The poster is bullish on Micron (MU) and the DRAM ETF. The core thesis is that AI data-center demand is transforming memory from a traditionally cyclical commodity into a long-term growth asset that is closer to a bottleneck in AI infrastructure. The evidence chain has three main layers: First, hyperscaler AI capex is not a one-off buildout, because buildings and power assets can last for decades, but chips such as GPUs, HBM, DRAM, and SSDs typically need to be replaced after 3-5 years due to depreciation, wear, or falling behind on performance. The poster therefore believes this creates a sustained refresh cycle. Second, Micron’s demand mix is shifting from consumer electronics such as PCs and smartphones toward hyperscalers, enterprises, and data-center customers, while part of HBM supply is locked in through multi-year contracts. Based on this, the poster argues demand is more “sticky” and that downcycles will be milder than in the past. Third, the poster extrapolates from the latest Q2 FY2026 EPS of $12.20 and roughly $19 EPS around Q3 guidance, assuming about $20 EPS per quarter over the next four quarters, or roughly $80 NTM EPS. Using about 1.1 billion shares, the poster derives about $88B in net income, then compares MU’s valuation multiple with large tech companies such as NVDA, MSFT, AAPL, GOOGL, and META. The poster argues MU is still priced at only about 8x forward P/E, and that if it rerates to 10x, 15x, 20x, 25x, or 30x, the corresponding share prices would be about $800, $1200, $1600, $2000, and $2400, respectively. The poster included no images and explicitly acknowledged that much of the body text was AI-generated. In terms of positioning, the poster claims to be long MU and DRAM. The most important implicit assumptions in the post are that AI capex will grow for a long time or at least remain elevated, tight supply-demand conditions in HBM and high-end DRAM will spread and persist, enterprise customers will not cut orders due to ROI, macro conditions, or excessive prices, and Micron’s margins will not quickly mean-revert as capacity expands and prices fall. The biggest flaw is that the poster approximates sustainable earnings power by annualizing quarterly EPS from an extremely strong upcycle, and compares Micron’s P/E with software/platform-style large tech companies, downplaying the highly cyclical risks from manufacturing capex, depreciation, inventory, spot pricing, and supply expansion. In addition, the 3-5 year replacement cycle itself may simply define a “hardware procurement cycle” and does not necessarily eliminate cyclicality.

The comment section had a fair amount of discussion, but the quality was clearly split. Many users focused on the point that the “AI-generated DD is too long and too low-effort.” One of the top-upvoted comments directly mocked the poster by saying that if the poster was too lazy to write it themselves, others would be too lazy to read it, showing strong community dislike for AI slop. In the substantive discussion, the most influential bearish comment was a long, highly upvoted rebuttal arguing that the $88B net income assumption is a hallucination-like extrapolation: Micron’s net income at the peak of the 2018 memory supercycle was about $14B, so even if FY2026 Q2 was a record quarter, annualizing a high-cycle quarter and assigning it a long-term tech-giant multiple is far too aggressive. That comment also noted that a 3-5 year lifespan does not eliminate the cycle; it instead defines the hardware purchasing cycle, and hyperscalers may enter a digestion phase after filling data centers in one large wave. It also argued that enterprise capex is not inherently stable, and if macro conditions weaken or AI ROI falls short of expectations, large-customer orders could fluctuate even more violently than consumer demand. Supporters emphasized that AI demand for memory, especially HBM, is a real bottleneck; inference performance is constrained by memory bandwidth and capacity; and hyperscale customers are locking in high-priced supply through long-term agreements. They argued that rising memory prices do not necessarily suppress demand, because memory may remain a key component for improving AI performance per dollar. The OP personally responded multiple times: in response to the claim that the cycle only has one quarter left, OP asked the other person to explain the basis for that view and said they were looking at longer-term valuation; regarding the risk of alternatives to HBM, OP argued that if a more effective solution emerges, the major memory manufacturers are also likely to participate in discovering or producing it; in response to the rebuttal that enterprises will find ways to reduce memory costs, OP argued that developing in-house alternatives or bypassing memory manufacturers is not easy, is costly, and takes a long time, and that tech giants are more likely to focus on their own strengths and pass costs on to customers; regarding the AI bubble, OP said that earnings from companies such as Google have already proven that AI can make money, and argued that physical construction may slow, but spending on chip maintenance and replacement will not disappear. The rough consensus in the comments was that AI memory demand and Micron’s recent results are very strong, and that supply tightness is real. The main disagreement is whether this is enough for memory to permanently escape cyclicality, and whether MU deserves a valuation multiple close to large tech or NVDA. There were also many memes and sarcastic comments, such as “Short MU and DRAM it is” and “how high could Micron’s stock go if it made electric vehicles,” whose real meaning was mostly to mock the poster’s overly linear price-target extrapolation rather than present serious investment arguments.