#股票交易分享挑战 Which is the most undervalued among Micron, SanDisk, SK hynix, and CXMT?
First, the framework: “undervalued” is not about whose P/E is the lowest, but whose price reflects the least expectation of a cyclical peak and offers the most downside protection.
First, the three-day market action: From August 12-14, the sector surged after SanDisk’s Investor Day ignited sentiment ($93.9 billion in long-term contracts + a long-term framework)—at the August 14 close: Micron $971.66 (+11.9% over three days), SanDisk $1,641.11 (+29.1%), SK hynix KRW 1.65M ≈ $1,165 (+15.9%), and CXMT ¥55.18 ≈ $8.18 (+9.6%).
All valuations and scenario ranges have been recalculated using this new baseline.
I. Valuation comparison: the hottest names are the most expensive
Valuation and market-cap comparison of the four companies
Several key readings (August 14 closing prices):
•Next-fiscal-year forward P/E: hynix 3.6x < SanDisk 6.2x < Micron 6.3x < CXMT approximately 30-34x (estimated based on annualized projected 2026 earnings);
•Trailing 12-month P/E excluding non-recurring items: hynix approximately 9.0x < SanDisk approximately 22.2x ≈ Micron 21.5x << CXMT approximately 1,970x (the 2025 earnings base is extremely low; this is not a typo);
•Market cap: Micron approximately $1.11 trillion, hynix approximately $827 billion, CXMT approximately $547.4 billion, and SanDisk approximately $239.6 billion;
•Balance sheet: hynix has approximately $49.1 billion in net cash (5.9% of market cap; after approximately $28.3 billion from the July ADR fundraising is received, pro forma net cash will be approximately $77.4 billion, or 9.4%), SanDisk has zero interest-bearing debt + a $15.5 billion buyback authorization (6.5% of market cap), and Micron has approximately $19.6 billion in net cash;
•Sell-side price targets are only useful as sentiment references—their targets at cyclical peaks are always lagging.
Consider the divergence: Micron’s lowest sell-side target is $361 and the highest is $2,200, a sixfold difference; for SanDisk, the sell-side average is $2,054 (+25% versus the current price), while independent platform TIKR’s neutral model is $1,335 (-19% versus the current price)—the same company, two different worlds.
CXMT’s forward P/E is 5-9 times that of the three overseas peers. Its IPO price-to-book ratio was 2.4x, promoted at the time as a “56% discount to overseas peers”—but after rising another 537% post-listing, the discount has long since become a premium (price-to-book is now approximately 15.3x).
What you are paying for “domestic substitution” is a sentiment premium, not an undervalued bargain.
II. Three-scenario calculation: how much are the upside and downside drawdowns? Assumptions: Bull case = continued upward revisions to AI demand + supply discipline remains intact; base case = prices peak and flatten in the first half of 2027 (Citigroup’s August 7 forecast of a peak in Q2 2027 is broadly aligned); bear case = downward demand revisions + concentrated capacity releases (CXMT’s capacity will reach 420k wafers/month in 2027, while hynix approved KRW 54.3 trillion in new-factory investment in August) + geopolitical shocks, with prices falling 10%-15% quarter over quarter for two consecutive quarters.
Upside from August 14 current prices (bull/base/bear):
•Micron: +44%/-7%/-56%
•SanDisk: +31%/-11%/-67%
•SK hynix: +42%/+5%/-53%
•CXMT: +118%/+15%/-62%
Risk/reward ratio (bull-case gain ÷ bear-case loss): Micron 0.78, SanDisk 0.46, hynix 0.78, CXMT 1.89.
Two issues must be made explicit:
First, after the three-day surge, Micron and SanDisk’s base-case target prices are already below their current prices. The model has not become more bearish—the target-price assumptions are exactly the same as in the August 11 version—but the current prices have outrun the base case. Under the base-case path, holding either for 12 months produces a negative expected return. This is the most important marginal change this week. Second, CXMT’s high risk/reward ratio is a trap. Its upside depends on the assumption that an “A-share scarcity premium” will persist, which is unrelated to fundamentals, while the July 2027 lock-up expiration for its IPO shares (when the free float will expand sharply from 6.73%) represents a supply overhang hanging overhead. A caveat is also needed regarding hynix’s “cheapness”: its three-scenario valuation multiples (3.2-4.5x) are systematically about 30%-35% lower than Micron’s (5-7x), representing an implicit “Korea discount.” It listed on Nasdaq (SKHY) on July 10, ostensibly to hedge this discount—but one month after listing, the ADR was instead trading at approximately a 30% premium to the Korean shares (August 5 data), meaning U.S. investors paid a 30% access-channel fee. If the discount were to fully converge, hynix’s base-case target price could rise another approximately 50%—but the discount has persisted since 2016, so convergence can only be treated as a free option, not a base-case assumption.
III. Could the bear case really fall that much? Four sets of data provide validation
The first reaction to a bear-case drawdown of -53% to -67% is always “that’s too extreme.” Let’s test it using four independent data sets:
1. This cycle has already provided a real-world test: From June 25 to July 29, 2026, based on closing prices, SanDisk fell 56.5%, hynix 54.7%, and Micron 39.1%—while contract prices were still rising quarter over quarter and no company had lowered guidance. Pure sentiment unwinding caused declines of this magnitude. If this can happen midway through an upswing, a genuine downcycle will only be deeper.
2. Two complete historical downturns: Micron fell 45% in 2H 2018-2019 and 46% in 2022-1H 2023 (monthly closing prices). This cycle’s stock-price gains (8-10x) far exceed those of the previous two cycles (2-2.5x), and a higher base implies a deeper mean reversion.
3. Implied drawdowns from the most bearish sell-side targets: Micron $361 (-63%), SanDisk $1,000 (-39%), and hynix KRW 1.2 million (-27%). The small group of most bearish analysts implies an average decline of 43%, while sell-side targets are systematically too high near peaks.
4. Price-to-book anchor: If Micron falls to the bear-case target of $425, it would still correspond to approximately 4.8x book value—whereas the bottoms of the 2018 and 2022 cycles were only 1-1.5x. In other words, the bear case does not even reach the “historical extreme bottom”; it merely assumes a normal cyclical reversion.
There are also positive buffers (to avoid a one-sided argument): hynix’s pro forma net cash is 9.4% of market cap, and it completed a KRW 12.24 trillion share buyback and cancellation in February; SanDisk has a $15.5 billion buyback plus $93.9 billion in guaranteed long-term contracts. Conclusion: the bear-case magnitude should not be reduced; the historical evidence is sufficient. What should genuinely raise the alert level is the new fact that the base-case scenario is already below the current price.
IV. The unpleasant facts must be stated upfront: risk list
Industry-level:
•Low-P/E trap: “Cheapness” is entirely built on the consensus that next fiscal year’s earnings will double again, while price momentum has already been cut in half; Citigroup has publicly forecast a price peak in Q2 2027—once prices flatten, consensus estimates will be systematically revised downward;
•An 84.6% gross margin is an extreme in memory history (the previous-cycle peak was approximately 46%), and mean reversion will only be delayed; SanDisk’s own long-term target is only approximately 80%;
•Supply side: CXMT will account for approximately 17% of global DRAM capacity in 2027; if Samsung’s HBM4 passes NVIDIA certification, it will compress the HBM premium; hynix’s $28.3 billion ADR fundraising will be invested entirely in new capacity;
•Weakness is already visible on the consumption side: PC OEMs are resisting double-digit price increases, SanDisk’s consumer revenue fell 32% quarter over quarter, and Jefferies flagged inventory accumulation in its peripheral businesses; cloud providers are experiencing memory inventory accumulation due to server CPU shortages.
Company-level (the single most critical issue for each):
•Micron: Mainland China historically accounted for approximately one-quarter of direct revenue, and it faced procurement restrictions in 2023—tighter export controls or accelerated domestic substitution would hit it most directly;
•SanDisk: a pure flash-memory single-medium company, with no memory/HBM hedge; it has been publicly listed for only 18 months and has not experienced a complete downcycle; NBM long-term contracts establish a floor but also cap the upside elasticity of approximately two-thirds of its bits;
•SK hynix: the Korea discount may not converge because of the ADR listing (it currently instead reflects an approximately 30% ADR premium); the new share issuance causes approximately 2% dilution; net profit propped up by one-off gains could mislead those who focus only on net profit;
•CXMT: the greatest valuation risk among the four; technologically several years behind the three giants, with HBM not yet in mass production (the prospectus does not mention HBM at all); depreciation during the capacity-expansion peak will suppress margins; accumulated unabsorbed losses total ¥36.6 billion, leaving no short-term dividend capacity; lock-up expiration in July 2027.
V. Final answer: Should you buy, how much can you afford to lose, and when should you admit you are wrong?
Valuation ranking: hynix > SanDisk ≈ Micron > CXMT.
The biggest difference from before: hynix listed on Nasdaq (SKHY) on July 10, meaning U.S. brokerage accounts can, for the first time in history, directly buy the cheapest of the four—but with the ADR trading at approximately a 30% premium to the Korean shares, the valuation advantage remains after paying the access-channel fee, though it has narrowed significantly. SanDisk is the most balanced pure-play combination of “odds + protection” on the U.S. main board (zero debt + buybacks + guaranteed long-term contracts), but it is a “late-cycle momentum + buyback + long-term contract” trade, and its investment thesis must be revalidated every earnings season.
①Should you buy: At the industry level, price momentum has been cut in half, share prices have swung sharply after earnings, and after the three-day surge the base-case targets are already below current prices (Micron -7%, SanDisk -11%), so the marginal risk/reward of chasing the rally is poor. A better strategy is to wait for a sell-off caused by quarter-over-quarter prices turning negative, or at least wait for Micron to confirm approximately $50 billion in next-quarter guidance at the end of September.
②How much can you afford to lose: The bear-case drawdown is 53%-67%, and June-July already provided a real-world “drill” of -39%--57%. Set the position-size limit at a level where losing half the money would not affect your life. This industry should never be leveraged at any price.
③When to admit you are wrong (exit and reassess upon any one trigger): Contract prices turn negative quarter over quarter for two consecutive quarters; leading cloud providers cut AI capital-expenditure guidance or long-term contracts are defaulted on/renegotiated at lower prices (with particular attention to SanDisk’s NBM fulfillment); CXMT/Samsung capacity deployment significantly exceeds expectations; at the individual-stock level—SanDisk gross margin falls below 70%, Micron’s next-quarter guidance comes in below $50 billion and is then cut again, hynix’s net profit continues to depend on non-operating income, or CXMT’s quarterly gross margin falls more than 5 percentage points quarter over quarter or shareholders engage in concentrated selling ahead of the lock-up expiration.
No one can guarantee that a price curve that has risen sevenfold will continue to do so. But rather than asking “can it keep rising,” the more important question is: if it falls, do I know what to do? Writing down stop-loss conditions before buying is the only survival rule worth carrying away from cyclical stocks.$CXMT $SK Hynix
First, the framework: “undervalued” is not about whose P/E is the lowest, but whose price reflects the least expectation of a cyclical peak and offers the most downside protection.
First, the three-day market action: From August 12-14, the sector surged after SanDisk’s Investor Day ignited sentiment ($93.9 billion in long-term contracts + a long-term framework)—at the August 14 close: Micron $971.66 (+11.9% over three days), SanDisk $1,641.11 (+29.1%), SK hynix KRW 1.65M ≈ $1,165 (+15.9%), and CXMT ¥55.18 ≈ $8.18 (+9.6%).
All valuations and scenario ranges have been recalculated using this new baseline.
I. Valuation comparison: the hottest names are the most expensive
Valuation and market-cap comparison of the four companies
Several key readings (August 14 closing prices):
•Next-fiscal-year forward P/E: hynix 3.6x < SanDisk 6.2x < Micron 6.3x < CXMT approximately 30-34x (estimated based on annualized projected 2026 earnings);
•Trailing 12-month P/E excluding non-recurring items: hynix approximately 9.0x < SanDisk approximately 22.2x ≈ Micron 21.5x << CXMT approximately 1,970x (the 2025 earnings base is extremely low; this is not a typo);
•Market cap: Micron approximately $1.11 trillion, hynix approximately $827 billion, CXMT approximately $547.4 billion, and SanDisk approximately $239.6 billion;
•Balance sheet: hynix has approximately $49.1 billion in net cash (5.9% of market cap; after approximately $28.3 billion from the July ADR fundraising is received, pro forma net cash will be approximately $77.4 billion, or 9.4%), SanDisk has zero interest-bearing debt + a $15.5 billion buyback authorization (6.5% of market cap), and Micron has approximately $19.6 billion in net cash;
•Sell-side price targets are only useful as sentiment references—their targets at cyclical peaks are always lagging.
Consider the divergence: Micron’s lowest sell-side target is $361 and the highest is $2,200, a sixfold difference; for SanDisk, the sell-side average is $2,054 (+25% versus the current price), while independent platform TIKR’s neutral model is $1,335 (-19% versus the current price)—the same company, two different worlds.
CXMT’s forward P/E is 5-9 times that of the three overseas peers. Its IPO price-to-book ratio was 2.4x, promoted at the time as a “56% discount to overseas peers”—but after rising another 537% post-listing, the discount has long since become a premium (price-to-book is now approximately 15.3x).
What you are paying for “domestic substitution” is a sentiment premium, not an undervalued bargain.
II. Three-scenario calculation: how much are the upside and downside drawdowns? Assumptions: Bull case = continued upward revisions to AI demand + supply discipline remains intact; base case = prices peak and flatten in the first half of 2027 (Citigroup’s August 7 forecast of a peak in Q2 2027 is broadly aligned); bear case = downward demand revisions + concentrated capacity releases (CXMT’s capacity will reach 420k wafers/month in 2027, while hynix approved KRW 54.3 trillion in new-factory investment in August) + geopolitical shocks, with prices falling 10%-15% quarter over quarter for two consecutive quarters.
Upside from August 14 current prices (bull/base/bear):
•Micron: +44%/-7%/-56%
•SanDisk: +31%/-11%/-67%
•SK hynix: +42%/+5%/-53%
•CXMT: +118%/+15%/-62%
Risk/reward ratio (bull-case gain ÷ bear-case loss): Micron 0.78, SanDisk 0.46, hynix 0.78, CXMT 1.89.
Two issues must be made explicit:
First, after the three-day surge, Micron and SanDisk’s base-case target prices are already below their current prices. The model has not become more bearish—the target-price assumptions are exactly the same as in the August 11 version—but the current prices have outrun the base case. Under the base-case path, holding either for 12 months produces a negative expected return. This is the most important marginal change this week. Second, CXMT’s high risk/reward ratio is a trap. Its upside depends on the assumption that an “A-share scarcity premium” will persist, which is unrelated to fundamentals, while the July 2027 lock-up expiration for its IPO shares (when the free float will expand sharply from 6.73%) represents a supply overhang hanging overhead. A caveat is also needed regarding hynix’s “cheapness”: its three-scenario valuation multiples (3.2-4.5x) are systematically about 30%-35% lower than Micron’s (5-7x), representing an implicit “Korea discount.” It listed on Nasdaq (SKHY) on July 10, ostensibly to hedge this discount—but one month after listing, the ADR was instead trading at approximately a 30% premium to the Korean shares (August 5 data), meaning U.S. investors paid a 30% access-channel fee. If the discount were to fully converge, hynix’s base-case target price could rise another approximately 50%—but the discount has persisted since 2016, so convergence can only be treated as a free option, not a base-case assumption.
III. Could the bear case really fall that much? Four sets of data provide validation
The first reaction to a bear-case drawdown of -53% to -67% is always “that’s too extreme.” Let’s test it using four independent data sets:
1. This cycle has already provided a real-world test: From June 25 to July 29, 2026, based on closing prices, SanDisk fell 56.5%, hynix 54.7%, and Micron 39.1%—while contract prices were still rising quarter over quarter and no company had lowered guidance. Pure sentiment unwinding caused declines of this magnitude. If this can happen midway through an upswing, a genuine downcycle will only be deeper.
2. Two complete historical downturns: Micron fell 45% in 2H 2018-2019 and 46% in 2022-1H 2023 (monthly closing prices). This cycle’s stock-price gains (8-10x) far exceed those of the previous two cycles (2-2.5x), and a higher base implies a deeper mean reversion.
3. Implied drawdowns from the most bearish sell-side targets: Micron $361 (-63%), SanDisk $1,000 (-39%), and hynix KRW 1.2 million (-27%). The small group of most bearish analysts implies an average decline of 43%, while sell-side targets are systematically too high near peaks.
4. Price-to-book anchor: If Micron falls to the bear-case target of $425, it would still correspond to approximately 4.8x book value—whereas the bottoms of the 2018 and 2022 cycles were only 1-1.5x. In other words, the bear case does not even reach the “historical extreme bottom”; it merely assumes a normal cyclical reversion.
There are also positive buffers (to avoid a one-sided argument): hynix’s pro forma net cash is 9.4% of market cap, and it completed a KRW 12.24 trillion share buyback and cancellation in February; SanDisk has a $15.5 billion buyback plus $93.9 billion in guaranteed long-term contracts. Conclusion: the bear-case magnitude should not be reduced; the historical evidence is sufficient. What should genuinely raise the alert level is the new fact that the base-case scenario is already below the current price.
IV. The unpleasant facts must be stated upfront: risk list
Industry-level:
•Low-P/E trap: “Cheapness” is entirely built on the consensus that next fiscal year’s earnings will double again, while price momentum has already been cut in half; Citigroup has publicly forecast a price peak in Q2 2027—once prices flatten, consensus estimates will be systematically revised downward;
•An 84.6% gross margin is an extreme in memory history (the previous-cycle peak was approximately 46%), and mean reversion will only be delayed; SanDisk’s own long-term target is only approximately 80%;
•Supply side: CXMT will account for approximately 17% of global DRAM capacity in 2027; if Samsung’s HBM4 passes NVIDIA certification, it will compress the HBM premium; hynix’s $28.3 billion ADR fundraising will be invested entirely in new capacity;
•Weakness is already visible on the consumption side: PC OEMs are resisting double-digit price increases, SanDisk’s consumer revenue fell 32% quarter over quarter, and Jefferies flagged inventory accumulation in its peripheral businesses; cloud providers are experiencing memory inventory accumulation due to server CPU shortages.
Company-level (the single most critical issue for each):
•Micron: Mainland China historically accounted for approximately one-quarter of direct revenue, and it faced procurement restrictions in 2023—tighter export controls or accelerated domestic substitution would hit it most directly;
•SanDisk: a pure flash-memory single-medium company, with no memory/HBM hedge; it has been publicly listed for only 18 months and has not experienced a complete downcycle; NBM long-term contracts establish a floor but also cap the upside elasticity of approximately two-thirds of its bits;
•SK hynix: the Korea discount may not converge because of the ADR listing (it currently instead reflects an approximately 30% ADR premium); the new share issuance causes approximately 2% dilution; net profit propped up by one-off gains could mislead those who focus only on net profit;
•CXMT: the greatest valuation risk among the four; technologically several years behind the three giants, with HBM not yet in mass production (the prospectus does not mention HBM at all); depreciation during the capacity-expansion peak will suppress margins; accumulated unabsorbed losses total ¥36.6 billion, leaving no short-term dividend capacity; lock-up expiration in July 2027.
V. Final answer: Should you buy, how much can you afford to lose, and when should you admit you are wrong?
Valuation ranking: hynix > SanDisk ≈ Micron > CXMT.
The biggest difference from before: hynix listed on Nasdaq (SKHY) on July 10, meaning U.S. brokerage accounts can, for the first time in history, directly buy the cheapest of the four—but with the ADR trading at approximately a 30% premium to the Korean shares, the valuation advantage remains after paying the access-channel fee, though it has narrowed significantly. SanDisk is the most balanced pure-play combination of “odds + protection” on the U.S. main board (zero debt + buybacks + guaranteed long-term contracts), but it is a “late-cycle momentum + buyback + long-term contract” trade, and its investment thesis must be revalidated every earnings season.
①Should you buy: At the industry level, price momentum has been cut in half, share prices have swung sharply after earnings, and after the three-day surge the base-case targets are already below current prices (Micron -7%, SanDisk -11%), so the marginal risk/reward of chasing the rally is poor. A better strategy is to wait for a sell-off caused by quarter-over-quarter prices turning negative, or at least wait for Micron to confirm approximately $50 billion in next-quarter guidance at the end of September.
②How much can you afford to lose: The bear-case drawdown is 53%-67%, and June-July already provided a real-world “drill” of -39%--57%. Set the position-size limit at a level where losing half the money would not affect your life. This industry should never be leveraged at any price.
③When to admit you are wrong (exit and reassess upon any one trigger): Contract prices turn negative quarter over quarter for two consecutive quarters; leading cloud providers cut AI capital-expenditure guidance or long-term contracts are defaulted on/renegotiated at lower prices (with particular attention to SanDisk’s NBM fulfillment); CXMT/Samsung capacity deployment significantly exceeds expectations; at the individual-stock level—SanDisk gross margin falls below 70%, Micron’s next-quarter guidance comes in below $50 billion and is then cut again, hynix’s net profit continues to depend on non-operating income, or CXMT’s quarterly gross margin falls more than 5 percentage points quarter over quarter or shareholders engage in concentrated selling ahead of the lock-up expiration.
No one can guarantee that a price curve that has risen sevenfold will continue to do so. But rather than asking “can it keep rising,” the more important question is: if it falls, do I know what to do? Writing down stop-loss conditions before buying is the only survival rule worth carrying away from cyclical stocks.$CXMT $SK Hynix
