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

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

SpaceX $SPCX closed at $134 yesterday, down over 4%, directly falling below the $135 issue price, with a market cap remaining at $1.77 trillion. The reason for the sell-off is simple: the lock-up period is still releasing shares continuously, and the entire arrangement won't end until 2027, with about 88% of the total 13 billion shares gradually unlocked. To put it plainly, the unlocking is an open secret; the market has long known that the shares would slowly come out. The fact that it still broke below the issue price means the current sentiment is weak. Looking at the promises made also brings a smile 😂: Trump just signed a memorandum paving the way for 1,000 launches per year by 2030; Musk is simultaneously claiming a 20x revenue increase by the late 2030s while promising to buy 220,000 Nvidia superchips. The promises are big, but the accounts haven't been realized yet, and the Starship recovery has been delayed by several months. My take: The unlocking was clearly laid out on the table; breaking below the issue price is the real thermometer of sentiment. This kind of stock will swing with news in the short term, so don't catch a falling knife or rush to sell. Wait for the shares to be fully absorbed before making a move 😂.

The same question: Can you combine an mp3 and a picture together as a video synthesis? Let's see how domestic large models and foreign large models handle it. Here, domestically they use: Zhipu and Doubao, abroad they use Gpt and Gemini The foreign ones are still impressive, they can actually do it, providing the materials and handling everything for you throughout. The domestic ones just can't do it, they give you the method and you have to figure it out yourself. 😂 How to explain this?

Wow, the Dow was slammed by Walmart for over 700 points last night, while Nvidia only dropped 0.33%, closing at $216.85, the most resilient on the floor. After touching 225 on August 13, it fell for three consecutive days, but the decline narrowed each day (-2.34% → -0.99% → -0.33%), indicating selling pressure is exhausting. Everyone is now waiting for the end-of-month earnings reports, with the market focused on one thing: how strong the guidance is. Just looking at Walmart, you can tell that beating revenue expectations doesn’t help if growth and guidance are weak—it still gets slammed. News flow hasn’t stopped; Jensen Huang’s daughter appeared at the Beijing Robotics Conference, physical AI is being laid out, and Google has also confirmed the next-generation liquid cooling solution. From a technical perspective, 215.7-216.2 is the short-term support; breaking below that points to 209-212. If it climbs back to 222-225 before earnings, it means someone is running ahead, and the actual report might easily disappoint. My take: if volume shrinks and it trades sideways before earnings, don’t mess around. Betting heavily on a one-sided move is just throwing money away. Wait for earnings to provide direction. This is not investment advice. #沃尔玛在美销售放缓，消费压力受关注

Bull run? Why did the market suddenly surge? BTC rose 11.7% in 24 hours, hitting $72,000, while ETH was even stronger, up about 20% to $2,300. The direct trigger was Trump meeting crypto executives and regulators at the White House, stating: regulation is needed, but it must not strangle the industry; this industry's importance may surpass the internet. As soon as he spoke, concept stocks soared first, with Strategy rising over 13% at one point, and Coinbase up nearly 10%. Looking at the bigger picture, the US Treasury has expanded bond repurchases in recent days, with Bassett stepping in to stabilize the bond market, easing liquidity expectations. Gold, silver, and Bitcoin all rose together, with gold surpassing $4,500. What the market lacks is not money, but certainty about regulatory direction. Trump's statement effectively changed the question from "Will there be regulation?" to "How will it be regulated?" The result is a textbook short squeeze: 180,000 liquidations in 24 hours, with liquidation amounts exceeding $3 billion, over 90% of which were shorts. CoinGlass says this is the largest short liquidation wave since 2021. The faster the price rises, the faster shorts die; liquidated positions then turn into buying pressure, accelerating the spiral. #BTC突破72000美元，本轮上涨能否延续？