When I was in Shenzhen a few days ago, I was chatting with some friends in the AI industry. Why hasn't DeepSeek heard much lately? Some say they need to keep a low profile because they have taken money from state-owned assets for financing, while others say they are holding back a big deal: at the same time as Apple announced a price increase across the board, DSpark came out like a savior: How to do it? After reading the paper, the solution has the consistent style of DeepSeek, which is mainly based on the same model and the same hardware. By improving the reasoning efficiency, the throughput can be significantly improved. Yes, it's just a systematic engineering optimization integration. The optimization details are included in the paper, so we won't delve into them, but is it worth exploring whether this is beneficial for the storage industry? 1) The transmission logic behind the price increase of hardware consumer goods such as Apple is due to the tight production capacity of HBM high bandwidth memory, and the price increase is too high for the production cost of Apple hardware devices, which is then passed on to consumers for price increases; However, the soaring cost of memory is due to the massive demand for HBM in AI data centers, which has led manufacturers such as Samsung, SK Hynix, and Micron to focus on high margin HBM wafers. This directly leads to a shortage of DRAM and NAND supply on the consumer end. This means that if the government implements subsidies or industry coordination policies in the future, it may alleviate the impact of the structural tilt in production capacity, as the government is sensitive to the inflation risk of rising prices of electronic consumer goods; 2) DSpark optimizes from the inference end engineering, allowing the same HBM memory resources to support more concurrent requests and higher token throughput, which directly alleviates the supply side demand for HBM memory in the short training period of large models, thereby reducing the transmission pressure of memory shortage to the electronic consumer end. Logically, make sense, but it can't solve the problem in the short term, because the problem of HBM memory shortage is now mainly the strategic needs of big factories such as capEx capital expenditure, which itself has a strategic foam component. Simply relying on reasoning and optimization to hedge the large-scale hardware demand generated by capital expenditure will have limited effect; The above. Overall, DeepSeek did showcase its technology at a critical juncture, but in the short term, the contradiction between production capacity and price in the storage field is very limited. In the long run, it is indeed a valuable exploration direction.