The AI Memory Gold Rush: Micron Prints Record $41.5B Quarter as SK Hynix Eyes $29.4B U.S. IPO

A Quarter That Has No Precedent

When Micron Technology filed its Q3 FY2026 10-Q with the SEC, the numbers were so large they looked like a typo. Revenue of $41.5 billion in a single quarter — a 346% increase over the same period last year. Gross margin of 85%. Net income of $28.2 billion. Diluted EPS of $24.67, compared to $1.68 in Q3 FY2025.

To put that in context: just twelve months ago, Micron's entire quarterly revenue was $9.3 billion. The company generated more net income in this single quarter — $28.2B — than many S&P 500 companies generate in an entire decade. This is not a recovery. It is a structural reset of what memory economics look like in the age of AI.

What's Driving the Numbers

The filing is explicit about the drivers. DRAM average selling prices rose approximately 260% year-over-year, with bit shipments up around 20%. NAND ASPs rose roughly 310%, with bit shipments up low-double digits. The gross margin expansion from 38% to 85% in twelve months is almost entirely attributable to pricing power — the kind that only emerges when demand structurally outpaces supply.

That demand is AI. Generative AI models require enormous volumes of High-Bandwidth Memory (HBM) — the 3D-stacked DRAM that sits alongside GPUs in AI accelerators. Each NVIDIA H200 or Blackwell chip requires multiple HBM stacks. Each data centre cluster requires thousands of chips. Hyperscalers — Microsoft, Google, Amazon, Meta — are building at a pace that has consumed virtually all available HBM capacity globally.

The breadth of the growth is striking. This is not a story about one hot segment masking weakness elsewhere. Every business unit grew over 250% year-over-year. Even automotive and embedded — historically the slowest-moving memory end-market — is accelerating as in-vehicle AI and ADAS systems demand higher-bandwidth solutions.

The Margin Story Is Even More Important

Revenue growth gets the headlines. The margin story is what matters for investors. Micron's gross margin of 85% in Q3 FY2026 is not just a record for Micron — it places the company in the same territory as software businesses. For a company that manufactures physical semiconductor wafers in billion-dollar fabs, an 85% gross margin represents extraordinary pricing power and operational leverage.

Management notes that margins improved "primarily due to increases in average selling prices and also benefited from continued strong execution and favorable mix." The favorable mix is a reference to the growing share of HBM in Micron's revenue — a product that commands significantly higher margins than conventional DRAM.

Enter SK Hynix: $29.4B U.S. IPO Targets July 10

If Micron's results confirmed the magnitude of the AI memory cycle, SK Hynix's decision to list in the United States is the market's vote on how long it lasts. The Korean chipmaker — the global leader in HBM by market share — is raising $29.4 billion in a U.S. IPO under the ticker SKHY, with the listing targeted for July 10.

SK Hynix holds an estimated 50%+ share of the global HBM market. Its HBM3E chips are the preferred memory solution inside NVIDIA's H100 and H200 platforms. It was the first memory maker to ship HBM3 at scale, and it is currently the only company with HBM4 in advanced qualification. In a market where NVIDIA alone consumed billions of dollars of HBM per quarter, being the dominant supplier is extraordinarily lucrative.

The IPO is not a fundraising exercise for a company in distress. SK Hynix is profitable and well-capitalised. The U.S. listing serves two strategic purposes: it provides direct access to U.S. capital markets at a time when AI infrastructure investors are actively seeking memory exposure, and it funds the next generation of HBM capacity — the packaging, advanced node transitions, and fab expansions required to maintain HBM leadership as demand accelerates through 2027 and beyond.

The HBM Oligopoly and What It Means for Investors

High-Bandwidth Memory is a structurally different product from conventional DRAM. It requires advanced 3D stacking technology — multiple DRAM dies bonded together with through-silicon vias — and sophisticated advanced packaging. Only three companies in the world have the technology and scale to produce HBM at volume: SK Hynix, Samsung, and Micron. No new entrant can replicate this capability within any investment horizon that matters.

This is what makes the current cycle qualitatively different from prior memory cycles. In the 2017–2018 DRAM super-cycle, high margins eventually attracted new supply. In the current cycle, the technological barriers to HBM production are so high, and the lead times for new fab capacity so long (three to five years from groundbreaking to production), that supply cannot catch up with demand in the medium term. Every major AI model training cluster being built today — and there are dozens under construction globally — requires HBM that can only come from three suppliers.

Micron is investing aggressively to maintain its position. The company's latest disclosures reference new DRAM wafer capacity in Idaho and New York under the CHIPS Act, an HBM advanced packaging facility in Singapore that broke ground in January 2025, and manufacturing modernisation in Japan focused on AI memory production. These are multi-billion-dollar, multi-year commitments that reflect genuine conviction in the duration of the cycle.

The Risk Case

No cycle is permanent, and the memory industry has a well-documented history of boom-bust dynamics. Several risks are worth monitoring.

The YMTC litigation risk in particular is worth watching. Micron discloses multiple ongoing patent infringement complaints from Yangtze Memory Technologies across multiple jurisdictions — U.S. courts, UK High Court, Unified Patent Court, and German regional courts. While none are currently material to operations, they are a reminder that the geopolitical dimension of the semiconductor war is playing out in courts as well as fabs.

The Bigger Picture

Two data points this week tell the same story about where we are in the AI infrastructure investment cycle. Micron's Q3 FY2026 10-Q shows a company generating $41.5B in quarterly revenue with 85% gross margins — numbers that would have been considered science fiction for a memory company eighteen months ago. SK Hynix's $29.4B U.S. IPO signals that the global leader in HBM believes U.S. investors will pay AI infrastructure multiples for memory exposure, and that the capital needed to sustain its advantage is best raised here.

The narrative around AI infrastructure spending has shifted. Eighteen months ago, investors debated whether AI capex would sustain. Today, the hyperscalers are in a race they cannot afford to lose, and memory is the bottleneck. The companies that make HBM — all three of them — are printing margins that have no historical precedent in the memory industry.

Whether you hold Micron (MU) or watch for SK Hynix (SKHY) post-listing, the underlying thesis is the same: the world needs more AI memory than anyone can currently build, and the companies that can build it are collecting the premium accordingly.