The biggest revelation from MU's financial report and conference call: the storage industry has changed Today Micron submitted an almost impeccable transcript. Q3 revenue was $41.46 billion, far exceeding market expectations of $35.59 billion. The adjusted EPS was $25.11, far exceeding the market expectation of $20.63. Even more exaggerated is the Q4 guidance: Revenue of $49-51 billion, market expectation of $42.9 billion. EPS of 30-32 US dollars, market expectation of 25.5 US dollars. This is no longer a beat, it's a direct blow to the Wall Street model. The phone conference revealed three core changes that will have an impact on the coming years. Firstly, storage has shifted from a cyclical product to a strategic resource The CEO said a sentence that left a deep impression on me: 'The performance of an AI system essentially depends on the performance and capacity of its memory subsystem.'. ” With the development of Agentic AI, multi-agent systems, and ultra long contexts, memory is becoming a bottleneck for the entire AI infrastructure. GPU is responsible for computation. Storage is responsible for memory. Without memory, even the strongest brain is useless. The CEO of MU even directly stated that AI has elevated memory to the position of a strategic asset. This is a very significant change. Strategic assets mean that customers don't just look at prices. And priority will be given to supply capacity, safety, and long-term cooperative relationships. Secondly, the tight supply of DRAM and NAND may continue beyond 2028 The management has made it clear that "we currently cannot see when memory supply can catch up with demand." They even further pointed out that it is expected that the tight supply and demand of DRAM and NAND will continue beyond 2027. Why? Because expanding production is too difficult. The construction period of new wafer fabs is long. Advanced nodes are becoming increasingly complex. HBM is consuming a large amount of DRAM production capacity. NAND manufacturers have even started transferring some cleanroom resources to DRAM production. What does this mean? Many people feel that the DRAM price cycle is coming to an end. MU is telling the market that this may not be an ordinary cycle. It is an industry change where the supply side is permanently restructured. Thirdly, the future demand source is no longer just AI servers The phone call gave a bigger story. Data Center: Server shipments are expected to increase by more than ten percentage points in 2026. Agentic AI is driving synchronous expansion of CPU cabinets and storage cabinets. AI is not just about increasing GPU demand. It is also increasing storage demand. Cars: The average storage capacity of L2+and above models has reached more than 5 times that of ordinary cars. This year, the proportion of L2+models will exceed 20%. It is expected to exceed 40% by 2030. Robots: The average storage capacity of humanoid robots is 10 times that of L2+cars. MU CEO directly stated that it is expected to initiate a storage demand cycle that will last for decades in the second half of this century. Additionally, there may not be sufficient pricing information in the market yet: Micron announced that it has signed 16 Strategic Customer Agreements (SCA). These agreements cover data centers, consumer electronics, and automobiles with a maximum term of 2030. More importantly, these protocols adopt the Take or Pay model. Simply put, customers must buy, and even if they don't buy, they have to pay. At present, these agreements have been locked in: about 20% of DRAM capacity and about one-third of NAND capacity. This means that Micron is gradually transforming its highly cyclical business model into a more long-term contract model. That's also why the CEO emphasized that MU is undergoing a business model transformation. Finally, let me share my opinion In the past, people used to think of the storage industry as cyclical stocks. Raising prices to make money. Falling prices result in losses. go round and begin again. But the AI era has changed this logic. HBM has increased the unit value, while Agentic AI has increased storage requirements. Autonomous driving has increased the demand for onboard storage. Robots have opened up a whole new terminal market. However, the supply side is limited by advanced processes, production capacity construction cycles, and capital expenditures. When demand grows exponentially and supply grows linearly. Periodic products may gradually become scarce resources. So I think the most important conclusion of this financial report is that MU has proven one thing to the market: storage in the AI era is becoming one of the most important and scarce assets in the entire AI infrastructure. That's also why I have always placed DRAM at the first level of AI investment logic. Storage comes first, optical interconnection comes second, and applications come third. At least for now, I still haven't seen any reason to change this judgment.