How did SK Hynix, the king of matchmaking battle suits in South Korea, counterattack Samsung?

Author: Deep Tide TechFlow

In a matchmaking agency in Seoul, a strange phenomenon has recently emerged.

Some men coming for blind dates deliberately place their business cards deep inside their suit pockets, only carefully taking out the card with the company name after confirming that the other person’s “behavior is decent.” The card bears four English letters: SK Hynix.

Source: Korean variety show

Korean matchmaking company Gayeon’s senior executive Kang Eun-sun publicly told the media that after the start of the semiconductor super cycle, the popularity of employees from Samsung Electronics and SK Hynix has continued to rise, “The market clearly prefers engineers with much higher actual income, surpassing some lawyers whose income is no longer as high as before.” On social media, there’s even a joke: “Hynix employees go on blind dates and humbly say they work at Samsung. Only when they meet a good character do they honestly admit they actually work at Hynix.”

Turning a work uniform into a “matchmaking armor” is driven by a set of numbers that make workers around the world turn red with envy.

By 2025, SK Hynix’s operating profit reached 47.2 trillion Korean won. According to a new agreement with the union last September, 10% of the operating profit goes into an employee bonus pool. With an average of 35k employees, each can receive about 140 million Korean won, roughly 650k RMB.

In the first quarter of this year, SK Hynix’s operating profit increased over 400% year-on-year to 37.6 trillion Korean won. Based on analyst forecasts from different countries, this year’s operating profit is expected to be between 210 trillion and 250 trillion Korean won, estimating the per capita bonus to be in the range of 2.9 million to 3.3 million RMB.

International investment bank Macquarie Securities further predicts that by 2027, operating profit could reach 447 trillion Korean won, with per capita bonuses possibly soaring to 1.29 billion Korean won, about 6.1 million RMB.

A more compelling story than “610,000 RMB per person” is: This company, long the second in Korea’s semiconductor industry, is the little brother standing behind Samsung.

What did it do right to let Samsung—whose screens and chips even Apple depends on—watch its position as the global storage leader slip away?

2008: From the brink of bankruptcy

Let’s rewind to 2008, when no one would have called Hynix a “future overlord.”

Its predecessor was Hyundai Electronics. In 2001, during the dot-com bubble burst, DRAM prices plummeted, and the company was saddled with a massive $14 billion debt, taken over by creditors, entering a five-year “workout” process—similar to “restructuring custody” in Korean discourse. For five years, factories, R&D budgets, and staffing were all squeezed tight.

By 2007, Hynix finally crawled out of this “custody” state, but was still weak and barely surviving.

At this moment, on the other side of the Pacific, a company called AMD came knocking.

AMD’s situation was also not optimistic. It was the second in the GPU market, squeezed by NVIDIA in gaming graphics cards. Its researcher Bryan Black was working on a “high bandwidth memory” (HBM) concept—stacking multiple DRAM chips vertically like building blocks, connected through TSV (through-silicon via) technology.

Why do this? Because AMD saw an overlooked problem: CPU/GPU computing speeds were increasing rapidly, but memory data transfer couldn’t keep up. After calculations, the compute units often had to wait for memory to deliver the next batch of data, a dilemma called the “Memory Wall.”

To illustrate roughly: a super chef slices 10 dishes per second, but the waiter can only deliver ingredients at 2 per second. The chef spends 80% of the time waiting. No matter how fast the chip is, if data can’t come in, it’s just idle.

AMD’s idea was straightforward: instead of expanding data channels horizontally (the traditional DDR approach), why not make the memory “taller,” stacking it vertically on the chip, so data can be delivered over shorter distances via wider buses? This vertical stack is what HBM is.

It sounded promising. But in 2008, there was no AI demand, no large model training, no “computing revolution.” The only visible application for HBM then was high-end gaming graphics cards. The market was small, manufacturing was extremely difficult, and each chip cost far more than regular DRAM.

AMD looked around but found no takers. Samsung was unwilling; it was fully committed to HMC (Hybrid Memory Cube), another vertical stacking tech it developed with Micron. Micron also refused, following Samsung’s lead.

The only company willing was Hynix, freshly out of bankruptcy, hesitant to take on big orders.

In 2009, Hynix officially launched HBM development. It wasn’t until 2013 that the first HBM chip was produced at Hynix’s factory in Icheon, Korea.

Who would have thought that this chip, 15 years later, would become the product all AI giants are rushing to buy, with capacity “sold until 2030”?

No one, including Hynix itself.

2012: A gamble by a chairman

Another key figure entered the story in 2012.

SK Group Chairman Chey Tae-won led a consortium initiated by SK Telecom, acquiring 21.05% of Hynix’s shares from creditors for about 3.4 trillion Korean won (roughly $472k). From then on, the semiconductor company was called SK Hynix.

What kind of person is Chey Tae-won? A description from the book Super Momentum says: After the acquisition, he personally met with 100 Hynix executives one-on-one in a short period. His first move wasn’t layoffs or cost cuts, but integrating SK Group’s management system with Hynix’s technological capabilities, then reopening paused investments in fabs (wafer factories) and process improvements.

The timing was crucial.

From 2012 to 2014, the entire DRAM industry was still shadowed by the downturn in storage prices in 2011-2012. Rational financial models advised management to be conservative, considering it a low point in the cycle. But Chey Tae-won made a counterintuitive decision: expand investments.

And most importantly, HBM.

Between 2011 and 2022, SK Hynix invested about 860 billion Korean won in HBM-related R&D, and about 1.5 trillion won in facilities and equipment. Much of this was during years of market downturn and when HBM’s commercial prospects seemed bleak.

What happened during this period?

HBM2 failed to meet performance expectations and had to be redesigned into “HBM2 Gen2.” The HBM team became a “department no one wanted to join,” with core members transferred away and morale low. AMD’s flagship R9 Fury X graphics card in 2015 used the first-generation HBM, but market response was lukewarm because of its high cost.

What made Hynix panic was the 2016-2017 period, when Broadcom approached Samsung to supply HBM2 for Google’s TPU second generation. If Samsung could meet 100% of Broadcom’s demand, Broadcom promised exclusive supply rights. This was supposed to be HBM’s first big entry into data centers.

But Samsung failed.

A report from Korea’s JoongAng Daily revealed the chaos: Google’s TPU project involved Broadcom (design), Samsung (memory), and TSMC (foundry). Samsung’s HBM had memory issues; engineers reported that TSMC refused to let them inspect the factory. The three companies kept passing the buck, delaying for up to half a year. “Such deadlocks were common between 2016 and 2017,” recalled an insider executive.

Later, Google gradually leaned toward working with SK Hynix. The first real deployment of HBM in data centers, however, was not Samsung’s victory.

But at that time, the world had no idea that HBM would become the most critical bottleneck in the AI era.

Chey Tae-won later said in an interview with Super Momentum: “We are at a crossroads.”

He didn’t elaborate on why he persisted. Looking back, the logic might be: Hynix’s fate had already flirted with bankruptcy in 2001. Unlike Samsung, which had diversified businesses to spread risk, Hynix only had memory. To be the global leader in this field or remain forever the little brother behind Samsung.

So “betting on HBM” wasn’t a choice for Hynix—it was a necessity.

2022: Jensen Huang hands over a matchstick

In June 2022, SK Hynix began mass production of HBM3. That year, the first batch of HBM3 chips was installed in a GPU called H100, from a company valued at about $300 billion, which was still “important but not game-changing” in gaming and data center markets—NVIDIA.

In November, OpenAI released ChatGPT.

What happened next is well known. The demand curve for AI computing power shifted from a gentle slope to an almost vertical rocket trajectory. Every GPU used for training large models needed HBM as its closest “data transporter.”

At that moment, Hynix’s 14-year gamble was suddenly revealed.

By Q2 2025, Hynix held 62% of the global HBM market share. Samsung fell to 17%, even behind Micron (21%).

By the end of 2025, Hynix’s annual operating profit reached 47.2 trillion Korean won, surpassing Samsung Electronics’ 43.6 trillion won for the first time in history.

NVIDIA CEO Jensen Huang visited SK Hynix’s booth at Computex Taipei in August 2025, leaving a handwritten note on the display: “JHH LOVES SK HYNIX!” This photo was circulated repeatedly by Korean media. In engineer culture, there’s no more direct official endorsement than this.

Privately, SK Hynix engineers gave HBM a new full name: they say HBM actually stands for “Hynix Best Memory.”

Where did Samsung lose?

So, the question is: what went wrong for Samsung, the giant that once crushed Japanese DRAM factories and pushed Micron into a corner?

Samsung missed the early HBM layout, chose the wrong technical route (NCF packaging vs. Hynix’s MR-MUF packaging), and HBM3E kept failing NVIDIA’s certification tests…

These are facts, but not the root cause. The fundamental reason is more awkward and ironic: Samsung was too successful, so it couldn’t afford to lose or take big risks.

Compare the two companies at that 2008 point. Hynix had just come out of bankruptcy, with only memory as its business, lacking diversified cash cows.

It had to bet on HBM not because it was visionary, but because it had no other choice. Any opportunity to shed the “little brother behind Samsung” label had to be seized.

And Samsung?

In 2008, Samsung was on the verge of its peak. Its mobile business was about to take off with Galaxy, its semiconductor business was the world’s top in DRAM and NAND, and its display division was about to land Apple’s OLED orders. Its cash flow was abundant, its business scope vast, and stakeholders’ interests highly complex.

For a company like that, what was HBM in 2008? A high-risk gamble with a tiny market, long return cycle, and potential conflict with its main tech route (HMC). Any rational finance committee wouldn’t approve full commitment.

This is the classic “Innovator’s Dilemma”: successful large companies are always constrained by their own success. They have too much to lose, too much market to protect, to risk everything on a seemingly unreliable new direction.

The deeper irony is that Samsung was not unaware of HBM. It started investing in HBM research as early as 2011, and in 2016, it led mass production of HBM2. But each time, Samsung didn’t go all in. Its efforts were split among HMC, GDDR, LPDDR, enterprise SSDs, and more. When Hynix’s HBM team was “marginalized but still fighting,” Samsung’s HBM team was also “marginalized,” just without anyone fighting for them.

By 2024-2025, Samsung finally realized it had to go all-in on HBM—but it was too late. The technological gap had widened, and the customer relationships built by NVIDIA and SK Hynix had formed a formidable moat.

In a New Year’s speech in 2026, Samsung Semiconductor Vice President Jun Young-hyun said: “Customers tell us, Samsung is back.”

“Back,” itself, is a form of self-admission.

Two questions

What does SK Hynix’s story mean? At least two issues deserve attention.

First, why are similar stories more likely to happen in Korea than elsewhere?

SK Hynix’s success didn’t come out of nowhere. It’s rooted in a special industrial soil—Korea’s chaebol system, which, despite decades of criticism, objectively allows a company to rely on a decision-maker’s will to bet on a 20-year return cycle, continuously funding it even when no commercial prospects are visible for a decade.

When Chey Tae-won acquired Hynix in 2012, no Wall Street analyst was shouting “quarterly earnings, quarterly earnings” in his ear. He didn’t need to prove HBM’s ROI every quarter to the board.

Such long-term decision-making cycles are increasingly rare in today’s US-driven tech companies. They are also a key variable for Chinese hard-tech firms like Yangtze Memory and Changxin Storage to break through. Technology isn’t the biggest obstacle; capital and willingness of decision-makers to sit through ten years of cold storage are.

Second, is Hynix’s “second-place fate” already over?

Not necessarily.

By Q4 2025, Samsung had regained the world’s top spot in total memory revenue. It is accelerating in HBM4, with key certifications nearing completion. Counterpoint’s director MS Hwang believes Samsung might overcome last year’s quality issues with HBM4 and make a significant turnaround.

Longer-term, Hynix’s current moat also has vulnerabilities. Its customer base is highly concentrated (NVIDIA orders dominate), its MR-MUF packaging faces warping issues beyond 16 layers, and expansion costs in 2026-2027 will drag on free cash flow. Chinese manufacturers are catching up; Changxin Storage’s HBM is expected to mass produce by 2027. Once that line is operational, the global HBM oligopoly could shift again.

But none of this changes one fact: SK Hynix has already demonstrated a possibility—an “underdog” long considered forever behind the giants, can turn itself into a pioneer of a new era in those 20 years of ridicule.

A market law is reaffirmed: when everyone chases certainty, betting on an uncertain long-term direction often yields the greatest alpha.

This was HBM in 2008, the EV industry chain in 2018, and perhaps another overlooked thing in 2026.

Don’t ask “who is today’s SK Hynix,” but rather: who is doing today what SK Hynix was doing in 2008, but everyone laughs at?