Post

#闪迪投资者日后，长期目标成焦点 80% gross margin, making NAND. You read that right. This is the long-term goal given at SanDisk Investor Day: FY28 to FY30, non-GAAP gross margin 80%, operating margin 75%, free cash flow margin 50%, mid-to-high double-digit revenue growth, and all remaining money distributed to shareholders. Goldman Sachs said "far exceeds expectations" after reading it, the stock jumped 15% that day, closed at 1554, and kept rising after hours. But you need to know, its latest quarterly gross margin is already 84.6%. So 80% is not "something to strive to reach," but "something to figure out how to maintain." That’s the key. Here’s an interesting thing. Last week’s earnings report showed record revenue, gross margin 84.6%, far exceeding expectations, yet the stock dropped 8% after hours. This week, after a meeting discussing the three-year goal, it rose 15%. Where is the logic in this market? In the NAND flash business, gross margins have basically fluctuated between 20% and 50% over the past twenty years. At worst, 20% or even negative; at best, about 50%. What does 84.6% mean? It’s even more outrageous than Nvidia. Historically, such gross margins have never been sustained. Every 3-4 years there’s a cycle: shortage and price hikes, capacity expansion, oversupply, crash, repeat. After the pandemic, NAND capacity was oversupplied, and Samsung, Micron, and SanDisk all had negative gross margins. Also, look at neighboring Hynix $SKHYNIX. On August 6, SanDisk’s earnings guidance missed expectations, causing the memory sector to drop, and Hynix fell 10%. Then news came that Nvidia cut Rubin Ultra’s HBM configuration in half, from 12 layers 384GB to 8 layers 192GB, and the next day Hynix dropped nearly 5% again. That’s a total 15% drop in two days, wiping out about 180 trillion Korean won in market value. Hynix’s Q2 NAND revenue was about 21 trillion won, up 344% year-over-year, the strongest by far, but a single HBM order cut caused a collapse. What does this show? AI storage isn’t a monolith; the logic is completely different between GPU training and inference. $SNDK SanDisk is betting on the latter. NBM new business model, no spot sales, signing long-term agreements with customers, with minimum purchase volumes and financial guarantees. Already signed 8 customers, accounting for half the bit volume in FY27, two-thirds in FY28. Using contracts to lock in cycle fluctuations. There’s also demand-side factors: AI inference explosion, enterprise data center flash market expected to reach 1.2ZB by 2030, BiCS10 QLC density up 60% over BiCS8, HBF high bandwidth flash is pushing forward. While Hynix is still shaken by HBM order cuts, SanDisk has already pitched the story to the inference side. But I still say: the biggest trap for cyclical stocks is "this time is different." NBM customers are of course willing to sign when supply is tight, but in the next downturn, can minimum purchase volumes really be enforced? Also, on August 7, Hynix just approved 19 trillion won to expand NAND capacity, and the Dalian second plant is rumored to restart; SanDisk isn’t the only one expanding supply. I personally hold SNDK on Solana, trading 24/7. I was there for the 8% drop last week and the 15% rise this week — much more exciting than US stocks — US markets close but it keeps running; checking my phone at night might catch a full roller coaster ride. In the first week after launch at the end of June, on-chain trading volume hit $70 million; Solana now accounts for 97% of all tokenized stock trading volume on-chain, RWA is no longer just hype. Whether the long-term goals can be realized will only be known in two or three years. But the market is willing to pay for this story now, and that’s enough. Purely for discussion, not advice.