By｜Beyond the Headlines, Authors | Banjun, Huahua
America has finally made a move against South Korea.
This time, it’s not about cars, steel, or tariffs.
It’s about memory chips.
On June 25th, Samsung and SK Hynix were hit with a class-action lawsuit in a California federal court. Sharing the defendant’s bench with them was Micron, an American company.
Not even sparing its own.
The accusation is collusion to create a "RAMpocalypse." The three companies are alleged to have slashed traditional DRAM production capacity under the guise of AI transformation, driving up memory prices by 700% over four years.
Four days later, South Korea’s response arrived. Jin Jung-gwan, South Korea’s Minister of Trade, Industry and Energy, announced an investment of 800 trillion won: four new wafer fabs, all-in on memory for the next fifteen years.
America takes action on one side; South Korea doubles down on the other.
This is no ordinary antitrust case. The first true resource war of the AI era has begun.
1. South Korea Steps into the AI Profit Center
Let's look at some core numbers.
The global AI net profit pool in 2026 is estimated at around $637 billion. According to Altimeter’s calculations, the profit distribution looks like this:
The US takes 49%, roughly $314 billion. The core is Nvidia, single-handedly capturing $207 billion.
South Korea takes 35%, roughly $223 billion. Samsung and SK Hynix together account for $222 billion.
The US and South Korea combined account for 84% of global AI profits. All other countries share the remaining 16%.
The global AI business today is essentially a profit split between two countries: the US and South Korea. Nvidia takes the largest slice, the Korean duo takes the second largest. All other nations combined get less than 20%.
South Korea’s 35% profit comes from an incredibly concentrated source: HBM (High Bandwidth Memory).
For high-bandwidth memory, there are only three global giants capable of mass production. SK Hynix holds 57%, Samsung 22%, and Micron 21%.
Nvidia is the GPU king. But GPUs need HBM to function. An H200 requires 141GB of HBM, a B200 requires 192GB. Each GPU needs 6 to 8 HBM chips. Without HBM, a GPU just spins its wheels.
In other words, in the AI era, the hardware bottleneck isn’t GPU computing power, but memory bandwidth.
This creates a supply chain dynamic: the better Nvidia sells, the more money South Korea makes. The faster Nvidia expands production, the greater South Korea’s demand.
Every time GPT is trained, South Korea profits. Every time an Agent is deployed, South Korea profits. Every time a new AI data center is built, South Korea still profits.
How much exactly? SK Hynix’s operating profit margin in Q1 2026 was 72%, surpassing Nvidia’s 65% and TSMC’s 58%, setting a new global record for the semiconductor industry.
Net profit of 40 trillion won in a single quarter. Over 2 billion RMB net profit per day.
From July 2025 to April 2026, over nine months, South Korea added 100 companies with market caps exceeding one trillion won. It took South Korea 10 years to achieve the same growth previously. (Extended reading: Nine Months, South Korea Adds 100 Trillion-Won Companies)
The total market capitalization of the South Korean stock market has more than doubled, surpassing $5 trillion, overtaking India to become the world’s sixth-largest stock market. The KOSPI index has risen 70% this year, breaking 7,000 points.
South Korean household paper wealth has increased by over 1,000 trillion won, approaching 40% of annual GDP. A nationwide stock-buying frenzy has seen many citizens invest in Samsung and SK Hynix, doubling their paper assets.
South Korea is experiencing a national-level wealth creation movement driven by AI chips.
2. Why America is Turning Hostile
Back to that lawsuit.
On the surface, it’s consumer advocacy. DRAM prices rose 700%, consumers couldn’t take it anymore, and sued the manufacturers for monopoly.
But look closely, there’s a detail.
The defendants aren’t just Samsung and SK; Micron is on the defendant's bench too.
American consumers are suing their own country’s chip company along with the Korean ones.
Why?
Because America’s real worry isn’t a single company, or even the price itself.
It’s a bigger issue: AI is turning memory into the new oil.
For the past two decades, the US has firmly controlled key positions in the semiconductor supply chain: CPUs, GPUs, EDA, operating systems, software ecosystems.
Today, it suddenly finds that a critical layer of AI infrastructure is held by South Korea.
Samsung and SK Hynix have "previous convictions." In 2005, the two companies pleaded guilty in the US for DRAM price-fixing, paying a combined $731 million in fines, with several executives serving prison sentences.
The lawsuit specifically dredges up this history, attempting to establish a pattern of "systematic collusion."
And Micron? Micron wasn’t charged in that case. This time it’s listed as a co-defendant, but it’s an American company, has factories in the US, receives federal CHIPS Act subsidies, and is building new fabs in Idaho and New York.
Look at this situation.
On one hand, the US government is pouring money into Micron for domestic fab construction—a $50 billion domestic investment plan, $6.1 billion in federal subsidies. The Trump administration is even considering converting subsidies into equity investments, taking direct stakes as it did with Intel. Trump publicly praised Micron at rallies.
On the other hand, consumers are suing Micron along with the two Korean companies.
Kill a thousand enemies, lose three hundred of your own. But America did it anyway.
The reason is, what truly rankles America is that South Korea, with just two companies, is taking 35% of the global AI industry’s profits. And American consumers and businesses are footing the bill.
3. The Memory Price Surge Isn't Just Collusion
First, let’s be clear about one fact: memory prices are experiencing a strong cyclical upswing, and it won’t stop in the short term.
American financial group Jefferies predicts memory prices will rise another 40% to 50% quarter-on-quarter in Q3 2026, and another 30% to 40% in Q4. For the full year 2027, a year-on-year increase of 40% to 45% is expected. A significant slowdown might not occur until 2028 at the earliest.
But who caused this price surge?
The lawsuit says it’s collusion by three manufacturers. But what’s really happening is more complex than the complaint suggests.
In the past, how much DDR memory versus HBM a fab produced was balanced based on market demand. If DDR demand was high, more DDR was made.
AI has disrupted that balance.
HBM chips occupy twice the wafer area of regular DDR chips. In 2026, HBM is expected to account for 25% of global DRAM wafer capacity, with demand growing 70% annually. Global total capacity is only growing 14%, with traditional DRAM allocation growing a mere 10%.
Global large model companies are frantically expanding data centers. Nvidia’s GPUs are selling at capacity limits, with each GPU requiring massive amounts of HBM. The three manufacturers shifting capacity to HBM is commercial rationality—HBM offers 72% margins, while regular DRAM might be 20% to 30%.
Faced with that profit gap, any enterprise would choose to pivot to HBM.
The problem is, the three of them combined control over 95% of the global DRAM market. When all three make the same decision simultaneously, even if they never sat in a conference room to discuss it, the effect is indistinguishable from collusion.
This is a structural issue of an oligopolistic market.
The result is: the more AI flourishes, the higher the cost for ordinary computers, phones, and servers. Apple is merely the first to hand this bill to consumers; it won’t be the last.
And American consumers don’t care about all that. They just know a DDR5 memory stick that cost $200 four years ago now costs $1,400.
4. The Lawsuit is Just the Surface
If you only look at the surface, this is an antitrust lawsuit. It might involve fines, settlements, or drag on for years.
But in the broader picture, this lawsuit’s function is to apply pressure.
America’s real demand is clear: memory manufacturing must be reshored.
Micron received $6.1 billion in CHIPS Act subsidies, aiming to invest $50 billion domestically in the US by 2030, achieving 40% domestic DRAM production capacity.
The logic is straightforward: 35% of global AI profits flow to South Korea because HBM manufacturing capability is there. If Micron can capture more share, those profits return to America.
A single lawsuit, triple effect:
First, apply legal and public pressure on Korean companies, increasing their operational costs in the US market.
Second, buy time for Micron. Micron’s New York fab is already delayed, pushed from 2028 to 2030 for production. Micron needs time to catch up. A lawsuit tying up competitors helps itself.
Third, reinforce the national security narrative. If the court finds Korean companies guilty of price manipulation, subsequent policy interventions in the memory chip supply chain gain legitimacy.
Plainly speaking, America insists on pushing the lawsuit because the long-term gains outweigh the short-term losses.
5. South Korea Doubles Down Instead
After the American antitrust lawsuit, South Korea offered no explanations. It directly responded with a national-level industrial investment plan.
South Korea’s official announcement of a trillion-dollar semiconductor investment on June 29th is essentially doubling down on the construction of the Yongin Semiconductor Mega Cluster. This isn’t just building a few fabs; it’s about concentrating the entire HBM chain—from design, materials, packaging to mass production—within the same region.
This highly integrated supply chain can shorten R&D iteration cycles, creating generational advantages over new, geographically dispersed overseas capacities.
Financially, the two semiconductor giants’ total investment plan exceeding a trillion dollars is backed by robust cash flow.
South Korea’s logic is clear: You file a lawsuit, I expand production. You try to tie me down with law, I outpace you with scale.
Moreover, South Korea has an advantage America cannot replicate: speed.
SK Hynix’s profit margin is 72%. Samsung’s operating profit in Q1 2016 was 57 trillion won. They have ample cash flow to support expansion without waiting for government subsidy processes.
Micron’s New York fab won’t be operational until 2030. South Korea’s new fabs could start shipping by 2028.
This time gap is South Korea’s moat.
Minister Jin Jung-gwan said one thing: the global memory chip market size will grow to four times its current size in the next five years. South Korea’s assessment is that market growth is far faster than America’s catch-up speed. As long as AI demand continues to explode, South Korea’s production lead will continue to widen.
6. Symbiosis or Conflict?
From a longer perspective.
In the past, the world competed over who had more users.
Today, the world is beginning to compete over who possesses more capacity to produce intelligence. GPUs, HBM, electricity, data centers—these things once hidden in server rooms are becoming new national-level strategic resources.
Nvidia is the biggest winner of the AI era. No one disputes that.
But SK Hynix and Samsung are upstream from Nvidia. Every Nvidia GPU requires 6 to 8 HBM chips. Without HBM, the GPU is a piece of useless silicon.
This is a dependency America cannot unilaterally sever. Nvidia cannot avoid using Korean HBM; Micron only has a 21% share, and its capacity is still under construction. At least until 2028, Nvidia’s reliance on the Korean supply chain won’t fundamentally change.
America’s dilemma is that it nurtured Nvidia, but Nvidia’s supply chain delivers the largest slice of profit to South Korea.
GPU profits go to America. HBM profits go to South Korea. Together, they devour 84% of global AI profits. AI has, for the first time, placed a country of less than 52 million people on the second tier of the global tech profit chain.
Symbiosis, because they need each other. Nvidia needs HBM; South Korea needs Nvidia’s orders.
Tension, because profit distribution is never static. When scale is large enough, the distribution method itself sparks conflict.
Today’s lawsuit is just the first public expression of this tension.
So America suing Samsung and SK today isn’t just about price. South Korea’s frantic fab-building isn’t just about making money.
They’re not fighting over chips, but over their position in the next generation of the world’s industrial system.
Words from 【Beyond the Headlines】:
I checked the outcome of that 2005 DRAM price-fixing case.
Samsung pleaded guilty, fined $300 million. SK Hynix pleaded guilty, fined $185 million. Including Elpida (a Japanese memory company), total fines reached $731 million. Several executives went to prison.
Twenty years ago, America sued Korean chip companies because memory was hurting Dell and HP’s procurement costs. That was a PC-era story.
Today, America is suing Korean chip companies because the HBM shift is pulling the global AI profit center towards Seoul.
The case is the same, but the era is not. Last time it was a trade dispute; this time it’s industrial sovereignty competition.
I don’t know how this lawsuit will be categorized looking back five years from now. But I have one judgment:
It might be the first landmark event in the resource competition of the AI era.
GPUs, HBM, electricity, data centers—these things are becoming the oil, steel, and railways of the new era.
And we all know the stories of oil, steel, and railways. Every time, they eventually become matters of the state.
This time will be no exception.
