By | Beyond the Headlines, Author | Hua Hua
In the past year, the most profitable company in the global AI industrial chain wasn't OpenAI, not Google, not even Nvidia.
It was a South Korean company.
Its name is SK Hynix.
In the first quarter of this year, Hynix's net profit margin was 77%, even higher than Nvidia's. It made a net profit of about 40 trillion won in a quarter, roughly $300 billion.
And the scale of its recent U.S. IPO fundraising was $26.5 billion.
That is to say, it earned more money in one quarter than it raised in the IPO.
The question is, why would a company not short of money at all still run to the U.S. to list?
I. Not for Financing, but for Repricing
On July 10, SK Hynix officially listed on Nasdaq, with an issue price of $149. Its stock closed up nearly 13% on its debut, giving it a market cap of $1.23 trillion.
This IPO raised $26.5 billion, surpassing the $25 billion raised by Alibaba's IPO in 2014, setting a new record for foreign companies listing in the U.S. It is the second-largest equity offering in the history of the U.S. stock market, second only to SpaceX's $75 billion fundraising last month.
The numbers are big. But in the context of SK Hynix's size, they aren't that significant.
$26.5 billion, it earns that back in a single quarter. The shares offered accounted for less than 3% of its total market cap. The company only issued 2.5% in new shares.
This doesn't look like the actions of a cash-strapped company.
However, just three years ago, no one imagined Hynix would become the most profitable company in today's AI supply chain. The predecessor of Hynix was Hyundai Electronics, which in 2002 was on the verge of bankruptcy due to a memory industry downturn. Creditors had even negotiated a sale to Micron, which was ultimately blocked by the board and employees.
For the next decade, it struggled under creditor trusteeship until it was acquired and renamed by the SK Group in 2012. In the 2023 memory downcycle, Hynix recorded a historic net loss of $6 billion.
It took only two years to go from the largest loss in history to a 77% net profit margin in Q1. This reversal itself reflects the extreme supply-demand imbalance in the AI era.
In its prospectus for this U.S. IPO, Hynix explicitly stated the funds raised will primarily be used for wafer fab construction in South Korea. That's what they say.
But SK Hynix now generates cash inflows at the $30 billion level in net profit per quarter alone. It is fully capable of funding expansion with its own cash.
So why come here?
The answer lies in valuation.
SK Hynix, listed on the Korea Exchange, saw its Q1 2026 revenue increase 198% year-on-year, and operating profit surge 405% (net profit grew about 398% YoY). Despite the explosive performance, constrained by the cyclical nature of memory and the liquidity discount in the Korean stock market, its reasonable forward P/E ratio midpoint is only about 8x, and its secondary market valuation has long been suppressed below this level.
At the same time, U.S.-based Micron Technology, which also sells memory chips and has a far smaller HBM market share than SK Hynix, maintains a P/E ratio above 23x.
The same industry, the same logic, but the valuation is three times different.
HSBC said bluntly in a research report that over the past 13 years, SK Hynix has been the world's cheapest AI asset, undervalued by an average of 35% compared to Micron.
The reason is simple: Hynix's stock was locked in South Korea.
Coming to Nasdaq isn't for the $26.5 billion; it's to let global capital reprice it.
II. South Korea Can't Hold Hynix Anymore
There's a specialized term in South Korea's capital market: Korea Discount.
This term has existed in investment circles for over a decade. Its meaning is simple: companies of comparable quality are valued lower on the Korean exchange than they would be on a U.S. exchange.
The reasons aren't complex. Insufficient liquidity in the Korean stock market, high barriers for overseas capital entry, limited institutional investors, a high proportion of retail investors, and a weak dividend culture. Combined, these factors naturally lead global capital to discount the Korean market.
Samsung Electronics is the most classic example. One of the world's largest semiconductor manufacturers, its market cap is less than half that of TSMC.
But SK Hynix's situation is more extreme than Samsung's.
Today, everyone knows about GPUs. But what truly makes AI run isn't just GPUs. GPUs handle computation; HBM feeds the data. As GPUs get faster, what actually limits model training speed is no longer the computing power itself, but whether data can be delivered to the GPU fast enough.
HBM is what solves that bottleneck.
And the world's strongest HBM company is SK Hynix.
According to the latest data from IDC, SK Hynix holds 56.4% of the global HBM market share. That's more than double that of its closest rival, Samsung Electronics.
As one of the core links in Nvidia's AI chip supply chain, about 70% of HBM4 orders for Nvidia's next-generation flagship NVL72 Vera Rubin platform have already been assigned to it.
Technologically, Hynix has been leading its rivals by a full generation in HBM since 2021. Samsung's HBM3 only passed Nvidia certification in July 2024, and only for use in the H20 chip targeting the Chinese market. Micron simply skipped HBM3.
SK Hynix's position in the core hardware of the AI era is almost akin to Nvidia's in GPUs.
Then look at its valuation: 8x P/E.
Now look at Nvidia: close to 40x.
The gap between them can't be explained by technology or performance. The only explanation is the listing location.
The essence of the problem lies in the different valuation philosophies of the two markets. The Korean market believes in cash flow and values manufacturing. The U.S. market believes in growth and values the future.
The same company, placed in different valuation systems, commands a completely different price.
Reuters cited a comment from Roundhill Investments CEO Dave Mazza: SK Hynix is one of the most important companies in the world, but U.S. institutions previously found it difficult to easily buy its stock.
It's that simple.
The world's largest AI memory monopolist, inaccessible to most global institutional investors.
In the past, SK Hynix was mainly bought by domestic Korean capital. After coming to Nasdaq, its buyers will become global pension funds, mutual funds, ETFs, sovereign wealth funds, and U.S. long-term capital. These pools of money are larger, have longer holding periods, and have a higher tolerance for growth.
HSBC predicts that after listing on Nasdaq, SK Hynix's price-to-book ratio could rise from the past average of 2.8x to 3.4x. Simply changing the listing location could add hundreds of billions of dollars to its market cap.
This isn't a financing act; it's a migration at the capital market level.
III. A Giant Earning $2 Billion Daily Also Fears Burning Cash
Hynix's choice to go to Nasdaq at this juncture has a very direct factor: AI is burning too much cash.
Each generation of models costs more to train than the last; each generation of GPUs is more expensive than the previous one. HBM prices are rising. Data center construction costs are rising. Power costs are rising. No company dares say that after this round of investment, it won't need to raise more funds in the future.
In this environment, whoever can continuously secure large amounts of cheap capital can keep investing. Conversely, for Hynix to continue expanding capacity and maintaining technological leadership, it equally needs the support of the capital market.
There is also an imminent industrial reason: the underlying logic of memory chips is changing.
Starting from the next-generation HBM4, memory chips are no longer pure memory; they need to incorporate logic processes. Hynix has already clarified that the underlying base die for HBM4 will no longer be produced in-house but outsourced to TSMC, using its customized 2nm or 3nm advanced process nodes.
This means that capital expenditure in the semiconductor industry is undergoing an exponential change. In the past, competition in the memory industry was about internal competition. Hynix only needed to iterate its own memory processes, improving cost efficiency through technological enhancements within its own fabs.
But in the HBM4 era, competition has become cross-sectoral warfare. It now needs to simultaneously pay for TSMC's expensive top-tier wafer fabrication fees, advanced packaging co-development investments, and its own high construction costs for the Yongin industrial cluster in South Korea.
This qualitative shift in technology nodes has created a financial black hole.
Against this backdrop, Micron Technology in the U.S., though with a smaller market share than Hynix, previously leveraged its local capital advantage to skip the HBM3 generation entirely, pouring large amounts of capital raised in the U.S. market into R&D and capacity expansion for the next generation, attempting a technological leapfrog. Samsung Electronics, while temporarily lagging in certification, is also making a full-throttle bet with its massive group financial resources.
In the memory industry, leading by one generation is life; lagging by half a generation is death.
Although Hynix is currently making staggering profits relying on Nvidia's monopoly orders—earning a net profit of $2 billion daily based on Q1 income—the semiconductor industry is cyclical.
Once the pace of AI infrastructure construction experiences a阶段性 slowdown, or competitors achieve breakthroughs in advanced processes, the defensive capital Hynix can mobilize domestically in South Korea will be extremely limited.
Relying solely on retained cash flow to compete against this level of global technological competition is like facing heavy artillery with bare hands.
Recently, the South Korean government and the semiconductor industry announced a $576 billion investment plan, with Samsung and Hynix taking on the core share, aiming to double domestic memory capacity within the next five years.
This level of capital consumption has surpassed the carrying capacity of any single regional market. Hynix must connect to the world's largest capital pool, one that also has the highest tolerance for high-tech R&D spending.
New York is the only choice.
IV. Technology Decides Victory; Capital Decides Survival
Hynix's migration reveals that AI competition is shifting from a contest of technology and products to a stage of competing in capital efficiency.
This brings us to the most brutal accounting logic of the two capital markets.
In the Korean stock market, Hynix's P/E is 8x; in the U.S. market, its peer Micron's is 23x. For the same $1 in profit, in the Korean market, it can only support $0.8 in refinancing or equity pledge capacity. But in the U.S. market, this $1 profit can be leveraged into $2.3 in capital leverage.
The disparity in capital amplification effects is something technology simply cannot compensate for. A company listed in the U.S., even if temporarily behind in technology, can raise several times more capital than its competitors with minimal equity dilution, then use that money to fund R&D, grab capacity, and lock down the supply chain.
This efficiency gap of "overwhelming rivals with money" is the most concealed nuclear weapon.
This is the true nature of the U.S. capital market. It is not just a trading venue; by offering the world's highest valuation premium, it forcibly assimilates the world's most prominent AI assets into its ecosystem.
Once Hynix's trading volume and market cap weighting dominate on Nasdaq, global institutional investors will anchor to its U.S. stock price. The local shares in Seoul will lose their pricing function, becoming mere shadows.
For Hynix, the people who truly decide its worth today are not in Seoul, but in New York.
This isn't about choosing between countries; it's about the brutal logic that has evolved in capital systems for centuries. Models can be trained, GPUs can be bought, HBM capacity can be expanded.
But Hynix's choice tears away the final card: The endgame of AI is evolving into a capital black hole, forcing the world's top assets to converge towards a single center.
In this game, winning the first three rounds doesn't matter. As long as you haven't broken free from dependence on super-large-scale capital, you must accept assimilation into this pricing system.
This is the barrier beyond technology that is truly difficult to break.
Words from 【Beyond the Headlines】:
Many people see Hynix listing again.
Their first reaction is that a Korean company has finally gone to the U.S. for financing.
I, however, think it tells another story.
For decades, capital chased manufacturing. Wherever factories were built, capital followed.
Today, manufacturing is starting to chase capital instead.
Hynix hasn't left South Korea. All its new fabs are still mostly built in South Korea, and the funds raised will continue to be invested back into South Korea.
What has truly left is the stock. Or rather, the company's identity.
Hynix no longer wants only to be the best company in South Korea, but the most important asset in the global AI era.
The gap between these two identities is precisely the gap between the Korea Exchange and Nasdaq.
Hynix hasn't fled South Korea.
It has simply positioned itself ahead of time at the endpoint of capital flows in the AI era.
