Anyone who tried to buy computer memory in 2026 has felt it. The price of RAM has roughly doubled this year, and in some categories it has climbed far more. The accepted explanation is the artificial-intelligence boom, which has pulled the memory industry's capacity toward the specialized chips that feed AI data centers.

On June 25, 2026, a class action in the Northern District of California offered a sharper account: that the shortage is not only a side effect of AI demand, but in significant part a coordinated scheme by the three companies that control almost the entire market.

The case, Garciaguirre v. Samsung Electronics, names Samsung, SK Hynix, and Micron, the three firms that together account for roughly ninety percent of global DRAM revenue, and brings claims under Section 1 of the Sherman Act alongside several state antitrust laws.

Filed by the antitrust boutique Bathaee Dunne on behalf of consumers and small businesses, it is an indirect-purchaser action, brought by the people and companies at the end of the supply chain who ultimately pay for the memory inside the devices they buy.

The allegations are unproven, and the defendants have not yet responded.

The allegation

The core claim is direct. Since October 2022, the complaint alleges, the three DRAM makers coordinated to restrict supply and raise prices for conventional memory, the standardized DDR4, DDR5, and LPDDR modules built to industry specification that sit inside ordinary phones, laptops, and servers.

Over four years, the complaint says, the price of that commodity memory rose roughly seven hundred percent, an increase the plaintiffs argue makes no economic sense unless the three firms were acting in concert rather than competing.

They point to a moment in October 2022 when one of the three announced production cuts even as prices were already falling, the kind of move that hurts a company acting alone but pays off if everyone cuts together.

The mechanism: AI memory as cover

What makes the case distinctly a product of this moment is its theory of how the alleged restriction worked. The memory industry has spent the past two years racing to build high-bandwidth memory, or HBM, the stacked, premium chips that AI accelerators consume in enormous quantities.

That shift is real and well documented: by some estimates the three makers have redirected the large majority of their combined capacity toward HBM, which sells for many times the price of conventional memory.

A single HBM module can fetch sixty to a hundred dollars where a comparable amount of standard DDR5 runs a few dollars.

The complaint takes that undisputed industry trend and casts it as the instrument of the conspiracy.

Because HBM consumes several times the wafer area per bit, every wafer moved to HBM withdraws a disproportionate share of conventional supply, so even a modest, coordinated pivot can starve the commodity market while each firm points to AI demand as the reason.

Plaintiffs allege the three used the HBM transition, together with a coordinated exit from older DDR3 and DDR4 production, to tighten conventional DRAM supply and hold prices up.

The same demand surge that makes the shortage look inevitable is, in the complaint's telling, what makes the alleged coordination so hard to see.

The facts the plaintiffs lean on

Antitrust law does not punish a concentrated industry for moving in the same direction; parallel conduct alone is not a conspiracy.

Plaintiffs have to plead facts tending to exclude the possibility that each firm simply acted in its own rational self-interest, and the complaint is built to supply them.

Beyond the irrational-seeming 2022 cuts, it points to Micron's decision in December 2025 to shut down Crucial, its direct-to-consumer memory brand, at what the complaint calls the most profitable price point in the brand's history, an odd time to abandon a profitable channel unless the goal was to remove supply.

Its most pointed allegation concerns conduct in January 2026, when, according to reporting the complaint cites, all three makers simultaneously began vetting customers and policing orders, questioning buyers about who they were reselling to, how much they wanted, and whether their stated demand was real.

The complaint quotes an industry executive describing identical behavior across the market: “the three companies became stricter and asked us about who we will supply to, how much quantity, and if the demand of our customers are real.”

That phrase, three companies behaving as one, is the kind of detail antitrust plaintiffs build cases around.

A market with a history

The complaint's most powerful rhetorical asset is that this has happened before, with the same defendants. Between 1998 and 2002, Samsung, Hynix (now SK Hynix), Micron, and others ran a DRAM price-fixing cartel that the Department of Justice prosecuted. Samsung paid a $300 million criminal fine and Hynix $185 million, total criminal penalties exceeded $730 million, and several executives served prison time.

Micron avoided a fine by reporting the conspiracy under the leniency program. A parallel civil case brought by consumers and state attorneys general later settled for $310 million.

That history cuts two ways for the current case. It gives the plaintiffs a ready narrative of repeat offenders and a roadmap of how the prior conduct was proven, and it means courts and juries will not be hearing the idea of a memory cartel for the first time.

It also means the defendants know exactly how these cases are litigated and what a leniency posture is worth, which is part of why the present complaint reads as written in anticipation of a long fight.

Why it is hard to prove, and dangerous to face

The defense almost writes itself, because the lawful explanation is sitting in plain view. AI demand for memory is genuine and extraordinary, HBM is far more profitable per chip, and prioritizing it is the obvious independent business decision for any rational maker, whether or not the others do the same.

In a three-firm market, everyone watching everyone else and reaching the same conclusion is exactly what you would expect from a real supply shock.

Under the governing pleading standard, the plaintiffs must offer more than that shared logic; they must show something that points to an actual agreement. The coordinated order-policing, the counter-cyclical cuts, and the recidivism are aimed squarely at that bar.

The exposure, if the case advances, is large in a way few products can match.

DRAM sits inside nearly every electronic device sold, so the class of indirect purchasers is close to the entire consuming public, and the state antitrust statutes the complaint invokes allow trebled damages.

That combination, a near-universal class and treble damages against firms with a documented cartel past, is what turns a difficult case into one that commands attention regardless of how the early motions come out.

A Market Built for Suspicion

Strip the case to its frame and it sits at an uncomfortable intersection.

A market with only three meaningful sellers, a documented history of criminal collusion, a genuine and enormous demand shock, and a price increase measured in multiples is the precise setting in which lawful parallel behavior and unlawful coordination are hardest to tell apart.

That ambiguity is not a flaw in the complaint; it is the whole battlefield.

The plaintiffs will argue the pattern is too neat to be coincidence, the defendants will argue the AI boom explains everything, and the case will be decided on which facts a court believes tend to exclude the innocent story.

However it resolves, it marks the moment the memory shortage stopped being only an engineering and economics story and became a legal one.

We saw this one coming.

Rain Intelligence flagged the DRAM price-fixing exposure facing Samsung, SK Hynix, and Micron roughly three months before this complaint was filed, while the memory shortage was still being reported as a pure supply story.

That is the work: forecasting where litigation lands next and routing each opportunity to the attorney best positioned to win it, so your firm reaches the client first.

Book a demo to see it matched to your practice, clients, and jurisdictions.