#美股财报季 #Gate广场万圣节 Sandisk × Kioxia Earnings: Record Profits, 85% Gross Margin, 4x Forward P/E—Why Did Both Stocks Plunge?
From October 29 to 30, Sandisk (SNDK) and Kioxia (Tokyo: 285A / U.S. ADR: KXIAY) reported their latest earnings in succession, with both companies posting record profits. Yet on October 30, Sandisk shares fell 13.47% in the U.S., while Kioxia’s U.S. ADR plunged 17.11%. Storage companies are generating historic profits, but capital markets are selling off both stocks sharply. Is this divergence due to weakening NAND demand, or is the market starting to reassess how long these high profits can last and the risks of future supply?
I. Record Earnings—Why Is the Market Still Concerned?
First, Kioxia posted record profits, but growth is set to slow markedly next quarter.
For the quarter ended September, Kioxia’s revenue was ¥2.48 trillion, up 40.3% sequentially; Non-GAAP (non-generally accepted accounting principles) operating profit was ¥1.99 trillion, for a margin of about 80%. ASP (average selling price) rose by around the mid-20% range this quarter, while Bit shipments (measured by capacity) grew in the low double digits. But the company’s revenue guidance for next quarter is about ¥2.50 trillion, up only around 0.8% sequentially, with operating profit expected to be roughly flat. Management said the guidance was affected by the yen’s appreciation, enterprise SSD delivery cycles, and some delayed shipments. Data center products still have room for price increases, but the previous rapid pace of growth is unlikely to continue.
Second, Sandisk’s revenue and profit beat expectations, with sales volume contributing more to growth.
Revenue for FY2027’s first quarter was $11.031 billion, up 23% sequentially; Non-GAAP gross margin was 85.1%, and adjusted EPS was $49.51. Revenue guidance for next quarter is $12.2 billion–$12.6 billion, with the midpoint representing further sequential growth of about 12.4%. Management disclosed that about 60% of the sequential revenue increase came from volume and 40% from pricing; data center revenue was $4.802 billion, up 61% sequentially, making it the main growth driver.
Third, the sharp declines in both stocks need to be viewed in the context of the timeline before and after earnings.
Kioxia’s Tokyo-listed shares (285A) fell 7.14% on October 30, before the earnings release that evening; afterward, its U.S. ADR (KXIAY) fell 17.11%, while Sandisk dropped 13.47% the same day. The Philadelphia Semiconductor Index also saw a notable correction, as the market weighed concerns about AI capital expenditure by large U.S. tech companies and its long-term returns. Therefore, the entire decline cannot be attributed to a deterioration in the NAND industry, but Kioxia’s weaker near-term guidance and slower ASP growth did heighten investor concerns about the memory cycle.
II. AI Storage Remains in Short Supply, but High Prices Are Starting to Bite
The most important current change in the NAND industry is the divergence in different end markets’ ability to absorb high storage prices.
First, enterprise SSDs remain strong.
Sandisk’s data center revenue grew 61% sequentially, while SSD and Storage revenue accounted for about 65% of Kioxia’s total. AI inference, KV Cache (key-value cache), and large-scale data storage are driving demand for high-performance and high-capacity SSDs. TLC primarily serves use cases with higher performance and endurance requirements, while QLC is suited to data lakes and high-capacity storage. Kioxia has started mass production of 240TB QLC SSDs, and related Sandisk products have also entered the customer ramp-up phase. These are real commercial demand signals already in evidence. By contrast, new architectures such as high-endurance SLC storage and HBF (high-bandwidth flash) are still at the customer-discussion or product-development stage and should not be counted prematurely in mature earnings models. Second, high prices are starting to weigh on consumer end markets. Intel’s latest earnings report noted that rising memory and other component prices are suppressing PC demand. Sandisk’s Consumer business revenue also fell 29% sequentially this quarter, indicating that parts of the consumer market are under pressure. However, the memory Intel referred to includes DRAM and NAND, and Sandisk’s Consumer business is not equivalent to the entire PC market, so the declines cannot be directly attributed to higher NAND prices. In terms of demand scale, PCs account for roughly the low teens percentage of global NAND Bit demand, while servers already account for more than 40%. PCs are not the largest source of demand, but when the industry supply-demand gap is only a few percentage points, changes in PC demand can still affect price equilibrium.
Third, prices across the industry have started to diverge. The latest TrendForce data show that North American enterprise SSD contract prices are still rising, while NAND wafer contract prices held steady at October’s high levels, and client SSD price increases have also narrowed considerably.
NAND is therefore currently closer to a phase of strong enterprise demand and pressure on consumer markets than to a synchronized industry-wide peak. Going forward, the ability to sustain high profits will depend increasingly on enterprise demand growth, product mix, and changes in effective supply.
III. With Forward P/E at Just 4x, Are Memory Stocks Really Cheap?
After this week’s sharp share-price declines, one question has come into sharper focus: memory companies are posting record profits, while some have forward P/E ratios of only 4–5x. Does that mean the market is seriously undervaluing them?
First, a 4x P/E reflects forecast high earnings, not long-term profits that are already secured.
Take Sandisk as an example: it closed at $1,150.89 on October 30, and a post-earnings, same-source FY2028 adjusted EPS reference estimate of about $267.53 implies a P/E of just around 4.3x. This valuation looks extremely cheap, but FY2028 figures are still estimates. This cycle’s gross margin of around 85% reflects an extreme shortage environment. What the market is really questioning is how much profit the company can retain once NAND prices normalize. Second, share prices are determined jointly by long-term earnings power and the valuation multiple investors are willing to pay.
Assuming normalized EPS can remain at $250, even an 8x P/E would imply a share price of $2,000; but if normalized EPS is only $150, the same 8x P/E would imply $1,200; if it falls further to $100, that would imply only $800. These are scenario calculations, not forecasts of future earnings or price targets. They show that a 4x forward P/E could indicate undervaluation, or it could be a reasonable discount applied to profits near a cyclical peak. What determines valuation is not how high current profits are, but how long they can last and how much profits will fall in the next price downturn.
Third, long-term contracts and actual free cash flow are key to assessing whether the cyclical discount is excessive.
Sandisk uses NBM (New Business Model) long-term agreements to secure some demand from long-term customers, and has disclosed about $88.9 billion in RPO (remaining performance obligations); its Adjusted FCF (adjusted free cash flow) was about $5.7 billion this quarter. Kioxia also generated about ¥1.438 trillion in quarterly Core FCF (core free cash flow). Both companies are generating substantial, real cash flows—not just accounting profits. But their FCF measures are different, and future manufacturing investment, contract fulfillment, and funding arrangements for joint-venture factories must also be considered.
For Sandisk in particular, NBM can improve the visibility of revenue and purchase volumes, but it has yet to prove that all future contracts can sustain current gross margins over the long term. Contract pricing terms, renewal economics, and free-cash-flow returns after accounting for the full capital burden are key to determining whether it can command a valuation higher than that of a traditional cyclical stock. So the question investors should really be investigating is not “Why isn’t a 4x P/E driving the stock higher?” but rather: how much does the current share price discount normalized future profits, relative to the actual risks?
IV. Could New Chinese Supply Change the NAND Cycle Earlier Than Expected?
Alongside demand divergence, new supply from China also warrants close attention.
First, Yangtze Memory Technologies (YMTC) is now large enough to affect global supply and demand.
Reuters reported in April that YMTC is planning a new NAND factory, with Wuhan’s third plant expected to begin operations in late 2026 and planned to reach capacity of about 50,000 wafers per month in 2027. TrendForce expects Chinese manufacturers’ share of global NAND Bit output could approach 19% in 2027 as China expands capacity and global manufacturers upgrade process technology, potentially easing supply tightness gradually in the second half of the year. But there is a time lag between completing a factory, installing equipment, ramping yields, and producing Bit that can actually be sold, so planned capacity cannot be treated as supply that has already come online.
Second, new Chinese supply may hit the consumer market first before affecting global prices.
YMTC does not need to immediately enter leading North American cloud providers to change the competitive landscape through domestic PC, smartphone, SSD, and module markets. As overseas manufacturers reallocate available capacity, the pricing impact may gradually spread to other markets. Enterprise SSDs still face barriers such as qualification, reliability, and trade restrictions, so the competitive impact may lag. CXMT (ChangXin Memory Technologies) is also reportedly preparing a NAND R&D line, but it remains primarily a DRAM manufacturer, and its NAND production, which has not yet begun, should not be counted prematurely in supply forecasts.
Third, major overseas memory makers are still planning future capacity cautiously.
Kioxia and Sandisk previously announced plans to invest a combined total of more than $31 billion in Japan by 2032, subject to conditions including government support. Kioxia’s Kitakami K3 plant is planned to begin operations in FY2029; in the near term, the company is prioritizing use of existing factories and upgrades to BiCS technology. OpenAI CEO Sam Altman also recently said in the Cheeky Pint interview that memory makers have painful memories of expanding capacity only to see prices fall, which is why they remain cautious about investing today. New Chinese capacity, global process upgrades, and high prices suppressing consumer demand could gradually shift the NAND market from widespread shortages to localized shortages, even as AI demand remains strong.
V. Which Conclusions Are Clear Now, and Which Still Need to Be Tested?
Three conclusions are supported by fairly strong evidence, but two questions affecting long-term investment value remain unresolved. The three conclusions we can establish are: First, the NAND upcycle has not been disproven, and enterprise SSDs are a real source of growth.
Both companies posted record revenue and profits, Sandisk’s data center business grew 61% sequentially, and more of its revenue increase this quarter came from volume rather than price increases alone. AI inference and data center storage demand are already contributing to actual business results, not just industry expectations.
Second, the industry is already seeing divergence in demand and pricing.
Enterprise SSDs remain in short supply, while PC and other consumer end markets are starting to feel the effects of high component costs and weak demand; price increases for client SSDs have narrowed, and NAND wafer prices have stabilized. This means future industry profit growth can no longer simply rely on rapid price increases across all products.
Third, future supply expansion is a material competitive risk, not a distant hypothetical.
YMTC has the scale and technological capabilities to compete, new capacity is moving forward, and overseas manufacturers are also upgrading nodes. Even if AI demand continues to grow, increasing supply and changing consumer demand could still alter the future price balance.
Two key questions remain uncertain
First, when will NAND supply and demand rebalance in 2027–2028?
YMTC’s actual Bit output, new-factory yields, global node upgrades, and enterprise SSD procurement demand could all affect how long supply remains tight. There is not enough evidence to conclude that the industry has already peaked, nor can we guarantee that high prices will last.
Second, how much profit will Kioxia and Sandisk retain after normalization?
Sandisk’s NBM long-term contracts improve demand visibility, but cross-cycle gross margins, price protections, and the full manufacturing capital burden still need to be tested. Therefore, a forward P/E of around 4x could represent a potential investment opportunity, but it cannot yet be deemed proof that the market is mispricing the stocks.
What should we watch next?
Demand: Whether actual enterprise SSD Bit shipments and contract ASPs continue to grow, and whether PC and smartphone demand can stabilize.
Supply: The actual ramp-up, yields, and release pace of saleable Bit from YMTC’s new capacity, as well as node upgrades by global manufacturers.
Value capture: NBM contract fulfillment, gross margin trends, and how much free cash flow Kioxia and Sandisk can generate sustainably after accounting for the actual manufacturing capital burden.
From October 29 to 30, Sandisk (SNDK) and Kioxia (Tokyo: 285A / U.S. ADR: KXIAY) reported their latest earnings in succession, with both companies posting record profits. Yet on October 30, Sandisk shares fell 13.47% in the U.S., while Kioxia’s U.S. ADR plunged 17.11%. Storage companies are generating historic profits, but capital markets are selling off both stocks sharply. Is this divergence due to weakening NAND demand, or is the market starting to reassess how long these high profits can last and the risks of future supply?
I. Record Earnings—Why Is the Market Still Concerned?
First, Kioxia posted record profits, but growth is set to slow markedly next quarter.
For the quarter ended September, Kioxia’s revenue was ¥2.48 trillion, up 40.3% sequentially; Non-GAAP (non-generally accepted accounting principles) operating profit was ¥1.99 trillion, for a margin of about 80%. ASP (average selling price) rose by around the mid-20% range this quarter, while Bit shipments (measured by capacity) grew in the low double digits. But the company’s revenue guidance for next quarter is about ¥2.50 trillion, up only around 0.8% sequentially, with operating profit expected to be roughly flat. Management said the guidance was affected by the yen’s appreciation, enterprise SSD delivery cycles, and some delayed shipments. Data center products still have room for price increases, but the previous rapid pace of growth is unlikely to continue.
Second, Sandisk’s revenue and profit beat expectations, with sales volume contributing more to growth.
Revenue for FY2027’s first quarter was $11.031 billion, up 23% sequentially; Non-GAAP gross margin was 85.1%, and adjusted EPS was $49.51. Revenue guidance for next quarter is $12.2 billion–$12.6 billion, with the midpoint representing further sequential growth of about 12.4%. Management disclosed that about 60% of the sequential revenue increase came from volume and 40% from pricing; data center revenue was $4.802 billion, up 61% sequentially, making it the main growth driver.
Third, the sharp declines in both stocks need to be viewed in the context of the timeline before and after earnings.
Kioxia’s Tokyo-listed shares (285A) fell 7.14% on October 30, before the earnings release that evening; afterward, its U.S. ADR (KXIAY) fell 17.11%, while Sandisk dropped 13.47% the same day. The Philadelphia Semiconductor Index also saw a notable correction, as the market weighed concerns about AI capital expenditure by large U.S. tech companies and its long-term returns. Therefore, the entire decline cannot be attributed to a deterioration in the NAND industry, but Kioxia’s weaker near-term guidance and slower ASP growth did heighten investor concerns about the memory cycle.
II. AI Storage Remains in Short Supply, but High Prices Are Starting to Bite
The most important current change in the NAND industry is the divergence in different end markets’ ability to absorb high storage prices.
First, enterprise SSDs remain strong.
Sandisk’s data center revenue grew 61% sequentially, while SSD and Storage revenue accounted for about 65% of Kioxia’s total. AI inference, KV Cache (key-value cache), and large-scale data storage are driving demand for high-performance and high-capacity SSDs. TLC primarily serves use cases with higher performance and endurance requirements, while QLC is suited to data lakes and high-capacity storage. Kioxia has started mass production of 240TB QLC SSDs, and related Sandisk products have also entered the customer ramp-up phase. These are real commercial demand signals already in evidence. By contrast, new architectures such as high-endurance SLC storage and HBF (high-bandwidth flash) are still at the customer-discussion or product-development stage and should not be counted prematurely in mature earnings models. Second, high prices are starting to weigh on consumer end markets. Intel’s latest earnings report noted that rising memory and other component prices are suppressing PC demand. Sandisk’s Consumer business revenue also fell 29% sequentially this quarter, indicating that parts of the consumer market are under pressure. However, the memory Intel referred to includes DRAM and NAND, and Sandisk’s Consumer business is not equivalent to the entire PC market, so the declines cannot be directly attributed to higher NAND prices. In terms of demand scale, PCs account for roughly the low teens percentage of global NAND Bit demand, while servers already account for more than 40%. PCs are not the largest source of demand, but when the industry supply-demand gap is only a few percentage points, changes in PC demand can still affect price equilibrium.
Third, prices across the industry have started to diverge. The latest TrendForce data show that North American enterprise SSD contract prices are still rising, while NAND wafer contract prices held steady at October’s high levels, and client SSD price increases have also narrowed considerably.
NAND is therefore currently closer to a phase of strong enterprise demand and pressure on consumer markets than to a synchronized industry-wide peak. Going forward, the ability to sustain high profits will depend increasingly on enterprise demand growth, product mix, and changes in effective supply.
III. With Forward P/E at Just 4x, Are Memory Stocks Really Cheap?
After this week’s sharp share-price declines, one question has come into sharper focus: memory companies are posting record profits, while some have forward P/E ratios of only 4–5x. Does that mean the market is seriously undervaluing them?
First, a 4x P/E reflects forecast high earnings, not long-term profits that are already secured.
Take Sandisk as an example: it closed at $1,150.89 on October 30, and a post-earnings, same-source FY2028 adjusted EPS reference estimate of about $267.53 implies a P/E of just around 4.3x. This valuation looks extremely cheap, but FY2028 figures are still estimates. This cycle’s gross margin of around 85% reflects an extreme shortage environment. What the market is really questioning is how much profit the company can retain once NAND prices normalize. Second, share prices are determined jointly by long-term earnings power and the valuation multiple investors are willing to pay.
Assuming normalized EPS can remain at $250, even an 8x P/E would imply a share price of $2,000; but if normalized EPS is only $150, the same 8x P/E would imply $1,200; if it falls further to $100, that would imply only $800. These are scenario calculations, not forecasts of future earnings or price targets. They show that a 4x forward P/E could indicate undervaluation, or it could be a reasonable discount applied to profits near a cyclical peak. What determines valuation is not how high current profits are, but how long they can last and how much profits will fall in the next price downturn.
Third, long-term contracts and actual free cash flow are key to assessing whether the cyclical discount is excessive.
Sandisk uses NBM (New Business Model) long-term agreements to secure some demand from long-term customers, and has disclosed about $88.9 billion in RPO (remaining performance obligations); its Adjusted FCF (adjusted free cash flow) was about $5.7 billion this quarter. Kioxia also generated about ¥1.438 trillion in quarterly Core FCF (core free cash flow). Both companies are generating substantial, real cash flows—not just accounting profits. But their FCF measures are different, and future manufacturing investment, contract fulfillment, and funding arrangements for joint-venture factories must also be considered.
For Sandisk in particular, NBM can improve the visibility of revenue and purchase volumes, but it has yet to prove that all future contracts can sustain current gross margins over the long term. Contract pricing terms, renewal economics, and free-cash-flow returns after accounting for the full capital burden are key to determining whether it can command a valuation higher than that of a traditional cyclical stock. So the question investors should really be investigating is not “Why isn’t a 4x P/E driving the stock higher?” but rather: how much does the current share price discount normalized future profits, relative to the actual risks?
IV. Could New Chinese Supply Change the NAND Cycle Earlier Than Expected?
Alongside demand divergence, new supply from China also warrants close attention.
First, Yangtze Memory Technologies (YMTC) is now large enough to affect global supply and demand.
Reuters reported in April that YMTC is planning a new NAND factory, with Wuhan’s third plant expected to begin operations in late 2026 and planned to reach capacity of about 50,000 wafers per month in 2027. TrendForce expects Chinese manufacturers’ share of global NAND Bit output could approach 19% in 2027 as China expands capacity and global manufacturers upgrade process technology, potentially easing supply tightness gradually in the second half of the year. But there is a time lag between completing a factory, installing equipment, ramping yields, and producing Bit that can actually be sold, so planned capacity cannot be treated as supply that has already come online.
Second, new Chinese supply may hit the consumer market first before affecting global prices.
YMTC does not need to immediately enter leading North American cloud providers to change the competitive landscape through domestic PC, smartphone, SSD, and module markets. As overseas manufacturers reallocate available capacity, the pricing impact may gradually spread to other markets. Enterprise SSDs still face barriers such as qualification, reliability, and trade restrictions, so the competitive impact may lag. CXMT (ChangXin Memory Technologies) is also reportedly preparing a NAND R&D line, but it remains primarily a DRAM manufacturer, and its NAND production, which has not yet begun, should not be counted prematurely in supply forecasts.
Third, major overseas memory makers are still planning future capacity cautiously.
Kioxia and Sandisk previously announced plans to invest a combined total of more than $31 billion in Japan by 2032, subject to conditions including government support. Kioxia’s Kitakami K3 plant is planned to begin operations in FY2029; in the near term, the company is prioritizing use of existing factories and upgrades to BiCS technology. OpenAI CEO Sam Altman also recently said in the Cheeky Pint interview that memory makers have painful memories of expanding capacity only to see prices fall, which is why they remain cautious about investing today. New Chinese capacity, global process upgrades, and high prices suppressing consumer demand could gradually shift the NAND market from widespread shortages to localized shortages, even as AI demand remains strong.
V. Which Conclusions Are Clear Now, and Which Still Need to Be Tested?
Three conclusions are supported by fairly strong evidence, but two questions affecting long-term investment value remain unresolved. The three conclusions we can establish are: First, the NAND upcycle has not been disproven, and enterprise SSDs are a real source of growth.
Both companies posted record revenue and profits, Sandisk’s data center business grew 61% sequentially, and more of its revenue increase this quarter came from volume rather than price increases alone. AI inference and data center storage demand are already contributing to actual business results, not just industry expectations.
Second, the industry is already seeing divergence in demand and pricing.
Enterprise SSDs remain in short supply, while PC and other consumer end markets are starting to feel the effects of high component costs and weak demand; price increases for client SSDs have narrowed, and NAND wafer prices have stabilized. This means future industry profit growth can no longer simply rely on rapid price increases across all products.
Third, future supply expansion is a material competitive risk, not a distant hypothetical.
YMTC has the scale and technological capabilities to compete, new capacity is moving forward, and overseas manufacturers are also upgrading nodes. Even if AI demand continues to grow, increasing supply and changing consumer demand could still alter the future price balance.
Two key questions remain uncertain
First, when will NAND supply and demand rebalance in 2027–2028?
YMTC’s actual Bit output, new-factory yields, global node upgrades, and enterprise SSD procurement demand could all affect how long supply remains tight. There is not enough evidence to conclude that the industry has already peaked, nor can we guarantee that high prices will last.
Second, how much profit will Kioxia and Sandisk retain after normalization?
Sandisk’s NBM long-term contracts improve demand visibility, but cross-cycle gross margins, price protections, and the full manufacturing capital burden still need to be tested. Therefore, a forward P/E of around 4x could represent a potential investment opportunity, but it cannot yet be deemed proof that the market is mispricing the stocks.
What should we watch next?
Demand: Whether actual enterprise SSD Bit shipments and contract ASPs continue to grow, and whether PC and smartphone demand can stabilize.
Supply: The actual ramp-up, yields, and release pace of saleable Bit from YMTC’s new capacity, as well as node upgrades by global manufacturers.
Value capture: NBM contract fulfillment, gross margin trends, and how much free cash flow Kioxia and Sandisk can generate sustainably after accounting for the actual manufacturing capital burden.
