Micron Stock Jumps 15% After Hours — What a $41.5 Billion Quarter Tells Us About the AI Memory Revolution

Introduction

At approximately 4:00 p.m. Eastern Time on Tuesday, June 24, 2026, Micron Technology filed its fiscal third-quarter earnings with the Securities and Exchange Commission. Within the hour, the stock was up 15% in after-hours trading. By the time analysts had finished processing the numbers, the market had added more than $150 billion to Micron's valuation in a single session. The company now commands a market capitalisation above $1.35 trillion, making it one of the most valuable semiconductor businesses on earth, trailing only Nvidia in that category. Twelve months ago, Micron was a cyclical memory company most people associated with price wars and overcapacity. Today, it is a critical infrastructure provider for the AI economy, and its results prove it.

Revenue for the quarter came in at $41.46 billion, against a consensus analyst forecast of approximately $35.25 billion. That is a $6.2 billion beat, or 16% above expectations, on a number that was already expected to be a record. Non-GAAP earnings per share of $25.11 beat the consensus estimate of $20.28 by 23.8%. GAAP net income for the quarter was $28.24 billion, compared with $1.89 billion in the same quarter a year ago — a 14-fold increase in a single year. The quarter marked Micron's fifth consecutive revenue record. And then the company guided the next quarter to $50 billion, against analyst expectations of approximately $42.9 billion, with gross margins of 86%.

These are not incremental improvements on a baseline. They are numbers that require a complete revision of how the memory industry is understood.

DRAM, NAND, and the Price Story Nobody Saw Coming

The composition of Micron's revenue reveals as much as the headline figure. DRAM — the category of memory used in servers, personal computers, and AI accelerators — generated a record $31.3 billion in the quarter, up 343% from the same period a year earlier and up 67% from the prior quarter alone. DRAM bit shipments grew only in the low single-digit percentage range quarter-over-quarter. The volume increase was modest. The price increase was not. Average selling prices for DRAM rose in the low-60% percentage range versus the prior quarter. That is pricing power of an order that the memory industry has not historically been able to sustain, because memory has historically been a commodity market where the inability of any single producer to differentiate its product created relentless downward pressure on prices.

What changed that dynamic is not complicated to explain, even if the implications are still being absorbed. AI accelerators require high-bandwidth memory stacked directly on the accelerator package. That memory cannot simply be swapped out for a cheaper alternative from a different supplier. It is designed, qualified, and integrated into the system over months of engineering work. The number of companies capable of producing it — Micron, SK Hynix, and Samsung — is small. The demand coming from hyperscalers spending a combined $725 billion on AI infrastructure in 2026 is enormous. The result is not a market. It is an allocation process, and Micron's HBM4 capacity has been fully booked through the end of 2026.

NAND revenue — the category used for solid-state storage — also set a record at $9.9 billion, up 361% year-over-year. NAND average selling prices rose in the mid-80% percentage range versus the prior quarter, an even more dramatic pricing move than DRAM. Micron noted over $5 billion in data centre SSD revenue specifically, reflecting demand not just for memory attached to AI accelerators but for the storage infrastructure that feeds data into those accelerators.

Gross Margins at 84.6% — What This Number Actually Means

Micron's GAAP gross margin for the quarter was 84.6%, up from 39% a year earlier. To contextualise that movement, consider that the semiconductor industry in general considers gross margins above 60% exceptional. A software company running at 84% gross margins would be considered outstanding. For a manufacturer of physical chips — a business that requires billions in factory investment, yields that must be managed across complex processes, and raw material costs that cannot be fully controlled — reaching 84.6% gross margins represents something that requires a new vocabulary to describe.

The drivers are the same as the revenue beat: tight supply, favourable product mix, and the structural shift toward HBM4 and premium NAND products that carry higher margins than commodity memory. Micron's Cloud Memory Business Unit generated 83% gross margins. Its Core Data Centre Business Unit generated 87% gross margins. Even the Mobile and Client Business Unit, which historically runs at significantly lower margins than data centre products, achieved 87% gross margins in the quarter.

The operating margin for the quarter was 80.4% on a GAAP basis, producing operating income of $33.32 billion. Operating cash flow was $25.39 billion, representing 61% of revenue. Free cash flow after net capital expenditures of $7.1 billion was $18.3 billion — a quarterly record. The company ended the quarter with $30.2 billion in cash and investments and a net cash position of $24.4 billion. These are not the financial characteristics of a cyclical commodity supplier. They are the financial characteristics of a platform business with pricing power and structural supply constraints working in its favour simultaneously.

The Strategic Customer Agreements That Change the Business Model

Buried inside the earnings presentation, and receiving less attention than the headline numbers, was arguably the most significant long-term development in Micron's history: the announcement of 16 Strategic Customer Agreements, or SCAs, with a total value of $22 billion in cash and financial commitments, including approximately $18 billion in cash deposits.

These agreements are structured as take-or-pay contracts with binding commitments to purchase specific volumes over multi-year terms, typically spanning calendar years 2026 through 2030. They cover approximately 20% of Micron's DRAM volume and one-third of its NAND volume over the contract period. The customers are not publicly named, but the categories are not difficult to infer: hyperscale cloud providers and AI infrastructure companies that need guaranteed memory supply over a planning horizon that aligns with their own capital expenditure cycles.

The significance of these agreements is structural. For most of its history, Micron's business has been characterised by the volatility that comes from selling into a spot market where prices can fall by half in a single quarter and then take years to recover. Take-or-pay contracts with $22 billion in committed cash remove that exposure from a significant portion of the company's volume. CEO Sanjay Mehrotra described the SCAs as expected to "significantly enhance the durability and predictability of financial performance" — language that signals not just confidence in near-term demand but a fundamental change in how the company's revenue should be modelled by investors.

When a memory company secures multi-year take-or-pay agreements with hyperscalers, it is no longer a cyclical commodity supplier being valued on a trough-to-peak earnings multiple. It is an infrastructure provider with contracted revenue, and that is a categorically different business with categorically different valuation multiples.

Q4 Guidance: $50 Billion and 86% Gross Margins

If the Q3 results were the event that moved the stock 15% after hours, the Q4 guidance was the announcement that made analysts revise their models entirely. Micron guided Q4 fiscal 2026 revenue to $50 billion, plus or minus $1 billion, against analyst expectations of approximately $42.9 billion. The guidance midpoint represents a 20.6% beat versus consensus before the quarter has even begun.

The Q4 gross margin guidance of approximately 86% implies further expansion from the 84.6% reported in Q3 — a continuation of the same dynamic that has driven every quarter this fiscal year. GAAP EPS guidance of $30.73, plus or minus $1.00, implies GAAP net income in Q4 of approximately $35 billion, which would again represent a quarterly record by a wide margin. Management also guided Q4 free cash flow to exceed $30 billion, which would set another record.

The sequential revenue jump from $41.5 billion in Q3 to $50 billion in Q4 — a $8.5 billion increase in a single quarter — would, if achieved, represent the largest absolute quarterly revenue increase in Micron's history, in a year that has already produced the two largest absolute quarterly revenue increases in the company's history. The $17.6 billion sequential increase from Q2 to Q3 was described by Micron's management as the largest in its history, eclipsing the prior Q1-to-Q2 increase of $10.2 billion, which was itself described as a record at the time. The numbers are not stabilising. They are accelerating.

What Memory's Position Reveals About the AI Economy

The broader conclusion that Micron's Q3 results force is one that the industry has been slow to fully internalise: in the AI economy, the critical bottleneck is not the model, and it is not the compute chip. It is the memory.

Nvidia's Vera Rubin VR200 NVL72 rack, which will ship later this year and represents the most advanced AI training and inference hardware in commercial production, costs approximately $7.8 million per unit according to Morgan Stanley estimates. Memory and HBM components account for approximately $2 million of that total, or roughly 25% of the entire rack cost. In the previous Blackwell generation, memory represented closer to 5% of rack cost. The 5x increase in memory's share of system cost over a single product generation reflects both the technical necessity of higher-bandwidth memory for more capable models and the pricing power that a three-supplier market with insufficient production capacity can command.

Micron's data centre revenue exceeded $25 billion in the quarter, representing an annualised run rate above $100 billion from a single business segment. That number sits inside a total AI infrastructure market projected by Goldman Sachs to reach $7.6 trillion in cumulative investment between 2026 and 2031. Memory's share of that investment is not 5% or 10%. At current pricing dynamics and current product mix trajectories, it is approaching 25% at the system level. The companies that control memory supply — Micron, SK Hynix, and Samsung — collectively control a quarter of the most critical infrastructure investment cycle in modern economic history.

As Micron's CEO Sanjay Mehrotra said in the Q3 earnings press release: "Micron's record fiscal Q3 financial results and even stronger outlook for Q4 reflect the strategic value of memory in the AI era." That sentence is accurate, but it understates the case. The results do not merely reflect memory's strategic value. They quantify it: 84.6% gross margins, 80.4% operating margins, $28.24 billion in GAAP net income in a single quarter, and a company that has seen its annual earnings per share forecast rise from $7.68 in fiscal 2025 to a current full-year fiscal 2026 projection of approximately $57.71 — a 651% increase in one year.

Conclusion

Micron's fiscal third quarter of 2026 is a document about the AI economy more than it is a document about a semiconductor company. The $41.5 billion in revenue, the 84.6% gross margins, the 346% year-over-year growth, the $50 billion Q4 guidance, and the 15% after-hours stock move collectively tell a single story: the race to build AI infrastructure has created a structural supply shortage in the one component that every accelerator, every server, and every AI deployment on earth requires, and the companies that manufacture that component are now extracting the pricing power that structural scarcity always eventually produces.

For investors, the question heading into Q4 is whether $50 billion in revenue and 86% gross margins represent the peak of a demand cycle or the establishment of a new floor. Micron's management, the take-or-pay agreements with $22 billion in committed cash, and the company's statement that supply is expected to remain tight beyond 2027 collectively argue for the latter. For the broader technology industry, the question is more fundamental: when memory accounts for 25% of the cost of the most advanced AI rack on earth and a single memory company generates $28 billion in net income in a quarter, the narrative that AI is primarily a software story has become difficult to sustain. The physical bottleneck is real, its pricing power is enormous, and the results published on June 24, 2026, are the clearest statement yet of what that means in dollar terms.