Inside SKT’s push to turn SK Hynix’s fabs into an AI business

SKT's Hy-TF is not a digital-transformation project—it is a supply-chain repositioning play that converts wafer fabs from cost centers into AI data assets. The technical kill-chain: HBM-line yield curves, defect maps, and process windows are the scarcest training corpus for manufacturing AI. By piping this through telecom-grade infrastructure into a B2B service, SKT carves into ASML's and Siemens' industrial-AI TAM. Upstream EDA vendors face a data-flywheel squeeze—the party whose models run on live fab data will cannibalize their simulation and consulting revenue. Compliance risk is materially underpriced. A telecom operator touching core fab process parameters, then monetizing via B2B, triggers cross-border data flows, IP-attribution disputes, and customer-isolation requirements (especially NVIDIA HBM custom specs) under both Korean industrial law and the EU AI Act. This is a supply-chain security event, not an IT ticket. Competitive dynamics: TSMC's fab-AI stack is closed. SKT's differentiator is an open platform. If Hynix's process AI ships as SaaS to GlobalFoundries or Intel, foundry pricing power gets redistributed, and Siemens/ABB industrial-AI divisions take the first hit. 12–24-month read: Hy-TF likely lands its first commercial deployment by Q2 2026, targeting automotive and power-device fabs—lower process complexity, manageable data sensitivity, ideal beachhead. Endgame: SKT becomes the "AWS of semiconductor AI," with Hynix fabs as the training cluster.