The defining bottleneck of the artificial-intelligence boom is no longer the graphics processor. It is the memory that sits next to it.

That is the core argument of a Monday note from Jordi Visser, head of AI macro nexus at 22V Research. Borrowing a line from KAIST professor Kim Jung-ho — the engineer often called the "father of HBM" — Visser frames the whole trade in three words: "AI equals memory."

His thesis is that the market keeps measuring the AI cycle by how far chip stocks have run, when it should be measuring how far the intelligence stack still has to evolve. "AI started as a compute race," Visser writes. "It is becoming a memory race." While compute created intelligence in the training era; in the agentic era, "memory gives intelligence context."

From A Brilliant Brain to A Memory Palace

Visser leans on Kim's analogy: the GPU is the brilliant analyst, but memory is the desk, the filing cabinet, the library and the courier system. A brilliant analyst with no filing cabinet and a slow courier spends the day waiting for files. Scale that to the millions of AI agents enterprises expect to run in parallel — each one needing context, state and persistence — and demand for memory does not taper off. It compounds.

Micron Numbers Feed the Narrative

The earnings data is backing the story. Micron Technology Inc. (MU) reported fiscal third-quarter 2026 revenue of $41.46 billion on June 24, more than quadruple a year earlier and well above the roughly $35.8 billion consensus, sending shares up about 15% after hours and the market capitalization past $1 trillion. Management guided fiscal fourth-quarter revenue to approximately $50 billion.

CEO Sanjay Mehrotra said memory has become "a strategic asset" in the AI era and that "AI system performance is architecturally dependent on memory subsystem performance and capacity." More important for the cyclicality debate, Micron said it has now signed 16 strategic customer agreements — multi-year deals with committed volumes and pricing terms. According to Visser, those contracts represent roughly $100 billion in cumulative revenue for Micron through 2030. Long-term supply agreements with committed pricing suggest customers increasingly view advanced memory as a strategic resource rather than a commodity component.

A Market Dominated by Few Players

Here is what makes the trade so concentrated. The number of firms that can supply AI-grade memory at scale is tiny. The bullish case is reinforced by the industry's highly concentrated structure. In DRAM, just three companies control roughly 90% of global production. According to Counterpoint Research, in the first quarter of 2026, Samsung Electronics Co. led with about a 38% revenue share, ahead of SK Hynix Inc. at roughly 29%, with Micron third.

The grip is even tighter in high-bandwidth memory — the stacked DRAM bolted beside every Nvidia Corp. (NVDA) accelerator — where Counterpoint data has put SK Hynix above 50% share and the three makers together commanding more than 95%. Goldman Sachs estimates SK Hynix has locked up about two-thirds of orders for Nvidia's next-generation HBM4. That HBM lead was strong enough to push SK Hynix past Samsung in annual operating profit for the first time in 2025.

"From the charts alone, it is easy to say this must be late. But that is the wrong measuring stick. The better question is not how far the stocks have moved. The better question is how far the intelligence stack still has to evolve," Visser said.

Industry Faces New Antitrust Challenge

The concentration that underpins the bullish investment case is also attracting legal scrutiny. A proposed federal class-action lawsuit – Garciaguirre et al. v. Samsung Electronics Co., Ltd. et al. (No. 3:26-cv-06345) – filed on June 25 in the U.S. District Court for the Northern District of California accuses Samsung Electronics, SK Hynix and Micron of conspiring to restrict supply and inflate prices in the global DRAM market. The plaintiffs—which include consumers, small businesses and repair companies—allege the three manufacturers coordinated production cuts and pricing following the pandemic, artificially driving up memory prices despite slowing demand. They are seeking damages and injunctive relief under U.S. antitrust laws.