Micron, Samsung Face Antitrust Suit Over AI Memory Supply

Garciaguirre v. Samsung Electronics lands at an awkward point for any clean story about AI memory supply. Filed June 25, 2026 in the Northern District of California, Case No. 3:26-cv-06345, and assigned to Judge Noel Wise, the proposed class action accuses Samsung, SK Hynix, and Micron of using the industry’s shift toward high-bandwidth memory, or HBM, as cover for restricting commodity DRAM output.[1] That is a serious allegation, but it arrives with a serious constraint: in 2022, the Ninth Circuit rejected a similar DRAM theory against the same three companies as conduct “more likely explained by lawful, unchoreographed free-market behavior.”[2]

That collision is the legal center of the case. AI demand has made HBM strategically valuable, and memory manufacturers have an obvious business reason to shift scarce capacity toward a higher-margin product. The complaint’s task is to make that plausible business pivot look like something more: a coordinated restriction of DDR3 and DDR4 supply that allegedly pushed DRAM prices up roughly 700% over four years.[2] For anyone tracking the legal implications of tight memory supply for AI chips, the question is not whether the market became painful for buyers. It is whether the pleaded facts tend to exclude independent action.

The market structure matters, but it is not the violation

The three defendants are not fringe participants reacting from the edge of the market. TrendForce’s Q1 2026 DRAM revenue figures put Samsung at 38.5%, SK Hynix at 28.8%, and Micron at 22.4%, for a combined share just under 90%.[2] In antitrust pleadings, that kind of concentration gives a complaint oxygen. It can support opportunity, interdependence, and motive. It does not, by itself, plead agreement.

That distinction is where many popular accounts of alleged semiconductor collusion become too loose. A concentrated memory market can produce similar conduct without a cartel. If the same demand shock hits the same three producers, and if all three see HBM as more profitable than legacy DRAM, parallel capacity decisions may be exactly what one would expect from independent firms. Section 1 of the Sherman Act requires an agreement; oligopoly plus high prices is not enough.

The complaint therefore leans on plus factors. The publicly available summaries identify three that matter most: simultaneous October 2022 counter-cyclical production cuts, Micron’s alleged shutdown of its Crucial consumer brand “at the most profitable price point in its history,” and alleged customer-vetting and order-policing practices beginning in January 2026.[2] Those are the allegations doing the work. The broader AI shortage narrative is only the setting.

Why the HBM pivot is the complaint’s main legal move

HBM is the fact that makes Garciaguirre different from the prior DRAM litigation, at least in theory. Plaintiffs are not merely saying that Samsung, SK Hynix, and Micron reduced output in a down cycle and later enjoyed higher prices. They are trying to plead a mechanism: as AI accelerators demanded more HBM, the manufacturers allegedly shifted production away from commodity DRAM in a way that tightened DDR3 and DDR4 supply while preserving the appearance of ordinary product allocation.[2][3]

That theory has a certain commercial plausibility even before it becomes legally plausible. DRAM fabrication capacity is not infinitely flexible. HBM consumes advanced packaging and production resources, and buyers of AI systems have been willing to pay for performance. A manufacturer choosing HBM over low-margin commodity DRAM does not need a conspiratorial explanation.

The plaintiffs’ sharper point is that the HBM transition allegedly gave the defendants a common story under which commodity DRAM scarcity could be created or maintained. If a company says it is reallocating capacity to serve AI demand, the immediate business explanation sounds rational. If all three dominant firms do it in ways that are simultaneous, counter-cyclical, and reinforced by customer restrictions, plaintiffs will argue the conduct starts to look less like coincident optimization and more like coordinated output discipline.

That is still an allegation, not an established fact. As of July 21, 2026, the full complaint text was not publicly available through court records, and the most specific details of the theory are coming from press and case-analysis summaries rather than a complaint document that can be tested line by line.[1][2] That limits how confidently any outside reader can evaluate particularity, market definitions, class allegations, and the exact documents or statements plaintiffs say support inference of agreement.

The October 2022 cuts and Micron’s Crucial allegation

The October 2022 production-cut allegation is important because timing often separates a weak parallel-conduct complaint from a more difficult one. The summaries describe simultaneous counter-cyclical cuts by the three DRAM producers at a point when, plaintiffs say, the market incentives did not justify uniform output restraint.[2] In pleading terms, “counter-cyclical” is meant to do more than describe timing. It is meant to suggest conduct against independent self-interest unless coordinated.

That is not an automatic winner. Courts are cautious about calling parallel output reductions irrational simply because prices later rose. Semiconductor markets are cyclical, capital intensive, and exposed to inventory swings. A defendant can argue that cutting production in response to perceived oversupply or margin pressure is ordinary discipline, especially where the same macro signals are visible to all market participants.

Micron’s alleged conduct around Crucial is more specific and therefore more interesting. According to available summaries, the complaint alleges that Micron shut down its Crucial consumer brand “at the most profitable price point in its history.”[2] If accurately pled and supported, plaintiffs will use that as a self-interest point: why would a firm exit or restrict a consumer-facing channel at a historically profitable moment unless it was protecting a broader supply-restriction strategy?

The defense answer is visible from the same facts. A business can close, reduce, or reposition a channel for reasons that have little to do with collusion: allocation of scarce wafers, packaging constraints, channel conflict, enterprise prioritization, or a decision to serve higher-value AI and data-center customers. The legal question will be whether plaintiffs plead facts making the conspiratorial explanation more plausible than those independent alternatives.

Customer vetting is the allegation to watch

The January 2026 customer-vetting and order-policing allegations may become the most consequential part of the case if the complaint pleads them with precision. Public summaries describe allegations that the defendants began vetting customers and policing orders in ways that restricted access to DRAM supply.[2] That is a different kind of fact from a capacity shift. It potentially reaches into how scarcity was administered.

A capacity-allocation story can often be told as internal business judgment. Customer-vetting allegations can suggest implementation: who gets product, who is delayed, what orders are rejected, and whether firms appear to be enforcing a shared discipline rather than making independent sales decisions. Plaintiffs will want those facts to show coordination at the customer interface, not merely similar production incentives upstream.

But the same caution applies. In a real shortage, manufacturers often vet customers to prevent hoarding, gray-market resale, duplicate ordering, or contractual mismatches. If HBM and commodity DRAM capacity are tight, a producer may rationally allocate supply to strategic customers. The allegation becomes legally potent only if plaintiffs connect the vetting and policing to interfirm agreement rather than ordinary rationing under scarcity.

The 2022 Ninth Circuit dismissal is not just background

The prior Ninth Circuit ruling is the hurdle every new paragraph of Garciaguirre has to clear. In 2022, the court dismissed earlier DRAM claims against Samsung, SK Hynix, and Micron, concluding that the alleged parallel conduct was “more likely explained by lawful, unchoreographed free-market behavior.”[2] That language is not a casual observation. It tells defendants exactly how to frame a motion to dismiss.

Expect the defense position to be straightforward: AI demand is genuine; HBM is economically attractive; production resources are limited; and sophisticated firms in the same concentrated industry can independently reach the same conclusion at roughly the same time. Under that account, similar output cuts, HBM prioritization, and customer triage are the predictable consequences of market structure, not evidence of a meeting of the minds.

That is why the new complaint cannot survive on atmosphere. It needs factual enhancement. The HBM pivot is plaintiffs’ candidate for that role, but it must do more than rename the market cycle. To matter under the pleading standard, it has to help explain why the alleged conduct is less consistent with independent profit maximization and more consistent with coordinated supply restriction.

There is also a practical appellate shadow. Even if a district court is willing to let discovery proceed, the defendants will likely insist that the Ninth Circuit has already warned against converting interdependent oligopoly behavior into a Sherman Act case without particularized plus factors. The plaintiffs’ best answer is that AI-era HBM demand changed the economics and provided a new cover mechanism that the prior litigation did not plead.

The old DRAM cartel explains suspicion, not liability

The names of the defendants carry old antitrust memory. Samsung, Hynix, and Micron were involved in the 1998–2002 DRAM cartel matter, which produced more than $731 million in criminal penalties and prison sentences for executives; Samsung paid $300 million, Hynix paid $185 million, and Micron avoided fines by reporting under the leniency program.[2]

Plaintiffs understandably invoke that history. It gives the market a prior-conduct context, and it rebuts any suggestion that DRAM coordination is commercially unimaginable. For readers who remember the earlier cartel, the current allegations are not floating in a vacuum.

But prior misconduct cannot substitute for a current agreement. A two-decade-old cartel may help explain why plaintiffs scrutinize parallel behavior in this market. It does not prove that an AI-era HBM shift was coordinated. Courts are likely to treat the history as context at most unless the complaint ties it to present conduct through current communications, mechanisms, or other particularized facts.

Why the case is drawing attention outside antitrust circles

The suit has drawn broader technology coverage because memory scarcity is now connected to AI infrastructure, consumer PC pricing, and graphics-card supply frustration. Reports have framed the defendants as dominant DRAM “oligopolists” and have highlighted the complaint’s allegation that the AI memory boom masked an intentional commodity DRAM squeeze.[4][5] That attention is understandable, but it can blur the issue.

A buyer experiencing a severe price increase does not need an antitrust violation to feel harmed. The alleged 700% increase, if accurately characterized, is commercially dramatic.[2] It may affect downstream device makers, component resellers, cloud infrastructure buyers, and consumers. Still, antitrust injury and pleading sufficiency depend on the source of that increase. Scarcity produced by extraordinary AI demand is different from scarcity maintained by agreement among dominant suppliers.

Some commentary has asked whether the AI memory boom reflects a real shortage or a manufactured one.[6] Garciaguirre puts that question into a legal form, but the court will not be deciding market morality in the abstract. It will be deciding whether the complaint pleads enough factual content to make unlawful coordination plausible.

Procedurally, this is still an opening-stage private case

The docket posture is modest. Early activity includes pro hac vice motions, waiver-of-service returns, and stipulations for extensions of time through July 16, 2026.[1] No U.S., EU, South Korean, or Chinese regulator has been identified as having opened a parallel investigation into current DRAM pricing. For now, the risk signal is private class action litigation, not a coordinated public-enforcement campaign.

That matters for legal operations teams and in-house counsel assessing supply-chain exposure. The case is not a finding that memory manufacturers colluded. It is an early pleading-stage test of whether plaintiffs can use AI-era capacity allocation, HBM economics, alleged customer policing, and historical DRAM conduct to get past the same free-market explanation that defeated earlier claims.

What can be inferred now

Garciaguirre is best read as a bellwether, not as a verdict on the AI memory market. It tests whether the HBM-production pivot supplies the additional factual enhancement that prior DRAM litigation lacked. The strongest version of the plaintiffs’ theory is not that AI demand was fake. It is that genuine AI demand created a useful cover for coordinated commodity DRAM restriction.

The strongest defense version is equally coherent: AI demand changed the economics of memory production, HBM became a rational priority, and independent firms facing the same incentives made similar decisions. The Ninth Circuit has already shown receptivity to that kind of explanation in this market. Unless the new complaint pleads a more concrete mechanism of agreement, concentration and price movement will not do the job alone.

The case’s viability therefore turns on the details still least visible from the public record: how specifically the complaint pleads the October 2022 cuts, what facts support the Crucial allegation, what the January 2026 customer-vetting practices actually were, and whether those facts make coordinated supply restriction more plausible than ordinary, painful economics in an AI-constrained memory market.

References

1. Garciaguirre v. Samsung Electronics, Case No. 3:26-cv-06345, U.S. District Court for the Northern District of California, June 25, 2026, https://www.cand.uscourts.gov
2. DRAM Price-Fixing Lawsuit, Rain Intelligence, https://rainintelligence.com/blog/dram-price-fixing-lawsuit
3. Tom’s Hardware summary of the coordinated HBM-shift allegation, Tom’s Hardware, https://www.tomshardware.com
4. PC Gamer coverage of the “oligopolists” characterization, PC Gamer, https://www.pcgamer.com
5. Quartz reporting on the class action filing, Quartz, https://qz.com
6. Datafloq piece questioning whether the AI memory boom is a real shortage or a manufactured story, Datafloq, https://datafloq.com