SK Hynix Is Spinning Off Its AI Memory Bet for a $150B US Listing
Solidium, a grandchild company of SK Hynix, is preparing for a US IPO that could value the firm at up to $150 billion. The move reflects a growing strategy among Asian chipmakers to isolate AI-focused subsidiaries and list them on Wall Street.
SK Hynix Just Turned a Grandchild Company Into a $150 Billion bet on AI Memory
Solidium, an obscure corporate name that sits three generations removed from South Korea’s largest chipmaker, is preparing for a Wall Street debut that would make it one of the most valuable semiconductor companies listed outside of Taiwan.
Reuters reported this week that Solidium is holding “bake-off” meetings with investment banks, the industry term for the final round of pitches before an IPO book runs. The target: a valuation of up to $150 billion, raising roughly $15 billion in fresh capital.
If those numbers hold, Solidium would dwarf ARM Holdings’ 2023 listing, which valued the British chip-design firm at $54 billion, and even outpace Cerebras, the “Nvidia rival” that priced its IPO at $56.4 billion earlier this year.
The implications extend well beyond a single listing. They mark a structural shift in how Asian semiconductor companies are organizing themselves for the AI era—and how aggressively they’re willing to split off units and float them on US exchanges to capture that premium.
How Solidium Got Here
Solidium was born from Intel’s retreat. In 2021, SK Hynix acquired Intel’s NAND flash and solid-state drive business and established the company as a US-based subsidiary. For years it operated quietly inside the SK Hynix ecosystem, supplying enterprise SSDs for servers, cloud infrastructure, and data centers.
This year, the corporate tree grew another branch. SK Hynix created AI Company, a US-domiciled investment subsidiary, which became Solidium’s parent. The reorganization wasn’t ceremonial. It positioned Solidium to be evaluated and listed independently, rather than buried inside SK Hynix’s sprawling conglomerate structure.
That’s the playbook now. Separate the AI-relevant unit, list it on Wall Street, let a different investor base price the future without dragging the valuation of the legacy DRAM business along for the ride.
Why Wall Street, Why Now
The timing is no accident. NAND flash demand has surged alongside AI training and inference workloads. SK Hynix is reportedly considering building a NAND flash production facility in New York state—a move that would place it geographically closer to the data centers and cloud customers driving demand.
A US IPO solves a financing problem that Korean markets cannot. The KOSPI trades at valuations that discount semiconductor cyclicality. Wall Street, starved for pure-play AI infrastructure exposure, does not. Solidium’s projected $150 billion figure, whether or not it survives the final pricing, reveals where the market thinks the money is.
ARM’s listing proved the model works for chip-adjacent companies. Solidium’s would prove it works for the physical production side—the factories, the NAND wafers, the enterprise storage that AI data centers consume by the petabyte.
The SK Hynix Strategy in Focus
SK Hynix is already the dominant supplier of high-bandwidth memory to Nvidia’s GPUs. HBM, the stacked DRAM that powers AI training clusters, is where SK Hynix has been investing billions and beating Samsung Electronics to market. That relationship with Nvidia is the crown jewel.
But HBM is only one layer of the memory stack. Enterprise SSDs built around NAND flash represent a second, faster-growing front in the AI infrastructure buildout. SK Hynix’s decision to spin Solidium out through AI Company signals it wants to capture both trades separately—not bundle them into a single Korean equity story that foreign investors find opaque.
The move also insulates SK Hynix. If Solidium’s NAND business faces margin pressure or a cyclical downturn, it fails as an independent company rather than dragging down the parent’s consolidated results. If it succeeds, SK Hynix retains ownership while unlocking value through a public market that prices AI infrastructure differently than Seoul does.
Who Wins, Who Loses
SK Hynix wins if Solidium prices close to the $150 billion target. The company raises $15 billion in new capital, which can fund the New York NAND facility and potentially additional capacity expansions. SK Hynix’s ownership stake in a publicly traded AI-memory company becomes a liquid asset on its balance sheet.
Solidium’s employees and early investors win if the IPO is sized below the $150 billion ceiling and leaves room for post-listing appreciation. A bake-off this late in the process suggests the bankers are still calibrating demand. The final number could shrink.
Samsung Electronics loses visibility into a potential competitive threat. If Solidium thrives as a publicly traded NAND competitor with Wall Street capital behind it, Samsung’s integrated model looks less like an advantage and more like a constraint. Samsung has not announced a comparable spinoff.
Nvidia’s supply chain benefits in the short term but faces a new dynamic. SK Hynix now has a separate funding vehicle and a public-market incentive to expand NAND capacity aggressively. That strengthens the company’s negotiating position on HBM and enterprise storage alike.
What Happens Next
The bake-off is this week. Pricing decisions follow. If Solidium files formally, expect a roadshow targeting US institutional investors with an AI infrastructure narrative—a story that commands multiples Samsung and SK Hynix’s legacy businesses cannot easily tell from Korea.
The New York NAND facility decision will likely land before or alongside the IPO timeline. The two moves reinforce each other: a US listing justifies US production, and US production justifies a US listing.
Watch whether Samsung responds. A Korean peer announcing a comparable spinoff would validate Solidium’s strategy and signal that the AI-memory breakout is becoming industry-wide, not SK Hynix-specific.
The $150 billion figure is a peak estimate. The bake-off will reveal whether banks believe the market will pay full price for AI-grade NAND stories—or whether a cyclical memory business still drags on a growth valuation. Either way, the structure itself—the grandchild company, the US parent, the Wall Street listing—is the real signal. Asian semiconductors are no longer asking to be taken seriously by global investors. They are building the vehicles to force the issue.