Samsung Bets on AI Infrastructure, Seeking Transformation from Chipmaker to Full-Stack Data Center Supplier
nashnova research
Samsung Group is bundling its construction, battery, IT-services and shipbuilding subsidiaries into a single AI data-center supply chain, aiming to evolve from a parts vendor into an end-to-end builder and operator of data centers. This means → Samsung is not betting on any single chip — it is betting that the AI-compute boom can unlock an entirely new business line.
What exactly is Samsung trying to do?
The core pivot: move from "selling memory chips and components" to becoming a full-stack data-center provider — design, build, cool, power, and operate, all in-house.
In plain terms = Samsung used to sell bricks and cement. Now it wants to be the general contractor who delivers the whole building, from blueprint to property management.
The central wager: whether the explosion in AI-compute demand can turn subsidiaries that previously served only Samsung itself into competitive external suppliers.
What has Samsung done in the past year?
In September 2024, Samsung invested $1 billion in Helix Digital Infrastructure, a data-center operator backed by KKR and Nvidia.
It acquired European cooling-systems manufacturer FläktGroup, filling the thermal-management gap in its data-center stack.
It signed a partnership with OpenAI covering advanced memory-chip supply plus data-center design, construction and operations. Samsung Heavy Industries also said it would participate in floating data-center development.
How do the four subsidiaries divide the work?
Samsung Electronics: memory chips, foundry services, advanced packaging, and data-center cooling via FläktGroup.
Samsung SDS (IT services): data-center design, construction and operations, plus cloud services and GPU-as-a-service — selling GPU compute to clients the way a utility sells electricity.
Samsung C&T (construction): physical building of data centers and power plants. Samsung SDI (batteries): backup power systems.
This means → four subsidiaries each own one segment, assembling a complete chain from chip to power supply — but only if they can coordinate.
How big is Samsung SDS's external market?
Samsung SDS started as an internal IT department; its data centers mostly serve Samsung itself. The Financial Times reports the company is in early talks with Anthropic and OpenAI about becoming tenants.
Morgan Stanley projects Samsung SDS's data-center revenue will reach ₩6.6 trillion (~$4.9 billion) per year by 2031 — roughly six times the current level.
This means → growth comes from two forces: AI companies need to lease capacity, and governments worldwide are legislating data-localization requirements, creating a wave of local build-out demand.
How large is the market, and where does the money come from?
McKinsey estimates global data-center capacity demand could nearly triple by 2030, with AI workloads accounting for about 70% of total demand.
Meeting AI-compute needs alone could require $5.2 trillion in capital expenditure.
Joungho Kim, a professor at KAIST, notes: "U.S. Big Tech is strong in AI and software, but highly dependent on outside suppliers for hardware." This reflects the exact gap Samsung aims to fill.
What is the biggest risk?
In 2017, Samsung's cross-business coordination body — the "Future Strategy Office" — was disbanded after its leaders were implicated in a political corruption scandal. Since then, the group has operated without a unified command center, relying on smaller, business-line-specific teams.
In plain terms = all four puzzle pieces exist, but no one is holding the master blueprint to fit them together — that is the single biggest organizational obstacle to the full-stack strategy.
External uncertainty is real too: some U.S. states face political pressure to halt data-center construction, and investor confidence in AI-infrastructure spending has wobbled. Sanjeev Rana, head of Korea research at CLSA, says the market will see "ups and downs," but the overall spending trend is "structural."
市场有风险，内容仅供研究参考，不构成投资建议。