Net Profit Surpasses Samsung Electronics, How Does SK Hynix Soar in the AI Era?

When it comes to stock trading, rely on King Kylin Analyst Reports—authoritative, professional, timely, comprehensive—helping you uncover potential thematic opportunities!

(Source: Lishi Business Review)

Introduction: From the abyss to the peak, the starting point that changed everything was not some grand strategy, but SK Hynix’s nearly stubborn refusal 20 years ago.

Author: Wang Jian / Lishi Business Review / Producer

In January 2026, SK Hynix announced its fiscal year 2025 financial report: revenue reached 97.15 trillion won (about $700 billion), operating profit was 47.21 trillion won (about $330 billion), profit margin 49%, with a total market value exceeding $430 billion.

Benefiting from the booming demand for high-performance storage driven by artificial intelligence development, South Korea’s memory giant SK Hynix for the first time surpassed Samsung Electronics (about $303 billion), which had dominated the memory market for 33 years, in operating profit.

Immediately, SK Hynix announced a performance bonus of 1500% of basic salary for all employees, averaging about 1.1 billion won (roughly 540,000 RMB) per person. It seems like a “lucky story” that hit the AI track right.

But 20 years ago, SK Hynix’s stock price was only 125 won (about $0.0839), and its debt-to-asset ratio was a staggering 206%, on the brink of bankruptcy and liquidation. From the abyss to the peak, the starting point that changed everything was not some grand strategy, but that nearly stubborn refusal 20 years ago.

1

A life-and-death moment

In 1999, under strong leadership from the Korean government, Hyundai Electronics (the predecessor of Hynix) acquired LG Semiconductors, taking on nearly $14 billion in heavy debt, according to media reports at the time.

Just two years later, due to fierce internal struggles within Hyundai Group, this hot potato was completely divested and abandoned, renamed Hynix, left alone to the creditors.

In 2001, the divested business was renamed “Hynix” and became independent, with annual sales of only 4 trillion won (about $31 billion), but losses reached 5 trillion won (about $38.74 billion), and debt-to-asset ratio soared to 206%.

Faced with such dire circumstances, the parent company chose to wash its hands of it, handing it over to the creditors’ bank for management, and the company was on the verge of bankruptcy.

In any industry, this was almost a death sentence.

What truly pulled it back from the brink was a national-level rescue that remains controversial to this day.

To save this semiconductor company, the Korean government coordinated with KEB Exchange Bank, Korea Development Bank (KDB), and several commercial banks to carry out a large-scale debt restructuring.

Through debt forgiveness, loan extensions, interest rate discounts, debt-to-equity swaps, and other “combination punches,” Hynix was given a breathing space, but was soon recognized by the U.S. Department of Commerce as engaging in “actionable subsidies,” leading to anti-subsidy tariffs on imported Korean DRAM.

Unwilling to accept this, the Korean government took the case to the WTO dispute settlement mechanism, accusing the U.S. anti-subsidy measures of violating international trade rules.

In 2005, the WTO dispute panel largely supported Korea’s position, ruling that the U.S. Department of Commerce failed to properly prove that the bank actions constituted “government instructions or mandates.”

This four-year international trade dispute centered on the U.S. view that rescuing a dying private enterprise through national banks was not consistent with market economy principles, and amounted to “state-level fraud.”

But regardless, the debt restructuring at least kept Hynix alive for the time being.

But the real test was still ahead.

U.S. memory company Micron Technology soon proposed a takeover plan, intending to buy Hynix’s memory chip business for about $4 billion, without assuming roughly $6 billion in debt or other businesses.

In simple terms, Micron planned to buy Hynix’s “shell,” but leave the debt behind, with the remaining entity continuing to carry it.

From a financial perspective, accepting this plan would at least recover some cash and reduce creditor losses.

But faced with such an aggressive acquisition proposal, the Hynix board of directors voted unanimously against.

Meanwhile, Hynix’s management and union also surprisingly stood together, voting to refuse the sale.

In the end, Hynix chose a more painful path: agreeing to a court-led debt restructuring, laying off thousands of employees, divesting all non-core businesses, and placing management under the full supervision of the creditors.

Although Hynix’s situation was dire, many internal elites chose to stay and were willing to weather the storm with the company.

2

Perseverance behind success

Among the engineers who stayed was a young man named Kwak Noh-jung, who had recently joined the company.

He later recalled those years and said: “At that time, to save electricity costs, engineers had to turn off fluorescent lights one by one. Even in the company cafeteria, to cut expenses, we launched a paperless ‘handkerchief movement,’ and everyone took unpaid leave.”

What’s most noteworthy in this story isn’t the “hardship,” but the word “engineer.”

In a company that was almost unable to survive, the top technical talents not only did not leave but also did something seemingly incredible.

With no funds to buy new lithography machines, this group of engineers began physically modifying the old 8-inch wafer production lines, optimizing lithography processes, and ultimately managed to mass-produce advanced 0.13μm DDR memory chips on outdated equipment.

This secret project, called “Blue Chip Plan,” did not generate significant profits itself, but it achieved two crucial things: first, it validated their technical potential at very low cost, buying the company some breathing room; second, and more importantly, it protected the company’s “technological fire” and the cohesion of the R&D team during the darkest hours.

Therefore, for Hynix in distress, refusing Micron’s acquisition was never just a financial decision, but a declaration of “who we are”: even in despair, we will never give up our technological accumulation.

It was this accumulation that later became a key foundation for its continuous ascent.

Over the next decade, Hynix relied on a stubborn spirit to gradually climb out of debt.

But what supported this journey was not only the engineers’ persistence but also a rare thing in a purely market environment: capital patience.

In 2012, Chey Tae-won, head of SK Group, during a semiconductor industry downturn, acquired 21.05% of Hynix’s shares for about 3.4 trillion won, becoming its largest shareholder, and the company was renamed “SK Hynix.”

This deal faced fierce opposition within SK Group, with no one knowing when the investment would pay off.

But Chey Tae-won pushed through against all opposition and completed the acquisition.

In fact, even before SK Group’s takeover, he had injected personal assets as collateral into the company.

As a value-driven chaebol, Chey Tae-won was not reckless; his confidence stemmed from SK’s unique corporate system.

SK Group is one of Korea’s top conglomerates, spanning telecommunications, energy, chemicals, and more, with profitable businesses forming a stable internal ecosystem that could continuously “fund” the strategic gamble on semiconductors, freeing it from quarterly profit constraints.

More critically, as a core chaebol deeply tied to the national economy, SK Group could more easily secure long-term support from the government during industry crises or when seizing strategic high ground.

It’s this “internal circulation with external safety net” structure that gave Chey Tae-won the confidence to invest long-term in Hynix, waiting for an uncertain “spring.”

At this time, Kwak Noh-jung had been with the company for nearly ten years, rising from an ordinary engineer to a core management role overseeing semiconductor manufacturing.

And the next chapter of Hynix’s story would be written by him.

3

How excellence is cultivated

In 2013, shortly after SK Group’s acquisition, Kwak Noh-jung, as a core management figure overseeing manufacturing and R&D, led the launch of the world’s first HBM (High Bandwidth Memory) chip.

At that time, industry attention was focused on smartphone memory and server DRAM, and few thought this HBM chip was very important.

The earliest commercial application of HBM was in AMD’s high-end graphics card Fury X, but sales were limited, and market response was lukewarm.

This cold reception lasted nearly a decade until the explosive rise of AI technology.

It was AI’s rapid boom that magnified the core bottleneck HBM aimed to solve into the industry’s top challenge.

To understand why HBM later became so critical, first understand what AI large model training consumes.

Typically, training AI models requires massive computing power, mainly GPUs (graphics processing units). But GPUs are just engines; without enough memory, they can’t run.

A quick science note: ordinary DDR memory is like a rural road, data travels from memory to GPU via one vehicle at a time.

In contrast, HBM’s structure is completely different: it stacks multiple memory chips vertically, connected directly to the GPU through thousands of tiny vertical silicon vias (TSV, about 10 microns in diameter, 100 microns deep, with thousands per layer), effectively building an eight-lane highway next to the engine, vastly increasing data transfer speeds.

Reports suggest that OpenAI engineers estimated that during GPT-4 training, about 90% of the chip’s time was spent on data transfer, with only 10% on actual computation.

In other words, the bottleneck in AI computing power isn’t the GPU itself, but whether data can be delivered fast enough to the GPU.

This explains why, when ChatGPT emerged at the end of 2022, the entire AI industry’s demand for computing power exploded exponentially within months, causing NVIDIA’s GPU orders to be booked two years in advance.

This huge market demand suddenly made HBM jump from the sidelines to the most sought-after component in the entire supply chain.

And SK Hynix had already been on this path for ten years.

Over that decade, SK Hynix’s R&D on HBM evolved from the first generation to the third (HBM3), and now to HBM3E. Each generation saw significant improvements in bandwidth and energy efficiency, with each iteration’s accumulated experience embedded in engineers’ expertise and processes.

This accumulated R&D and technology became the most difficult barrier for competitors to copy.

Fast forward to 2022, when global doubts about AI’s rapid development arose, fearing that demand was a false proposition, and industry forecasts predicted memory demand would continue to shrink.

Within SK Hynix, the finance side began questioning the ROI of the HBM project, suggesting cuts or budget reductions. Fortunately, led by Kwak Noh-jung, the R&D team withstood the pressure.

Even so, by 2023, the global DRAM market remained severely oversupplied, with prices falling continuously, and the industry still struggling.

For SK Hynix, abandoning HBM at that time seemed a perfectly rational choice.

At this moment, Kwak Noh-jung, as the top executive overseeing semiconductor manufacturing, took the initiative to persuade the board to promote him to co-CEO, granting him absolute authority to push the project forward.

He had been with the company for 28 years, and no one knew better what the future AI demand truly looked like.

He once said: “Even if AI has to wait three more years, we must start preparing now.”

As a technical expert, it was not easy for him to say this during such tough times, but it also showed his sharp foresight.

And as an engineer who had experienced nearly the entire semiconductor era, he knew clearly that if they gave up on HBM development now, they wouldn’t be able to catch up when AI truly arrived.

Kwak Noh-jung’s passionate appeal was recognized by the board, and he was promoted to co-CEO with maximum authority.

His first move after gaining power was not to cut back but to increase: while the industry was generally reducing R&D spending and waiting for the cycle to bottom out, he boosted SK Hynix’s HBM R&D budget by about 30%.

This decision, made against internal opposition, later proved to be SK Hynix’s most correct choice.

4

Why SK Hynix?

In the second half of 2023, with the rapid iteration of AI technology driven by ChatGPT, HBM demand surged, with delivery times extending from weeks to months or even over a year.

By 2025, SK Hynix’s annual HBM capacity was fully booked, with orders from multiple companies including NVIDIA.

Some may wonder: when giants like Samsung and Micron are also eyeing this market, why did SK Hynix become the indispensable supplier?

It’s true that the storage chip industry has never been dominated by SK Hynix alone.

Samsung Electronics is the undisputed leader, with strong technological reserves, the largest global capacity, and R&D budgets several times that of SK Hynix.

Micron Technology, as the only U.S. DRAM manufacturer, backed by the domestic market, has also continued investments in HBM, gaining substantial results.

In terms of size, experience, and resources, SK Hynix does not have the advantage.

But NVIDIA and other companies still chose SK Hynix.

The reason isn’t price or capacity, but timing.

SK Hynix was nearly two years ahead of Samsung in HBM, and even longer than Micron.

Two years may not seem long, but in the semiconductor industry, it means hundreds of engineering batches’ yield accumulation, thousands of process adjustments, and the formation of technological barriers in packaging, stacking, and interconnection that competitors can’t replicate in the short term.

More importantly, SK Hynix did something most suppliers wouldn’t: it directly embedded engineers into NVIDIA.

During the critical development phase of NVIDIA’s H100, SK Hynix assembled a top-tier engineering team stationed at NVIDIA’s headquarters in Santa Clara.

During collaboration, their desks were right next to NVIDIA’s R&D engineers, and NVIDIA engineers could discuss technical issues face-to-face with SK Hynix experts by simply turning their chairs.

This scene went beyond traditional supplier relationships.

The usual logic is: you order, I produce, I deliver. But SK Hynix’s approach was: your problems are my problems; your R&D pace is my R&D pace.

Kwak Noh-jung described this relationship as “embedded partnership,” emphasizing that a company must “become an irreplaceable role for the customer.”

This “on-site team” is the best illustration of that.

This deep integration also created an enviable matching model: when NVIDIA’s H100 chips shipped, SK Hynix’s HBM3 had already completed full compatibility debugging; when NVIDIA’s H200 and B100 were released, SK Hynix’s HBM3E was always the first to pass verification.

This nearly synchronized R&D rhythm made it very difficult for later entrants to insert themselves.

In comparison, Samsung’s situation was more like that of a “good student” with worries.

5

What did Samsung and Micron lack?

Compared to the other two competitors, Samsung is indeed strong in everything—resources, technology, scale—but hesitated when it came to betting everything on a “niche elective” (HBM).

After all, in the past decade, HBM was a small market with no visible returns, while the traditional memory market driven by smartphones and cloud computing was large and profitable.

So Samsung invested huge sums and top engineers into these “main courses” that seemed to guarantee victory, and this judgment appeared entirely correct at the time.

But when the AI exam suddenly arrived, everyone realized that the “niche elective” was actually the key to victory.

Samsung wasn’t unprepared, just late to act.

By the time Samsung mobilized heavy resources to perfect HBM, it was already too late: the manufacturing “feel” and “experience,” such as thousands of tiny silicon vias, require extensive time and real production line “grinding” to master.

In 2024, Samsung’s HBM3E samples were rejected by NVIDIA due to yield issues, stuck at this “feel” stage.

Huang Renxun publicly said, “Waiting for Samsung to pass the test,” meaning: cooperation requires not just technical compliance but also stable large-scale production capacity, which is the real ticket to entry.

And SK Hynix had been quietly accumulating this “ticket” for ten years.

Micron’s story is another version—more like a shrewd “acquirer.”

In 2013, Micron acquired Elpida of Japan, instantly becoming a memory giant. But the catch was: Elpida specialized in mobile memory and knew almost nothing about HBM.

So when HBM became important, Micron had to build this product line from scratch.

Although Micron quickly caught up with the technology thanks to strong engineering, and became NVIDIA’s second certified supplier, the long-term “comrade” relationship between SK Hynix and NVIDIA was clearly different.

Therefore, when giants like Microsoft and Google invested hundreds of billions into AI, industry players were surprised to find that almost only SK Hynix could reliably supply top-tier HBM.

The result of this competition wasn’t just market share (SK Hynix now holds over 60% of the HBM market), but a complete reshaping of product value.

Because HBM’s profit margin far exceeds that of ordinary memory, SK Hynix is no longer selling “electronics commodities” with cyclical prices, but a “core component” with pricing power in the AI world.

Faced with booming demand, Samsung and Micron still had opportunities, but both needed time to bridge the gap of “experience” and “trust” created by different choices.

In today’s explosive AI development, time is precisely the most scarce resource.

6

First-mover advantage is never a permanent moat

SK Hynix’s victory seems on the surface to be about betting on the right AI track.

But more accurately, it’s about accumulating the right capabilities at the right time, then waiting for an unprecedented demand surge.

AI is just the last fuse; the explosives were laid long ago.

It’s worth noting that if the AI wave had arrived five years later, SK Hynix’s story might have been entirely different.

And the current CEO Kwak Noh-jung’s trajectory is a testament to SK Hynix’s 20-year history.

From an engineer willing to weather hardships with the company, to a leader insisting on R&D investment in HBM despite opposition, to a proactive co-CEO persuading the board to grant him absolute authority to push the project forward.

This is not just a victory of an individual or a company, but a perfect coordination of the entire system: government’s safety net, chaebol’s capital patience, engineers’ perseverance.

Lacking any one of these could have halted the story at some point.

The most critical “capital patience” sounds simple but is extremely difficult to sustain in market logic.

A purely listed company, under quarterly report pressures, finds it hard to make decisions like “continue to burn money supporting a project that won’t be profitable for ten years.”

The structural advantage of the SK chaebol system lies precisely here: internal cash flows from cross-shareholdings, policy-based credit support during crises, turning “counter-cyclical investment” that seems irrational into a sustainable institutional arrangement.

Viewed in a broader context, this isn’t just SK Hynix’s story.

The semiconductor industry has always been one of the most intense battlegrounds in geopolitical competition.

U.S. subsidies under the CHIPS Act, Japan’s advanced semiconductor manufacturing subsidies, Europe’s chip acts—all aim to leverage national strength to provide the kind of “capital patience” that markets cannot.

Similarly, Chinese memory chip companies are also on a similar path: Changxin Memory Technologies relies on policy support, continuously breaking through in mid- and low-end DRAM, gradually squeezing traditional manufacturers’ profits.

The next step for these competitors to truly challenge in the HBM track still depends on three key factors: whether their technological accumulation is deep enough, whether their capital patience is long enough, and whether engineers are willing to stay during the toughest times.

These factors won’t disappear with changing competition patterns, nor will they last forever just because of current leadership.

On February 14, 2026, just before Jensen Huang’s 63rd birthday, he hosted a dinner at a fried chicken shop in Santa Clara, California, inviting 30 core engineers from SK Hynix.

He personally poured drinks, thanked them one by one, then said: “Please ensure the supply of the highest-performance HBM4.”

The day before, on February 13, Samsung announced the first shipment of HBM4.

Because SK Hynix’s HBM4 requires increasing the stacking layers from 12 to 16, the technical difficulty has significantly increased, causing shipment delays due to technical issues.

Thus, Huang’s gesture was not just gratitude.

He was confirming: can that “irreplaceable role” continue to be irreplaceable in the next generation of products?

Although SK Hynix’s advantage was built over many cold winters, relying on engineers who never left, R&D investments that saw no immediate returns, and countercyclical decisions that defied human nature, it was accumulated little by little.

But in the face of rapidly changing technology, first-mover advantage is never a permanent moat.

Huang Renxun’s birthday dinner was also a reminder: in the face of the next technological paradigm, today’s barriers can be overturned and rebuilt at any time.