwell GPUs, across South Korea.
Bullish
This is likely bullish for the broader “AI compute” trade, but it’s not an immediate, direct catalyst for any single major token. SK Telecom’s rollout of Nvidia Blackwell GPUs expands real-world demand for AI training/inference compute and strengthens the GPU-as-a-Service playbook—an area where crypto-market narratives (decentralized/marketplace compute) have often sought to capitalize.
Short-term: traders may position in AI-compute proxies (GPU/compute-related tokens) on newsflow tied to Nvidia-class hardware and scaling roadmaps (1,000+ GPUs today, “gigawatt-scale” ambitions by 2027). However, because SKT is building a centralized, government-backed sovereign cloud (not a decentralized GPU marketplace), the “direct revenue/market share” link to decentralized token models is weaker.
Long-term: the government-supported “sovereign AI” requirement can create persistent procurement and capacity buildout in-country, which typically lifts sentiment toward AI infrastructure themes. Historically, when large operators announce hyperscale-like capacity expansions (e.g., major data-center and AI cluster deployments), market participants often respond by rotating toward liquid AI infrastructure/compute exposures—first on sentiment, then on follow-through if usage and commercialization become visible.
Overall, expect a mild-to-moderate positive tone for AI/compute-related assets, while broader market stability depends on whether this translates into measurable adoption and partnerships rather than just announcements.