美光科技

Micron Technology posted another blowout quarter, with sales rising 380% year on year to $54 billion. Growth was led nearly entirely by pricing growth. Gross margin expanded more than 40 points year on year to 87%, with incremental pricing-led revenue coming as practically pure profit.

Why it matters: The highest pricing growth we've ever seen in the modern memory market continues. Still, quarter-on-quarter pricing growth is decelerating, and we believe a peak is coming in early 2028. To us, it looks like we're in the final stretch of this unprecedented cyclical upswing. We've been bullish on pricing through 2027, with low supply coming online before 2028. It looks to be moderating more quickly than we expected, but is nonetheless extremely positive. We remain bullish on the next four quarters, but this deceleration indicates to us that a peak is on the horizon. We maintain that the current paradigm is a tremendous cycle, but a cycle nonetheless. We believe commodity pricing dynamics will hold, and as a massive amount of new supply comes online through 2030, there will be downward pressure on memory prices.

The bottom line: We lower our fair value estimate for no-moat Micron to $700, from $850, behind moderated short-term upcycle growth forecasts and a continued expectation for a steep downturn starting in 2029. Shares were flat after hours, and remain overvalued to us. The market appears to be pricing in a partial downturn, but not enough of one. We see memory supply doubling by 2028, driving a steep pricing decline across 2029 and 2031. After huge capital spending, we expect heady depreciation to commensurately drive margins back down to earth. The flat market reaction shows us that much of the current upswing has already been priced in. To us, the "easy money" has been made—1,100% appreciation since the start of 2025. We see significant risk for long-term investors, especially once the market starts to bake in a pricing peak.

The key debate on Micron is whether the current cycle is actually a structural, durable shift in memory dynamics. The structural bull thesis here rests on four tenets. First, that artificial intelligence is leading demand growth to new highs. Second, that long-term supply agreements, or as Micron dubs them, SCAs, will keep prices at high levels even under a higher supply backdrop. Third, that high-bandwidth memory is a differentiated product within DRAM. Fourth, that market participants will now act rationally in regards to supply, leveling out future cyclicality. We do not agree with this thesis.

On AI, it's true that it has greatly raised the demand growth profile for memory. However, demand growth is nothing new for memory—the only metric of import to memory investors is the ratio of supply to demand. High demand only benefits prices (and thus results) if it exceeds supply, as it does currently. If and when supply matches or exceeds demand, we still expect to see cyclicality manifest. Our work on global memory supply capacity plans indicates supply will exceed demand by the end of 2028, with steepening oversupply into 2030.

Secondly, we don't believe these long-term agreements are ironclad. The memory industry has attempted long-term agreements before, with little success in preventing a downturn (see Micron's 50% revenue decline in 2023). It's true that these new agreements have more teeth—Micron cited $32 billion in total nonrefundable deposits under its current set of agreements. Nonetheless, we estimate that the actual nonrefundable portions of these contracts represent less than 20% of the total agreed-upon dollar value. As such, we see these as a hedge, rather than a defense against a downcycle. If and when prices come down, we believe customers hold immense power to renegotiate terms, given memory is commoditized with four to six players competing for sales.

Third, we don't see HBM breaking commodity dynamics. We agree it's more differentiated than DRAM, with a logic base die that augments connection speeds. Nonetheless, we see the same dynamics as traditional DRAM, where Micron, SK Hynix, and Samsung all advance at a similar cadence and offer fungible products to customers like Nvidia. To us, talk of "custom HBM" is mainly marketing, and we even observe an increasing trend of memory suppliers outsourcing base die design to logic chipmakers, thus focusing on the more commoditized memory stack.

Finally, consider us flabbergasted if the industry has finally decided to collude into rational behavior regarding supply after two decades of volatile memory cycles. In a pricing upswing, every supplier builds to capture a piece of the pie. Not doing so would be to forego significant share and face a worse competitive position long-term. While it's true no supplier tries to overbuild, the lack of coordination and lack of accuracy in forecasting future demand mean that eventually supply will exceed demand. When oversupply hits and prices fall, suppliers readily take lower prices so they can ship their product and avoid costly underutilization manufacturing charges and obsolete inventory impairments.

As a truly final point on rational building during a cycle, the entrance of Chinese competitors—CXMT in DRAM and YMTC in NAND—makes rational industry building even less likely. These new entrants are publicly seeking to take aggressive share in their markets, augmenting and pulling forward the risk of oversupply globally, in our view.

We believe investors should see skyrocketing capital spending as a warning sign to the cycle turning. Micron's capital spending nearly quadrupled from 2024 to 2026, and we expect it to nearly double again in 2027. This spending takes two to three years to equate to actual memory output, adhering to our view of a glut of supply coming online starting in 2028. Even when prices fall, Micron will be on the hook for significantly higher depreciation from this building, which creates a double-jeopardy impact by further compressing margins in a downswing.

Our fair value estimate for Micron is $700 per share. Our valuation implies a fiscal 2027 adjusted price/earnings multiple of 4 times and an enterprise value/sales multiple of 2 times. Micron is often analyzed with price/book, which we put at 2.4 times against our fiscal 2027 estimates. We foresee price/book returning to its historical breakeven level of 1.0 between 2029 and 2030. The greatest drivers to our valuation are the massive upcycle we expect through fiscal 2028 for both DRAM and NAND prices, and the subsequent cyclical downturn we predict between 2029 and 2031.

We forecast Micron’s results to be cyclical in the long term and note that fiscal 2023 was among the most severe downturns we’ve seen for the chipmaker, with supply gluts across data centers, smartphones, and PCs following the postpandemic-era demand surge. Micron was also hampered in fiscal 2023 by the loss of revenue from memory sales to Chinese data center customers, due to retaliatory measures in response to other US government restrictions.

Through 2028, we see Micron benefiting from an unprecedented cyclical upswing. We see AI-driven demand for Micron’s HBM and DRAM chips as the firm’s largest growth driver, going forward. After more than 60% revenue growth in fiscal 2024 and nearly 50% growth in fiscal 2025, fiscal 2026 revenue more than tripled to $133 billion. We expect revenue to more than double again in fiscal 2027. We predict that both DRAM and NAND prices will peak in 2028, followed by a strong and enduring downcycle between 2029 and 2031.

While the DRAM and NAND markets can be volatile, we view long-term trends as showing growth in bit shipments (which equates to volume) and declines in price per bit. In DRAM, we expect high-teens bit growth and low double-digit annual pricing declines over the long term. In NAND, we expect mid-teens bit growth, and roughly 15% annual pricing declines over the long term. We see HBM boosting Micron’s DRAM bit growth and pricing, but for it to exhibit DRAM-like pricing declines after fiscal 2028 as higher supply drives down prices. Still, we observe HBM as margin-accretive, with roughly three to four times the price and only three times the cost as traditional DRAM on a per-bit basis in the long term.

In the short term, we believe prices and volume growth are far from these historical long-term trends. We expect above-trend supply growth through 2030 as the industry races to meet rapidly rising AI demand—which in turn drives downward pricing pressure in our model. After the brunt of the downturn, we expect worse-than-trend pricing declines for some years as overbuilt supply through 2030 leaves a lasting impact. We also believe aggressive investment from Chinese entrants CXMT (DRAM) and YMTC (NAND) will further pressure prices past the current upcycle.

Practically all of Micron’s costs of sales are fixed, leading to high gross margin leverage. We expect Micron to continue investing in manufacturing for the long term and gross margins to move in line with volumes. In the medium term, we see an impressive gross margin reaching the high 80% range with high pricing. In the long term, we think Micron’s elevated depreciation from the current build cycle will drive gross margins back down toward 30%.

Micron’s operating expenses are more flexible than its manufacturing costs, but still offer leverage. We forecast steady growth in research and development in the midsingle digits over the long term, and for other operating expenses to track top-line volume. Still, Micron’s operating margins will be heavily dictated by its gross margins. In the current upcycle, we forecast non-GAAP operating margins rising to the mid 80% range. At midcycle in the longer term, we think Micron can earn non-GAAP operating margins of around 20%. This implies a higher level than the historical midcycle, with Micron durably benefiting from the immense operating leverage it’s experiencing in the current upcycle.

We do not believe Micron possesses an economic moat. The firm operates in the highly capital-intensive industries of DRAM and NAND flash semiconductors, and it doesn’t earn good enough profit margins through the course of a cycle to give us confidence in enduring economic profits.

We view DRAM and NAND as commoditylike products prone to market supply-and-demand dynamics and steady pricing erosion that reduces industry profitability. While we see better margins in DRAM, which has consolidated to three primary players (versus six in NAND) we still don’t see good enough returns on invested capital over the course of a cycle to merit a moat. We do expect to see years of positive economic profits during periods of tight supply and strong demand that help support good pricing, but we see steep downcycles giving rise to poor profitability and eroding the economic profits built up in a market upswing.

The majority of Micron’s sales comes from DRAM chip sales. DRAM chips provide volatile random access memory, which is used to store memory during a system’s operation, and only while a system is turned on. DRAM is a critical component for data center servers, computers, smartphones, and industrial applications. We see DRAM chips as commodity products, primarily sold by three suppliers—Samsung, SK Hynix, and Micron—that occupy over 90% of the market. DRAM technology generations last years, and we see the leading oligopoly players moving in step to successive generations and participating in an industry standards-setting body. DDR5 is the latest technology transition for mainstream DRAM, and while a player like Micron will cite technological leadership or being first to market, we view these chips as fungible and time-to-market differences of a quarter or two as immaterial to competitive positioning.

As a vertically integrated producer, Micron has massive capital investment needs that weigh on its returns on invested capital. The DRAM industry typically exhibits high-single-digit annual pricing declines, with the dominant players focusing not only on product advancement, but also advances in cost efficiency to keep up with their peers. This rapid pricing erosion makes it difficult for industry players to earn strong profits, even with large market shares. Consolidation of the DRAM industry and rational acting by players have helped improve industry margins. Still, we don’t see current margin levels providing durable economic profit levels over the course of a cycle. The memory chip market is prone to bouts of undersupply that raise prices and profits, followed by gluts of supply that crater pricing. Downcycles can be severe, sometimes leading to negative returns on invested capital and negative profit margins.

Micron also has a meaningful NAND flash chip business that makes up the minority of sales. NAND flash chips provide nonvolatile memory for storage purposes. Compared with DRAM, NAND is used for longer-term storage of data for data centers and consumer hardware. NAND advancements happen at a much faster pace than those in DRAM, with suppliers putting out higher-density chip generations every 18-24 months. Similar to DRAM, while suppliers will tout technological differences like layer counts and being first to market, we view NAND chips as fungible and commoditylike among the leading players.

We view very similar dynamics in NAND as we do in DRAM: commodity products, high capital intensity, and volatile market cycles. NAND, however, is less consolidated of a market with six primary players. In our view, the more fragmented nature of the NAND market creates more volatile supply-and-demand dynamics with worse pricing and profitability in market downturns caused by oversupply. NAND is also even more capital intensive than DRAM is, which we attribute to faster advancement and shorter product generations, and exhibits steeper annual pricing declines typically in the midteens. All of this leads to a lower margin profile for NAND compared with DRAM, as well as lower returns on invested capital.

When memory markets are in an upswing and demand is strong, Micron’s sales growth and profitability can be impressive.

Micron is benefiting from immense growth in HBM revenue, due to high investments in AI infrastructure. This boosts the firm’s growth and margin profile.

We like Micron’s shareholder returns and view its balance sheet as strong for a cyclical firm.

Micron has a high fixed cost base that leaves it vulnerable to underutilization charges and major profit compression when memory markets enter a downturn.

We see DRAM and NAND as commoditylike products, and we foresee little ability for Micron to build durable differentiation against its competitors.

Micron is extremely sensitive to memory pricing, and weaker pricing growth in an upcycle, or a more volatile downcycle in the future, can greatly depress its results and valuation.