NextFin News - Samsung Electronics, SK hynix, and Micron have become the clearest beneficiaries of the AI memory boom, and now they are also the subject of a new antitrust fight over how that boom has been managed. A class-action complaint filed on June 25 in California accuses the three chipmakers of using the shift to high-bandwidth memory, or HBM, as a pretext to choke legacy DRAM supply and drive prices higher. The companies have not yet responded in court, and the allegations remain unproven, but the case lands at a moment when advanced memory is one of the most strategic bottlenecks in AI infrastructure.
The commercial backdrop is easy to see. SK hynix said in a January 5 outlook that the global semiconductor market could grow by more than 25% in 2026 to about $975 billion, with the memory segment rising 30%. The same company said the HBM market could reach $54.6 billion in 2026, up 58% from the prior year, and said the industry was entering a “memory supercycle” driven by AI infrastructure. That is the profit engine behind the trio’s recent strength: AI servers need more fast memory, HBM commands better pricing than legacy DRAM, and the supply chain remains tight enough to keep margins elevated.
The legal challenge is built around that same commercial reality. Plaintiffs in the California case say Samsung, SK hynix, and Micron together control roughly 90% of the relevant DRAM market and used the transition to HBM to wind down older products such as DDR3 and DDR4. The complaint says that helped push conventional DRAM prices up roughly 700% over four years. Those allegations matter because they try to turn a rational business shift into evidence of an unlawful agreement. That is a much higher bar than simply showing that the companies made the same profitable move.
What makes the story consequential is that both explanations can sound true at once. HBM is a real technological necessity for AI accelerators, and the suppliers that can meet qualification standards are few. At the same time, when only three firms dominate a critical input and all three pivot toward the same higher-margin product, the market can begin to look coordinated even if each company is acting alone. That is why the dispute is bigger than a single lawsuit. It goes to the core of whether AI demand is merely enriching the memory makers or whether the structure of the market is giving them unusual power over prices.
Micron’s own June 24 earnings prepared remarks reinforce how powerful the cycle has become. The company said it delivered an exceptional fiscal third quarter with records in revenue, gross margin, and earnings per share, and said its data-center revenue exceeded $25 billion on an annualized run rate of more than $100 billion. Micron also said the memory industry has been structurally transformed by the proliferation of AI. That language does not sound like a short-lived bump. It sounds like a business mix shift that could last long enough to re-rate the whole sector.
Samsung’s position is more complicated but still central. A 2024 report on Nvidia qualification said Samsung’s HBM3 chips had been cleared for use in Nvidia’s H20 processor for the China market, while testing for HBM3E was still underway. That mattered because qualification is one of the gates to higher-end AI memory sales. In other words, the competition is not just about capacity; it is about who can meet the technical standard fast enough to join the premium part of the market.
The result is a memory industry that is unusually attractive and unusually exposed. It is attractive because AI demand is still rising, advanced memory remains scarce, and customers need reliable supply. It is exposed because the same scarcity that boosts profits can invite antitrust scrutiny, especially when the companies involved have a history of DRAM collusion allegations. Samsung and Hynix were sanctioned in the 2000s in a separate DRAM price-fixing case, and that history will sit in the background even if it does not prove anything about the current complaint.
The central question for investors and customers is not whether memory demand is strong. It is. The question is how long the industry can keep converting that demand into pricing power before new supply, qualification success, or legal pressure changes the balance. For now, the answer is that the AI buildout still needs a relatively small group of suppliers, and that scarcity remains the most important fact in the market.
AI Memory Has Turned Scarcity Into A Profit Center
The first layer of the story is structural. AI systems consume far more memory than older computing workloads, and they need that memory to be faster and more efficient. HBM is designed for exactly that use case. It is stacked, high-bandwidth, and expensive to produce. That makes it less of a generic commodity and more of a specialized component whose availability can shape the pace of AI deployment.
That shift has consequences across the entire memory stack. If capital and production capacity move toward HBM, the older DRAM formats that still power consumer devices can become tighter. Buyers then face a double hit: they need to pay more for premium AI memory and more for the mainstream memory that remains. That is the economic foundation behind the plaintiffs’ theory of the case, even if it is not proof of coordination.
“The memory segment increasing at 30% growth.”
That line from SK hynix’s outlook captures the scale of the re-pricing. The company was not describing a narrow product cycle. It was describing a broader reallocation of value toward memory inside the semiconductor market. It also said that AI infrastructure is increasing the amount of DRAM and HBM installed per server, which helps explain why demand remains strong even after a long stretch of industry investment.
For the suppliers, the economics are straightforward. If customers need advanced memory and only a few companies can deliver it at scale, pricing power rises. If those companies can direct more wafers and packaging capacity toward the highest-margin products, their profitability improves. That does not require a cartel to be profitable. It only requires a market that is tight enough to reward discipline.
That is why this case should not be read as a simple morality tale about greedy chipmakers. The more realistic reading is that AI has created an industrial bottleneck and the memory makers are monetizing it. The legal system may eventually decide whether they crossed the line from parallel strategy into agreement. The market has already decided that the input itself is worth more.
The Lawsuit Matters Because It Tests The Boundary Between Strategy And Collusion
The California complaint is important not because it proves misconduct, but because it forces a test of the line between independent business judgment and illegal coordination. Plaintiffs say the three firms are so dominant that their combined behavior effectively controls the market. They also say the companies used the HBM transition to reduce legacy supply and raise prices. If they can show an actual agreement, the case becomes much stronger. If they cannot, the complaint may end up looking like a challenge to ordinary parallel conduct in a concentrated industry.
That distinction is crucial. Antitrust law does not punish firms for making the same profitable choice. It punishes agreements to do so together. In a three-player market, especially one where the product mix is shifting toward a more profitable segment, identical moves can be perfectly rational. The hard part is proving that the moves were coordinated rather than simply similar.
“The companies have not yet responded in court, and the allegations remain unproven.”
That caveat belongs at the center of any accurate account. The complaint is serious, but it is still only a complaint. The law will care about evidence of communication, coordination, or other facts that show a real agreement. The fact that the companies have all benefited from the same cycle is not, by itself, enough.
Still, the history is enough to keep the case alive as a market story. Samsung and Hynix have faced DRAM antitrust penalties before, and that backdrop makes the present environment easier to scrutinize. It also means the current cycle will be interpreted through a skeptical lens: if prices stay elevated and supply stays tight, the temptation to suspect coordination will remain high.
That is especially true because the AI market has made memory visible to investors in a way it rarely was before. HBM is now part of the core narrative around Nvidia, data centers, and AI capex. When a once-obscure input becomes a strategic asset, any evidence of persistent price strength or supply discipline tends to attract both capital and regulators.
What The Next Phase Depends On
The next phase of this story is less about demand than about supply execution. If SK hynix can keep its HBM lead, if Samsung can keep closing its qualification gap, and if Micron can keep converting AI demand into margin, the market can stay tight even if the legal noise gets louder. But if new capacity comes online faster than expected, the pricing environment could soften.
That is why the most important catalysts are operational rather than rhetorical. Qualification milestones, new capacity ramps, and customer agreements will matter more than public commentary. The AI memory boom is still real, but booms end when supply catches up or when customers resist the premium.
For now, the lesson is that AI has not just boosted chip demand. It has reorganized the semiconductor profit pool around the companies that control scarce memory. The lawsuit may or may not survive the legal process. The commercial re-pricing is already in place.
In the short term, the market is rewarding scarcity. In the long term, the question is whether the scarcity was simply earned — or whether a court decides it was arranged.
Explore more exclusive insights at nextfin.ai.
