Lying on the "Money-Printing Machine," Micron Rakes in $24 Billion in a Single Quarter

Ask AI · How Will HBM4 Mass Production Reshape the Storage Market?

Image may be AI-generated: Micron’s “Silicon-Based Money Printer.” Image generated by AI

Text | Su Yang

Editor | Xu Qingyang

On March 18 (local time), after market close, U.S. semiconductor company Micron announced its Q2 FY2026 results (covering December 2025 to February 2026), ending February 26, 2026.

Data shows that Micron’s revenue for the quarter reached $23.86 billion, nearly doubling year-over-year, with net profit soaring from $1.58 billion last year to $13.79 billion. Both figures significantly exceeded analyst expectations. Previously, LSEG surveyed analysts predicted an average revenue of $20.07 billion and EPS of $9.31, but Micron delivered an adjusted EPS of $12.20.

Micron’s Key Financial Data

What shocked the market even more was the outlook for the third fiscal quarter.

The company expects revenue to reach around $33.5 billion, plus or minus $750 million. Regarding adjusted EPS, Micron forecasts $19.15, plus or minus $0.40, with all data well above expectations.

Micron’s Q3 Outlook

However, after the earnings release, Micron’s stock fell about 2% in after-hours trading.

Behind this unusual decline is Micron’s simultaneous disclosure of aggressive expansion plans: Capital expenditures for FY2026 will exceed $25 billion, up from the previous forecast of $20 billion, with analysts estimating $22.4 billion beforehand.

The after-hours stock reaction highlights investor concerns about the cyclical “expansion-price reduction” pattern in the chip industry, along with unmentioned factors like buybacks and macroeconomic sentiment shifts.

It’s worth noting that Micron’s stock has doubled in 2025, up 63% so far this year. Among the top 10 U.S. tech companies by market cap, Micron is the only one to rise this year; Oracle has fallen 22%, while Microsoft and Tesla have experienced double-digit declines.

Contrasting market worries, Micron’s board clearly remains confident about the future.

On March 18, the company announced a 30% increase in quarterly dividends, from 11.5 cents per share to 15 cents. This move is particularly notable amid a significant rise in capital expenditures, reflecting management’s confidence in sustained cash flow.

“Driven by strong demand, tight industry supply, and our robust execution, Micron set new records in revenue, gross margin, EPS, and free cash flow in Q2,” said Sanjay Mehrotra, President and CEO of Micron, in the earnings report.

01 A “Flawless” Performance

From a financial perspective, Micron’s second quarter performance can almost be described as “flawless.”

Revenue of $23.86 billion not only far exceeds last year’s $8.05 billion but also significantly surpasses the previous quarter’s $13.64 billion.

Net profit reached $13.79 billion, with diluted EPS of $12.07; on a non-GAAP basis, net profit was $14.02 billion, EPS $12.20. Operating cash flow hit $11.9 billion, compared to $3.94 billion last year and $8.41 billion last quarter.

More notably, gross margin jumped from 36.8% a year ago to 74.4%, doubling within a year—a rare feat in the chip industry.

On a non-GAAP basis, gross margin reached 74.9%. In comparison, last quarter’s non-GAAP gross margin was 56.8%. The continuous rise in gross margin directly reflects the profitability boost from AI-related memory products.

Operating expenses were $1.62 billion, growing slower than revenue, with non-GAAP operating expenses at $1.42 billion. Operating profit margin reached 67.6%, with non-GAAP at 69%. A year ago, these figures were 22% and 24.9%, respectively.

As of the end of Q2, Micron held a total of $16.7 billion in cash, marketable investments, and restricted cash, ensuring ample funds for large-scale capacity expansion. Capital expenditures for Q2 netted $5 billion, with an adjusted free cash flow of $6.9 billion.

It’s important to note that Micron did not disclose a share buyback plan in this earnings report. This likely indicates that, at this stage, cash is prioritized for capacity expansion over buybacks.

02 Storage Price Increases, Micron Benefits

Looking at business segments, Micron’s growth spans nearly all areas.

Cloud computing revenue reached $7.749 billion, up over 160% YoY. Gross margin for this segment is 74%, with an operating profit margin of 66%, both significantly higher than 66% and 55% in the previous quarter.

Core data center segment revenue was $5.687 billion, up from $2.379 billion last quarter and $1.83 billion a year ago; gross margin is 74%, operating profit margin 67%.

Mobile and client segment revenue was $7.711 billion, compared to just $2.236 billion last year and $4.255 billion last quarter. Its gross margin is as high as 79%, with an operating profit margin of 76%. A year ago, gross margin was only 15%, and operating profit just 1%. This indicates that the AI-driven memory upgrade wave has penetrated consumer electronics, and Micron is enjoying the benefits of storage price hikes.

Automotive and embedded segments generated $2.708 billion, up from $1.720 billion last quarter and $1.034 billion a year ago, with growth potential gradually unlocking.

Micron stated in its earnings report: “As AI develops, we expect computing architectures to become more memory-intensive. We believe Micron is one of the biggest beneficiaries and drivers of AI.”

Mehrotra revealed more specific progress during the earnings call. HBM4 chips customized for NVIDIA’s Vera Rubin platform have entered mass production in the first quarter, and next-generation HBM4e products will be launched in 2027, alleviating previous market concerns about Micron’s ability to supply HBM4.

Meanwhile, the business model of the memory industry is also changing.

Historically, memory was a typical cyclical commodity with short contract cycles and volatile prices. Now, customers are seeking long-term supply guarantees. Micron mentioned that in recent months, memory companies have signed more long-term contracts to secure future capacity. This shift suggests that the performance stability of memory manufacturers may improve, and cyclical volatility could weaken.

03 New Capacity Expected to Materialize by 2027

Facing a market with persistent memory chip shortages, Micron has also launched expansion plans.

The company disclosed that capital expenditures for FY2026 will exceed $25 billion, meaning substantial investments in equipment and facilities are planned in the remaining months until August.

Mehrotra said during the earnings call that capital spending in FY2027 will “increase significantly,” with construction costs alone expected to rise over $10 billion compared to a year earlier.

Such massive investments aim to address ongoing supply shortages over the coming years.

Micron is building two large wafer fabrication plants in Idaho and New York to boost U.S. memory manufacturing capacity. Mehrotra revealed that the Idaho plant is expected to begin initial production by mid-2027. In January, Micron broke ground on a $100 billion mega-facility in New York, which will include up to four wafer fabs, with wafers expected to be produced in late 2028.

He emphasized, “In the AI era, memory has become a strategic asset for our customers. We are investing in global manufacturing bases to support their growing needs.”

Industry forecasts underscore the necessity of this capacity expansion.

UBS analyst Srini Pajjuri predicts that storage chip shortages will persist until 2027 or even extend into 2028. He also noted that any potential demand slowdown in PC and smartphone markets “will be largely offset by AI/data center demand.”

04 Major Orders Are Key

The global HBM (High Bandwidth Memory) market is currently dominated by three companies: Micron, Samsung Electronics, and SK Hynix. All are expanding capacity, but supply growth still lags behind the explosive AI demand.

For Micron, besides HBM4 already in mass shipment, the company is planning next-generation products.

NVIDIA has announced that its upcoming Feynman GPU in 2028 will use custom HBM. This makes Micron’s partnership with NVIDIA critical in the coming years. Any decision by NVIDIA to reduce purchases from Micron could significantly impact the company.

From a fundamental perspective, AI demand for memory seems just beginning to release.

A single NVIDIA Vera Rubin NVL72 system uses about three times the DRAM of the Grace Blackwell GB300 NVL72 rack. A single Rubin Ultra GPU will have 1TB of high-performance HBM4e memory—more than triple that of a single Rubin GPU.

NVIDIA CEO Jensen Huang estimated at GTC on March 17 that by 2027, the total procurement orders for Blackwell and Vera Rubin GPUs will reach $1 trillion.

Micron already indicated in December last year that all high-bandwidth memory for 2026 was sold out, which was reflected in its record-breaking Q2 results.

Three months ago, Wall Street called Micron’s performance “one of the biggest surprises in U.S. chip history,” and this surprise may continue for some time.

Summary of Micron’s Q2 Earnings Call:

Mehrotra began by stating that the strong quarterly results were mainly driven by AI-driven memory demand, structural supply constraints, and the company’s strong execution. He emphasized that memory is becoming a strategic asset in the AI era and has signed its first five-year strategic customer agreement.

On the technical front, the yield of 1Y DRAM nodes is reaching a record pace, expected to become the main production in mid-2026. The 9th-generation 3D NAND will achieve this goal simultaneously, and the company plans to introduce advanced EUV lithography at the 1b node.

Due to supply constraints, PC and smartphone shipments may decline by low double digits.

Product updates include mass shipment of HBM4 12-layer products compatible with NVIDIA’s next-gen platform, samples of 16-layer 48GB products, and plans for HBM4E mass production in FY2027.

To meet growing demand, Micron is accelerating global capacity expansion, with the acquisition of Powerchip Semiconductor’s Cuong factory completed, and a new NAND wafer fab in Singapore underway, expected to start production by 2028.

In capital expenditure, Micron expects FY2026 to exceed $25 billion, with a further significant increase in FY2027, including over $10 billion more for construction to support long-term investments in HBM and DRAM.

Q&A Highlights:

Q: The 81% gross margin guidance is impressive. How sustainable is this margin, especially with HBM4 added? What about in Q4 and beyond?

CFO Murphy: We provided a strong sequential guidance with a 600 basis point increase, but per our usual disclosure policy, we do not give specific gross margin guidance for Q4.

It’s important to note that supply tightness will persist beyond 2026. The gross margin after Q4 essentially reflects the long-term investment cycle benefits driven by AI, which is just beginning.

Meanwhile, the supply-side constraints we repeatedly mention will continue to play a role.

This 81% gross margin guidance fully accounts for the structural impact of gradually ramping HBM4 products. At current levels, relying solely on price increases sequentially will have diminishing marginal effects on overall gross margin.

Q: How does the five-year strategic customer agreement differ from traditional long-term contracts? Is it a multi-year volume and price commitment or a yearly renegotiation? Are there cancellation clauses if the cycle slows?

CEO Mehrotra: Strategic customer agreements are fundamentally different from typical annual contracts; they are true multi-year frameworks.

In the current extremely tight supply environment, customers’ long-term planning needs are more urgent than ever. They seek not only supply assurance but also clear visibility into capacity allocation over the coming years. This is the core driver behind such structural agreements.

Q: How do you view terminal market allocations? Are you worried about demand destruction in PCs and smartphones? How do you balance large and small customers?

CEO Mehrotra: All terminal markets are experiencing severe supply tightness, and demand remains strong across sectors. Even in price-sensitive consumer markets, while price hikes may suppress some demand, overall momentum remains.

Our strategy is to be a diversified supplier across all terminal markets.

From an industry perspective, data centers are becoming an increasingly large part of the TAM, and a bigger share of supply is flowing into this sector, which is the main growth engine. But PC, smartphone, automotive, and industrial markets are also key parts of our strategic layout. We aim to maintain a healthy, diversified portfolio across different end markets.

Q: Previously mentioned that some customers’ needs could only be met at 70%. What is the current situation?

CEO Mehrotra: In the last earnings call, we said that for some key customers, we could only meet 50% to two-thirds of their mid-term demand. The situation remains the same.

Q: Regarding the strategic customer agreement, is there a mechanism to limit gross margin risk during downturns?

CEO Mehrotra: Details cannot be disclosed. The agreements are multi-year with specific commitments, designed to provide visibility and stability. We have successfully completed one and are discussing similar arrangements with several other customers.

Q: How broad is your customer engagement for these agreements? Are they only with mega-clients? Do the agreements include upfront capital expenditure commitments? Are prices linked to investment returns?

CEO Mehrotra: The first strategic customer agreement was with a leading major client. Such agreements are valuable because they give us confidence to push forward with capacity investments. They include specific terms to help us better understand customer needs over the medium to long term. We are currently in discussions with multiple customers across various markets.

Q: Regarding HBM, the previous 40% CAGR (market size about $50 billion this year)—has there been any change? Do you see industry players shifting capacity from DDR5 to HBM?

CEO Mehrotra: Currently, non-HBM products have higher margins than HBM, but demand for HBM remains strong. We have not revised our overall market outlook for HBM. Data centers continue to demand DDR5, LPDDR, and HBM.

As AI demand in data centers continues to grow, we will dynamically adjust our product mix. While focusing on data centers, we also value maintaining share in other key markets. Micron’s full data center product line—including HBM, LPDDR, SoC-M, DDR5, and data center SSDs—has strong growth potential.

Q: Regarding enterprise SSDs, previously estimated to account for nearly half of NAND revenue, with 60% QoQ growth. February SSD revenue doubled QoQ and still accounts for 50% of NAND mix. Will G9 node improvements continue to support QoQ growth? How do you view the new high-bandwidth flash (HBF)? Will Micron invest in R&D?

CEO Mehrotra: Data center SSDs are a key growth area. Currently, NAND supply is tight, and demand is strong, making data center SSDs the main driver.

Micron’s SSD portfolio covers a wide range of capacity and performance needs, including TLC and QLC products. As part of our strategy to shift revenue toward higher-margin areas, we will continue investing in SSDs and are satisfied with current growth.

Regarding HBF, this technology offers advantages like high capacity, but it also faces physical limitations inherent to NAND, such as write speed, power consumption, and retention. It could be a solution for specific workloads, but it’s still early days.

Q: Are multi-year strategic customer agreements driven by the need for earlier, longer-term cooperation with GPU/XPU customers? Do they involve early design engagement, 12-18 month design cycles, IP sharing, process optimization? Do they require early customer involvement in design stages?

CEO Mehrotra: We cannot disclose specific customer types. But strategic agreements do foster closer R&D collaboration and roadmap planning, which is one benefit.

Q: Regarding LPU architecture and increased SRAM use, how do you view the long-term memory market? With more workloads relying on memory types beyond HBM, what is your outlook? Since much demand comes from long-term agreements and a few data center clients, how do you set benchmarks when expanding capacity? Do you have internal accelerators or forecasts? Do you communicate with customers to build bottom-up forecasts, ensuring supply sufficiency over three to five years without excess?

CEO Mehrotra: The emergence of LPU-like architectures makes AI infrastructure more efficient, which helps accelerate the overall AI market growth.

These innovations aim to: process workloads more efficiently, optimize token economics, improve inference speed and scalability, and reduce power consumption.

Any technology that enhances AI efficiency ultimately fuels demand. From this perspective, LPU is not a replacement for HBM or DRAM but a strong complement.

Without more and faster memory, AI expansion cannot proceed. An intuitive example: the DRAM demand for advanced AI accelerators has doubled over last year, and similar trends are happening in edge devices, smartphones, and PCs.

Q: The 81% gross margin guidance—historically, peak cycles hovered around 60%. How does this compare? What lessons do past cycles offer for future gross margin trends? How might customers react to such high margins?

CFO Murphy: Industry supply constraints and market tightness will persist beyond 2026, supporting short- and medium-term pricing. Whether margins will revert to historical averages depends on a new perspective.

Higher-performance memory reduces token costs and energy per token, increases throughput, and enhances AI model intelligence. This drives models toward more complex problem sets and AI applications, further increasing token consumption and memory demand.

The current high margins reflect a market revaluation of memory’s value—memory has become an efficient vehicle for monetizing AI capabilities. This logic extends from data centers to edge computing, broadening the scope.

Over the past year, we emphasized that supply-side constraints are multi-dimensional and hard to resolve quickly: industry inventory remains low; process node improvements are yielding fewer bits per wafer; HBM capacity utilization is rising; and new capacity requires building new wafer fabs—these physical constraints mean expansion cycles are measured in years.

Memory’s strategic value and supply bottlenecks form the two pillars of the current market landscape.

We are pursuing a dual approach: actively investing in capacity expansion and increasing R&D to drive technological iteration, continuously enhancing the value of memory products. We believe these efforts will underpin future margins. Customers are well aware of this, which is why more are signing long-term strategic agreements with us.

Special thanks to Jin Lu for contributing to this article.