Samsung SDS Courts OpenAI and Anthropic as Data Center Tenants
Samsung wants to build, power and run AI data centers, not just sell chips. Analysts see Samsung SDS data center revenue rising sixfold to $4.9 billion by 2031.
Published 5 min read

Samsung SDS is in early talks to sign OpenAI and Anthropic as tenants of its data centers, the Financial Times reported on Sunday, November 15. SDS is the IT arm of South Korea's Samsung group. A deal would push Samsung beyond memory chips into building, powering and running AI data centers for top model makers.
No terms, capacity or timing have been reported, and no agreement has been announced. Samsung SDS, OpenAI and Anthropic have not commented publicly.
What is Samsung SDS offering OpenAI and Anthropic?
A tenant, also called a colocation customer, rents floor space, power and cooling in someone else's building and fills it with its own servers. For AI labs, that is the fastest way to add computing power without building a campus from scratch.
Samsung SDS began as an in-house IT provider and still runs data centers mostly for other Samsung companies. Winning OpenAI or Anthropic would make it a supplier to outside AI customers, which is the shift the group is betting on.
Analysts expect Samsung SDS data center revenue to reach 6.6 trillion won, about $4.9 billion, a year by 2031. That would be roughly six times today's level. Morgan Stanley said in a recent report that SDS is well placed to gain from the construction boom, helped by countries that require data to be stored locally.
Where would an OpenAI or Anthropic deal actually land?
The clearest candidate is a site Samsung SDS started this month. On November 2, it broke ground on a dedicated AI data center in Gumi, South Korea, with 60 megawatts of initial power capacity that can grow to 120 megawatts.
The 46,000 square meter facility is due to be finished by the end of 2028. Samsung SDS said it will run its own GPU servers there and also offer colocation, where customers install their own GPUs, with hybrid air and liquid cooling.
That colocation model is exactly what a tenant agreement would use. The FT report did not say which sites are part of the talks, so Gumi is the obvious fit rather than a confirmed one.
How would Samsung build an AI data center end to end?
The FT described a one-stop offer stitched together from Samsung affiliates. Samsung C&T, the construction arm, would build data centers and power plants, while Samsung SDS would design and operate them.
Samsung SDI would supply batteries and backup power units. Samsung Electronics would provide memory, including high-bandwidth memory (HBM, stacked chips that feed AI accelerators), along with chip packaging and cooling. Shipbuilder Samsung Heavy Industries could build floating data centers at sea.
Much of this already has a paper trail with OpenAI. In an October 2025 letter of intent, Samsung Electronics became a strategic memory partner for OpenAI's Stargate project. Samsung SDS agreed to help design, build and operate Stargate data centers. Samsung C&T and Samsung Heavy Industries signed up to develop floating data centers.
Why does Samsung want to move beyond memory chips?
Memory is hugely profitable in a boom but swings hard in a downturn. Owning more of the data center stack gives Samsung revenue that does not depend only on chip prices.
The group has spent the past year buying pieces. In September 2026, six Samsung companies committed a combined $1 billion to Helix Digital Infrastructure, the KKR-backed AI infrastructure firm led by former AWS chief Adam Selipsky. In November 2025, Samsung Electronics closed its 1.5 billion euro purchase of German cooling maker FläktGroup.
Demand is the other half of the pitch. McKinsey, cited by the FT, expects global data center capacity demand to nearly triple by 2030, with AI taking about 70 percent. OpenAI and Anthropic are among the hungriest buyers, and Anthropic's IPO coverage has put its reported commitments, mostly for compute and chips, at nearly $901 billion.
What could stop the plan?
Coordination is the main doubt. Citi analyst Peter Lee told the FT that Samsung has a wide range of capabilities but has been relatively weak at combining them into one offer for customers.
Samsung shut its central Future Strategy Office in 2017 after a corruption scandal and now coordinates through smaller bodies split by business line. Analysts say that makes projects needing several affiliates at once harder to run.
Some think the parts business is the better bet. Sanjeev Rana, head of Korea research at CLSA, said AI infrastructure spending could last years or even decades. Still, he argued Samsung may earn more from high-margin components than from owning or operating data centers.
KAIST professor Kim Joung-ho sees an opening instead. He told the FT that big US tech companies are strong in AI and software but rely heavily on outside suppliers for hardware, which leaves room for Samsung's integrated package.
What does it mean for everyone else?
For most people, the link is memory. AI data centers are already soaking up DRAM supply, one reason RAM prices have climbed for PCs and phones. A Samsung that both makes the memory and hosts the AI labs would sit on both sides of that demand.
For the AI labs, another landlord in Asia would add capacity outside the crowded US market. Whether talks become signed leases, and at what scale, is the next thing to watch.
Samsung's AI infrastructure pieces at a glance
| Company | Role in the plan | Latest verified step |
|---|---|---|
| Samsung SDS | Designs and runs data centers | Broke ground on 60 MW Gumi AI data center, Nov 2, 2026 |
| Samsung Electronics | Memory, HBM, packaging, cooling | Closed FläktGroup deal, Nov 2025; $500M into Helix, Sep 2026 |
| Samsung C&T | Builds data centers and power plants | Floating data center work with OpenAI, Oct 2025 letter of intent |
| Samsung SDI | Batteries and backup power | Part of the $1 billion Helix commitment, Sep 2026 |
| Samsung Heavy Industries | Floating data centers | Floating data center work with OpenAI, Oct 2025 letter of intent |