A US class action alleges the three memory chip giants coordinated to shift factory capacity from commodity PC memory to the stacked memory inside AI chips, testing whether a parallel rush into a higher margin product can be sued the same as a price fixing cartel.
The DRAM lawsuit filed last week in federal court in San Jose is not really about a secret cartel. It is about whether three companies that dominate the global market for the working memory in PCs and servers can be sued under US antitrust law for doing something each of them had a perfectly ordinary business reason to do: shift their factories toward the high-bandwidth memory chips that power AI accelerators.
Seventeen individual consumers and small businesses sued Samsung Electronics, SK hynix, and Micron Technology in the Northern District of California, alleging that the three memory-chip makers, who together control roughly 90% of the global market for DRAM according to Korea Herald figures cited by ANI, "coordinately curtailed" production of older DDR3 and DDR4 chips to free up factory capacity for HBM, the stacked memory used inside AI training chips. The complaint, filed as Garciaguirre et al. v. Samsung Electronics Co., Ltd. et al., case number 3:26-cv-06345, appears on the court's case docket, on PacerMonitor, and on Law360.
That reframing is the real story. A traditional DRAM price-fixing case asks the court to find a forbidden agreement: a phone call, an email, a meeting in a hotel. This complaint, as summarized by TrendForce, Wccftech, and PC Gamer, asks the court to find something newer and less comfortable: that the three firms' parallel commercial decisions, made independently but visibly in step, can be reframed as a coordinated reallocation of capacity away from the cheap end of the market and into the expensive end.
The mechanism is the part the early coverage did not spell out. High-bandwidth memory is more profitable per wafer than conventional DRAM, and it also consumes more wafer area per chip. So when a memory maker moves a fab line from DDR4 to HBM, two things happen at once: capacity for the older, cheaper DRAM tightens, and HBM supply rises. The complaint acknowledges that the industry's overall DRAM wafer capacity is still expanding in 2026, as The PC Enthusiast notes in its summary of the case. Plaintiffs' theory is that the expansion is going into the wrong product.
That distinction matters because Section 1 of the Sherman Act, the federal antitrust statute at the center of the case, generally requires either an agreement to fix prices or some other "concerted action" that harms competition. A firm choosing to make more of a higher-margin product is, on its face, exactly the kind of decision antitrust law usually leaves alone. The plaintiffs' wager is that the choice stops looking unilateral when all three dominant firms make it at once, in a market with no real new entrants to discipline prices. As Seeking Alpha notes, the move echoes the price-fixing allegations that have shadowed DRAM for two decades. The legal hook this time is the structural parallel play, not an alleged phone call.
The defendants have a real answer, and that answer is also the reason this case will turn on expert economic testimony rather than on any document dump. Every major memory maker has publicly explained its HBM build-out for at least a year, and the pivot has been widely covered in industry research and in earnings commentary. That makes the concerted-action allegation harder to sustain, because each firm can point to its own independent business case for the reallocation. To survive a motion to dismiss, the plaintiffs will need an economist who can model what DRAM prices would have looked like if one of the three had stayed in DDR4 while the other two pivoted, and show that the gap is large enough to qualify as an antitrust injury.
The early-stage procedural terrain is not friendly to plaintiffs. Federal class actions of this kind typically face an uphill fight under the rule that an agreement under Section 1 needs more than parallel conduct that could just as easily be unilateral. Courts have repeatedly dismissed similar theories when the defendants' parallel behavior was publicly observable and individually rational. The plaintiffs here are betting that the specific combination of high market concentration, a documented capacity reallocation, and rising DRAM prices is unusual enough to clear that bar.
The stakes reach beyond DRAM itself. If the theory survives a motion to dismiss, it could give consumer plaintiffs a template for challenging any concentrated industry that chases a higher-margin segment in lockstep, whether in legacy components, cloud capacity, or other AI-adjacent supply chains. The complaint is in the earliest stage of litigation. The court has not yet shown any defendant filings on the docket. What to watch next is whether the plaintiffs' economic model survives the inevitable motion to dismiss, and whether any of the three defendants responds by publicly framing the HBM shift as a customer-driven, contractually-anchored decision rather than an internal capacity choice.