The era of AI is booming, and recently the US stock market's storage sector has been talked about every day, with NAND, DRAM, and HBM flying everywhere. Do you all know what they are? 😂 Here's a popular science post to boost your knowledge 1. NAND Pronounced /nænd/, roughly like "Nan-de" in Chinese. Full name NAND Flash, which is flash memory. Its main feature is that data is not lost when power is off, providing persistent storage. It's everywhere around us: SSD solid-state drives, USB drives, memory cards, the 256G or 1TB internal storage in phones—all of these use it. Large SSDs in enterprise servers also use NAND. SanDisk's core asset is NAND; AI datasets and vector databases ultimately rely on it. In short, it's used for storing data. 2. DRAM Pronounced /ˈdræm/, roughly like "Dram." Full name Dynamic Random Access Memory, which is the running memory. All data is cleared when power is off, but it's much faster than NAND. Corresponding products include computer RAM sticks, 12G/16G phone RAM, and server memory. Temporary data during GPU computations is stored in DRAM. In short, it's used for running processes; it doesn't store data long but runs fast. 3. HBM No need to memorize the pronunciation; just say the letters one by one. Full name High Bandwidth Memory, a high-end upgrade of DRAM. While regular DRAM chips are laid flat on the motherboard, HBM stacks multiple layers of chips vertically, maximizing bandwidth. High-end AI GPUs like Nvidia H100/H200 and AMD MI series come standard with HBM. Note that HBM is not a separate category; it is still part of the DRAM family, just a high-end form. Production mainly relies on three companies: Micron MU, Samsung, and SK Hynix. Finally, just remember roughly: NAND is for storage, DRAM is for running processes, and HBM is a special fast-running memory for AI. Is everything clear now? #AI 🦧 #USStockMarketKnowledge

Is Puzi launching a coin today? Are you waiting for it~ Last time when Trump was launched, it directly drained the entire financial market Then everything crashed, and the market was bearish for a year

The Bitcoin bull market is here. Will there be another surge in Bitcoin ecosystem narratives? Can all the new protocols like inscriptions, small images, rgb++ finally start working? It’s been held back for so long; it’s time for a decent rebound. The Bitcoin assets in my portfolio have really been rusting away for too long~ Back then, it was all kinds of monsters and demons rushing in, but looking back now, I was really stupid—knew how to buy but not how to sell!

Wake up, happy weekend. What is everyone doing today? Today I’m taking my family out to have some fun~

Very interesting, the pricing of large models between China and the US has diverged sharply this round. On August 21, OpenAI officially announced that the API and credit pricing for GPT-5.6 Sol will be reduced by more than 20% over the next three months. Domestically, it's the opposite: DeepSeek leads the price increase, followed by Zhipu, Kimi, and MiniMax, with Morgan Stanley reporting that the average API input price for domestic models rose to 4.9 yuan per million tokens in Q2, compared to 3.3 yuan in Q1 2025, and the output price has increased even more to 21.9 yuan. My first reaction was that I read it wrong, but after double-checking, it's correct. The key is the main force behind the price increase: independent model vendors are leading the rise, while large companies with their own computing power remain inactive or adjust prices covertly. To put it plainly, the computing power cost is fixed; independent vendors can't sustain low prices to gain market share and have to raise prices; OpenAI, on the other hand, is proactively lowering prices to capture volume. The intentions on both sides are different. This round of price increases looks like a correction, but in fact, it's just that independent vendors can't hold on any longer. #OpenAI二季度营收67亿美元，亏损扩大