Samsung SDS in Early Talks to Rent Data Centre Space to Anthropic and OpenAI, FT Says

Analysts expect SDS's data centre revenue to rise to 6.6 trillion won, about $4.9 billion, by 2031, close to six times the current level.

The strategy depends on combining affiliates that have mostly worked for Samsung itself. Under the model, the group's construction arm would build data centres, SDS would design and run them, its battery company would supply energy storage and its semiconductor unit would provide advanced memory chips and other components. Kim Joung-ho, a professor of electrical engineering at KAIST, said the large US technology companies are strong in AI and software but rely heavily on outside suppliers for hardware, which gives Samsung room to offer an integrated package.

The group has moved in that direction over the past year. In September it invested $1 billion in Helix Digital Infrastructure, a data centre operator backed by KKR and Nvidia. It has also bought a European maker of cooling systems.

Samsung's work with OpenAI goes back further. Last year the group agreed to supply advanced memory chips to the US Stargate data centre programme, and SDS took on design, development and operating work on those data centres. Samsung C&T and Samsung Heavy Industries have said they will work with OpenAI on floating data centres.

Morgan Stanley considers SDS well placed to benefit from a surge in data centre construction, helped by governments tightening requirements for data to be stored locally. McKinsey estimates that global demand for data centre capacity could nearly triple by 2030, with AI workloads accounting for about 70 per cent of it, and that AI compute alone could need up to $5.2 trillion in capital spending.

The spending has its sceptics. Recent results from technology companies have focused investors on the cost of servers, chips, power and cooling. Election wins by Democratic candidates who pledged to stop data centre construction in their states have raised the prospect of tighter US rules. Sanjeev Rana, head of Korea research at CLSA, said AI infrastructure spending is a structural trend that will continue for years, and possibly decades, despite short-term swings.

Samsung's harder problem may be internal. Until 2017 the group had a Future Strategy Office that coordinated decisions across affiliates. It was closed after a political corruption scandal involving group leaders, and Samsung has since relied on smaller coordinating bodies organised by business line. Analysts say that structure makes large projects harder to execute, because AI infrastructure needs affiliates to pool technical, commercial and financing capabilities. Peter Lee, an analyst at Citi, said Samsung has a wide range of capabilities but has been relatively less successful at bringing them together into a single offer for customers.

Some analysts think the better bet is supplying the parts. Samsung already sells high-bandwidth memory, advanced packaging substrates and backup battery units for servers. Rana said building on those higher-margin component businesses could pay off more for the group, and that owning or operating data centres is not necessarily where Samsung can make the best use of its strengths.