
Masayoshi Son did not need a signed deal to light up the wires. He only needed to ask Wall Street a wild question.
According to The Information, SoftBank Group (SFTBY) CEO Son recently approached major U.S. institutions about financing a possible buyout of SK Hynix (000660.KS) at roughly $1 trillion. Reuters and Channel News Asia carried the same core block on Friday into Saturday. SoftBank's own answer landed almost as fast as the leak. A spokesperson said the company explores many opportunities, and this one is no longer under active consideration.
SK Hynix said it has no knowledge of the matter. It also reminded everyone that any foreign purchase would need prior Korean government approval.
My take is simple. This was never a real near-term merger story. It was a feasibility check on the one piece of the AI stack Son still does not control: high-bandwidth memory. The ask itself is the tell. The kill statement is the fact.
What the $1 trillion number actually means
SK Hynix already sits in the high hundreds of billions of dollars of market value on the Seoul exchange. Korean coverage put the company around 777 trillion won, or roughly $580 billion, when the story hit. A trillion-dollar package would not be a polite premium. It would be a control moonshot on the company that still leads the world in HBM, the stacked memory that feeds advanced AI accelerators.
SoftBank is not sized like a buyer that can casually write that check. Tokyo shares of SoftBank Group slid sharply into the Nov. 20 session while the broader AI complex sold off. SoftBank already carries a massive OpenAI position, about $64.6 billion invested for roughly a 13% stake after the October tranche, plus its long-running Arm story and Vision Fund baggage. Debt, margin loans, and collateral on those holdings are part of how SoftBank funds the dream. Asking banks whether a trillion-dollar Hynix package is even financeable fits Son's style. Pretending SoftBank can close it cleanly against a Korean industrial champion does not.
If that package had ever become a live offer, it would have blown past any completed tech deal you can name offhand. It did not. SoftBank said so. Treat the headline as ambition, not arbitrage.
Why Hynix sits in Son's crosshairs
Son has spent years trying to own more of the AI supply chain, not just apps. Arm (ARM) gives him a design foothold. OpenAI gives him a model lab. Data-center and power bets fill out the physical layer. What he still does not own is the memory choke every GPU stack needs.
SK Hynix is that choke for a reason. It still leads HBM by revenue, even after Samsung clawed back share. Recent industry tallies put SK Hynix around half of HBM revenue in the second quarter of 2026, with Samsung and Micron (MU) filling most of the rest. Nvidia (NVDA) remains a defining customer. HBM4 is ramping into the next wave of accelerators. That is the business Son was sniffing around: scarce AI memory with pricing power, not a random Korean ticker.
The hard part is whether Japan-based SoftBank can buy the crown jewel of Korea's memory industry the way a private-equity shop buys a software company. It cannot without Seoul's blessing and without SK Group's willingness to sell a strategic asset. Neither showed up in this report. Hynix's "no knowledge" line and the government-approval reminder were the polite way of saying the door is not open.
What investors should actually watch
Do not rebuild a SoftBank-Hynix merger model because one outlet said Son asked about financing. SoftBank already walked it back. The useful work sits on two ledgers.
First, SK Hynix and the HBM complex. The franchise does not need a SoftBank buyout to stay important. Watch HBM pricing, HBM4 yields, how much supply still runs through Nvidia versus Google (GOOGL), AMD (AMD), and cloud custom chips, and whether Samsung and Micron keep closing the gap. A dead SoftBank idea does not free a single wafer. It only shows how central memory still looks to the people writing the biggest AI checks.
Second, SoftBank as a leveraged AI portfolio, not as a Korean industrial acquirer. The stock already moves with Arm marks, OpenAI paper value, and how expensive it is to fund the next tranche. That is the real SoftBank risk and reward. Son still wants vertical control. Desire and executable capital are different things.
Micron and Samsung live in the same HBM fight. Arm remains SoftBank's public chip-design crown jewel. None of that turns a shelved financing talk into a deal calendar.
What would change my mind
I would revisit this if SoftBank reversed the kill line in public, if SK Group or Korean regulators opened a foreign-control path, or if banks leaked real underwriting work instead of "Son asked a question." Until then, treat every recycled "$1 trillion Hynix" headline as old theater.
Bottom Line
Son briefly shopped a trillion-dollar path to own SK Hynix because AI memory is the vertical SoftBank still lacks. SoftBank says the idea is dead. Korea would have blocked a real foreign control fight anyway. Price Hynix on HBM power and industrial policy. Price SoftBank on Arm, OpenAI, and balance-sheet stress. Do not price a buyout that never cleared the first phone call.

Comments
Log in or sign up to join the conversation.