A class action lawsuit filed in the Northern District of California on June 25, 2026, accuses Samsung, SK Hynix, and Micron of coordinating to restrict conventional DRAM production since 2022, resulting in a roughly 700% cumulative price increase. The three companies collectively control approximately 90% of the global DRAM supply.
The complaint alleges the producers shifted wafer capacity toward High Bandwidth Memory to meet AI datacenter demand, tightening supply for standard DDR memory. Plaintiffs point to the companies’ 2005 criminal price-fixing guilty pleas as precedent and argue that none of the three moved to expand conventional output despite record margins, which they characterize as evidence of coordination rather than independent business judgment.
DRAM costs 700% more than it did in 2022. That single number is the engine behind a class action complaint filed in the Northern District of California on June 25, 2026, and it will make uncomfortable reading for the three companies that, between them, control roughly 90% of the world’s DRAM supply.
Case 3:26-cv-06345, Garciaguirre et al v. Samsung Electronics Co., Ltd. et al, names Samsung, SK Hynix, and Micron. The accusation is that since 2022 they have coordinated to restrict conventional DRAM production. Not through a back-room handshake, the plaintiffs concede, but through a collective pivot toward High Bandwidth Memory that left the standard DDR market starved and prices at record highs. Judge Noel Wise has the case. Prosecuting it is Bathaee Dunne LLP, an antitrust boutique that already secured a judgment against Google over digital advertising collusion.
The fourteen consumers and three small businesses behind the filing (PC retailers and tech service shops) are not the first to accuse this trio of bending memory prices, and that is exactly the argument. Samsung pleaded guilty in 2005 to DRAM price-fixing spanning April 1999 through June 2002, paying a $300 million criminal fine, then the second-largest antitrust penalty in US history. SK Hynix pleaded guilty in the same DOJ case to $185 million in fines. Counting Elpida, total penalties hit $731 million, and several executives went to prison. Micron walked away without a fine because it was the whistleblower who tipped off the regulators, a fact that now cuts both ways: it proves the industry has a documented capacity for coordinated supply manipulation, and it hands Micron a legally distinct seat at what is otherwise a shared defense table.
The plaintiffs openly lean on that history, calling it a “systematic and repetitive pattern of collusive behavior.” That phrasing may or may not survive a motion to dismiss. What it does accomplish is sparing the court the trouble of imagining, from scratch, an industry capable of collusion.
The mechanism in the complaint is technically precise. Around 2022, all three producers started shifting wafer capacity away from conventional DDR3 and DDR4 and toward HBM, chasing the explosion in AI datacenter demand, most of it flowing through Nvidia, where SK Hynix is the dominant HBM supplier. Roughly 25% of global DRAM wafer capacity has shifted to HBM, and that figure understates the damage, as an HBM die occupies about twice the wafer area of a standard DDR die. A fab floor that once stamped out X conventional dies now turns out roughly X/2 HBM dies, none of which ever land in the consumer DRAM market.
Total DRAM wafer capacity is projected to grow by about 14% by 2026, but conventional capacity is climbing only around 10%, so the gap for standard memory keeps widening even while the industry expands. The cumulative price increase since 2022 is the 700% headline, and Jefferies sees more pain coming: 40 to 50% quarter-over-quarter in Q3 2026, 30 to 40% in Q4, and another 40 to 45% year-over-year across all of 2027. The earliest credible window for relief is 2028, and Lenovo has gone on record calling elevated memory prices “the new normal” through at least 2030.
The defendants have an easy answer ready. HBM demand from AI infrastructure is real and gigantic; all three have announced new fab construction, and pivoting toward a hotter product is what rational firms do in a competitive market. The plaintiffs’ rebuttal is where the case lives. In a genuinely competitive commodity market, the record margins a supply squeeze produces should tempt at least one of the three large producers to crank up conventional output, undercut the other two, and grab share. None of them did. That restraint, the complaint argues, is the observable fingerprint of coordination. You do not need a smoking-gun hotel-room meeting if the behavior simply does not square with independent competitive decision-making.
This is an oligopoly-incentives theory, not a direct conspiracy claim, and antitrust courts have wrestled with exactly this kind of argument since Brooke Group and the tacit-collusion literature that followed. Whether the plaintiffs can drag enough out of discovery to survive summary judgment, meaning internal communications, capacity planning documents, and pricing strategy memos, will decide whether any of this reaches trial.
Gross margins on HBM are estimated at three to five times those of conventional DRAM, and that gap tightens the screw. The plaintiffs argue the trio chose to chase that margin instead of expanding conventional capacity not because doing both was technically impossible, but because keeping conventional supply tight propped those prices up too. Nothing about the two product lines forces an either-or; a company can build out both. The open question is whether the decision not to reflects ordinary business judgment or something more deliberate.
Apple’s recent price bumps on iPad and Mac lines show up in the complaint as a concrete downstream effect and a proximate trigger for the suit, a clean illustration of how DRAM costs reach consumers who never buy a memory module in their lives.
Certify this as a full class action, and the plaintiff pool balloons to include every consumer and business that bought a DRAM-containing product since 2022, which is to say, anyone who bought a phone, laptop, desktop, or server in the past four years. US antitrust law entitles prevailing plaintiffs to treble damages, three times the actual harm proven. With DRAM baked into hundreds of billions of dollars of electronics sold every year, the liability has no comfortable ceiling. Even a conservative overcharge estimate, tripled, runs into tens of billions.
Micron’s position is the strangest piece of the puzzle. The company can argue that its 2005 cooperation with the DOJ proves it is no habitual colluder. The plaintiffs can argue that the same cooperation gave Micron an intimate, working knowledge of how DRAM price-fixing operates, which would make any later participation more culpable, not less. How Micron’s counsel handles that history against Samsung’s and SK Hynix’s counsel will be one of the sharper strategic subplots here.
The headline figures, in one place:
| Metric | Figure |
|---|---|
| DRAM price increase since 2022 | ~700% cumulative |
| Big Three combined market share | ~90% |
| Wafer capacity shifted to HBM | ~25% |
| HBM die area vs. standard DDR die | ~2× |
| Projected Q3 2026 price increase | +40–50% QoQ (Jefferies) |
| Projected Q4 2026 price increase | +30–40% QoQ (Jefferies) |
| Projected 2027 full-year increase | +40–45% YoY (Jefferies) |
| Earliest expected price relief | 2028 |
| 2005 Samsung criminal fine | $300 million |
| 2005 SK Hynix criminal fine | $185 million |
| Total 2005 case penalties | $731 million |
Class certification, discovery, and probably years of litigation stand between this filing and any damages award. The conditions it describes are unusually substantive for a memory-industry antitrust case, more so than anything since the 2005 DOJ prosecution: a market this concentrated, a paper trail of prior collusion, and a capacity shift that serves a legitimate business purpose while producing results indistinguishable from coordinated supply restriction. The one thing worth doubting is whether the plaintiffs can ever close the gap between story and proof. The economic narrative is compelling, but that alone wins nothing. They need admissible evidence that the three of them actually coordinated, and right now, there is no public sign that they have it.