South Korean tech giant SK Group doubled down on AI with another restructure, this time with a U.S.-based and AI-driven unit formed from its booming chip division.

SK Hynix revealed this week it would be spinning out a new venture by the tentative name of AI Company (AI Co.), reflecting what Hynix saw as its changing nature from pure-play memory provider to chip design manufacturer serving the whole AI ecosystem.

Citing Gartner projections on AI infrastructure driving global AI spend to $4.7 trillion by 2029, Hynix positioned the unit as a partner in the AI data center ecosystem, responsible for overall performance and efficiency.

Officially launching in February, the $10 billion investment will center around its U.S.-based Solidigm SSD arm. Originally Intel’s NAND and SSD business, the firm was acquired by SK Hynix for $9 billion in 2020 to be merged with its own NAND operations.

Hynix cited Solidigm as “already a key player in the AI datacenter ecosystem,” an assertion disputed by Alexander Harrowell, principal analyst for advanced computing at Omdia, who sees the restructure as a response to U.S. export restrictions keeping higher-end chips out of China.

“SK Hynix is being put under pressure to move some work into the U.S., where Micron and Samsung already have fabs. SK Hynix has Solidigm and a substantial R&D site in Santa Clara, but nothing on that scale,” Harrowell told SDxCentral.

Max Smolaks, research analyst at Uptime Institute Intelligence, meanwhile, saw the move as an attempt to replicate a certain U.S.-based chip giant.

"Investor enthusiasm for all things AI now extends to memory chipmakers, and shares in SK Hynix have more than tripled in the past 12 months. The company is leveraging this attention to establish itself as an active participant in the AI ecosystem, and not just a component supplier. This is what Nvidia managed to do, and it might just work," Smolaks said.

The SK Group subsidiary’s move also echoes its repivot of SK Telecom (SKT), which last year consolidated its AI operations into an internal independent business unit called AI Company-in-Company (AI CIC). The move saw SKT separate management and focus between traditional mobile operations and its burgeoning AI activities for the South Korean market.

The plans for AI Co. differ in the company being a brand new subsidiary, with a U.S. base at that. But in Harrowell’s view, Hynix's pitch appears more a vehicle for portfolio investments than an operating company for a memory fab, with the analyst highlighting how the investment is structured as a capital call reserve.

“The $10 billion commitment is small for a volume semiconductor product and is structured in the way VC funds are – you don’t fork out the cash when you sign, instead you sign a commitment to come up with it when a deal is struck," Harrowell explained. “I think they would rather not move manufacturing if they can help it, but would be delighted to roll some of their profits from the AI chip boom – until a couple of months back, they were the exclusive supplier of high bandwidth memory (HBM) in Nvidia’s data center GPUs – into startups or acquisitions in the U.S.”

According to reports this week, rival Samsung Electronics is in the final qualification phase to supply HBM4 to Nvidia, finally putting it on par with Hynix.

Chaebols ch-ching

It is unlikely Hynix is too concerned regarding that development, as on the same day AI Co. was unveiled, the company posted full-year operating profits of $33.3 billion, with quarterly operating profits growing 137.2 % to $13.4 billion for the final three months of 2025.

Rival Samsung posted similar record-breaking profits this week, with the company’s memory division disclosing operating profits of around $17.4 billion for its fiscal 2025.

Both chabeols warned the ongoing memory chip shortage is likely to continue into next year, with Park Joon Deok, head of SK Hynix's DRAM division, noting "supply constraints and strong demand for server-related products" impacting orders on the consumer side.

The warning added to reports from last year that the company expects growth for commodity DRAM to be constrained through 2028, excluding HBM and SOCAMM (small outline compression attached memory module), which it manufactures with its recently opened M15X fab in Cheongju, South Korea.

In Samsung's earnings call this week, Kim Jae-june, head of Samsung Electronics’ memory business, told analysts that supply is expected to remain tight due to limited cleanroom expansion across the industry, while the supply shortage "is anticipated to persist due to strong demand linked to AI.”

To mitigate this, Kim noted that Samsung plans to invest in new fabrication facilities to expand cleanroom capacity ahead of demand and will continue "a proactive investment strategy going forward.”