The storage sector, which was the hottest sector before June, has undergone a round of adjustment and is now the most controversial sector. The core market is also unsure whether to view storage as a cycle or growth. Previously, it was believed that there were signs of growth in the market, but now it has returned to a cyclical pattern, and it must be irreversible. Goldman Sachs' latest report on storage is still worth reading. It outlines the eight core concerns in the current market and provides us with a framework to view storage. The following image is an analysis of the eight major doubts in the market by Goldman Sachs report. Tell me my understanding: 1. Which of these eight are the most convincing and which assumptions are the most radical? The three most convincing factors 1) The LTA structure has indeed undergone substantial changes Five year contract take-or-pay、 The price bottom line and large advance payments have been publicly disclosed by the company, which is the most different aspect of this cycle from historical cycles. 2) Original factory inventory is still low Inventory levels for 2-4 weeks are significantly lower than historical downturns, and both Samsung and Hynix have publicly stated that customer demand still exceeds supply capacity. 3) HBM expansion will squeeze the supply of ordinary DRAM The increase in HBM trade ratio, advanced packaging constraints, insufficient clean room space, and longer construction cycles of new factories make it difficult to achieve rapid expansion of both HBM and regular DRAM production simultaneously. The two most radical and in need of verification 1) By 2027, comparable HBM products will increase in price by approximately 60% There is logic in the direction, but the magnitude is very sensitive to factors such as GPU shipments, HBM4 yield, customer bargaining, LTA price ceiling, and the expansion discipline of the three suppliers. 2) Changxin's influence has long been limited to the mainland of China It may be established in the next year, but when viewed from the perspective of three to five years, this judgment is clearly optimistic. Unless there is a clear polarization in the storage supply chain due to political reasons. The real risk for Changxin is not necessarily to immediately compete for HBM, but to lower the long-term price center of ordinary DRAM. 2. The most valuable aspect of Goldman Sachs' report is that it proposes a possible framework: The market is still using the previous storage cycle of "spot dominance, supplier blind expansion, and rapid inventory accumulation" to value a new cycle increasingly dominated by AI server demand, advanced packaging constraints, long-term contracts, prepayments, and price bottom lines. Among them, the most solid bullish basis is: The LTA clause has been significantly strengthened; Original factory inventory is extremely low; HBM expands production and squeezes ordinary DRAM; Insufficient NAND capital expenditure; The demand for server storage is still growing. The most important thing to be vigilant about is: The price increase of HBM in 2027 may be too aggressive; Long term cooperation not only protects the bottom, but may also limit the top; The "one-time miss" of Hynix in the second quarter must be repaired in the third quarter; Changxin's global impact is limited in the short term, but it cannot be ignored in the medium to long term. 3. Several points to consider: 1) Will the peak of the cycle be later than market expectations? As long as the AI server HBM、 The demand for server DRAM and enterprise SSD continues to grow, and 2027 may not necessarily be the end of profitability. 2) Will the bottom of the next cycle be significantly higher than in the past? LTA、 Prepayments, price floors, and low inventory will reduce the probability of the past "price collapse inventory explosion huge losses". 3) Valuation logic needs to shift from peak profits to normalized profits Will normal profits (rather than peak profits) be maintained for a long time in the future? One point of consideration is that it is understandable to view DRAM and NAND as cyclical products, but HBM is actually highly bound to GPU and packaged together. To view HBM as a cyclical product, the premise is to also consider Nvidia GPU as a cyclical product (obviously GPU is not a cyclical product). For storage enterprises, there are shadows and signs of a new cycle now, but further proof is needed. The core still depends on demand. After new production capacity is gradually implemented in the next few years, will demand continue to grow rapidly. This content is sponsored by @ BITstocks_CN. Buy BIT-16000+US stocks and ETFs on the US stock market, hold real positions, and enjoy dividends.