Looking at $SAMSUNG and $SKHYNIX , I’m more inclined to see today’s price level as the end of the downtrend; the chances of another sharp move lower are not high. Based on how I read the chart, today marks the end of this decline. People who look only at RSI may say it needs to break below the lows for confirmation, but my view is that a much deeper decline from here is less likely.
I first noticed $Lobster at the end of February, when that Binance Chinese-language post came out. The community jumped on it immediately, and it surged to over 150% in just a few hours. I underestimated it, thinking it was just a short-term trade. But after pulling back 93% from its 200x high, it rebounded 50% today—more than 200x from its lowest point. My first judgment was wrong: I underestimated the power of cultural memes.
I saw HSBC slash its price target for $ZHIPU by two-thirds, citing intensifying competition, soaring R&D costs, and tougher anti-distillation measures from major US AI companies. Another blow, underscoring just how dependent Chinese AI startups are on distillation. $ZHIPU probably isn’t an exception.
I don’t think Soros Fund’s Q3 increase in $INTC was just a token move: they went from 100,000 shares straight to 2.83 million, a clear jump in their position. That scale of increase shows they were genuinely raising their allocation to Intel in the third quarter, not just testing the waters.
$SNDK When this wave of Q3 13F filings came out, a lot of people said, “The big players have abandoned it.” But I noticed a few things that almost nobody mentioned. First, the buying: more institutions bought than sold—952 bought in, while 645 sold; 283 opened new positions and 805 added to their holdings, while 514 reduced their positions and 239 exited completely. But on the other hand, the number of institutions holding $SNDK fell from 1,835 to 1,544, and their combined holdings also dropped from 111 million shares to 8,540. So don’t jump to conclusions based on the headline alone. The shift in the underlying mix is more complicated than the word “abandoned” suggests.
I think this cyclical stock has already reached the top of its cycle; its P/E ratio has already been pushed above 5x, and it really is expensive. With capital expenditures that high, I wouldn’t be surprised at all if it’s losing money two years from now. Also, it is indeed lagging in HBM, and its EUV expansion has only just started, exactly the opposite of $SAMSUNG and $SKHYNIX .
The market looks pretty divided today. $VIX surged nearly 19 points, and as volatility picked up, investors started getting selective. Airlines and consumer stocks were among today's winners. $LUV rose more than 3%, and Starbucks was up around 2% as well. On the other hand, some sectors took a real beating. Satellite communications stocks were broadly lower, and memory chip stocks fared even worse: $SNDK plunged more than 8%, with its peers in the sector getting hit too. In this kind of split market, would you catch a falling knife or chase the strong performers? What do you think?
Tencent has released Hy4 Flash: a hosted model with a million-token context window, and input costs just $0.07 per million tokens—about one-twelfth the price of the full version of Hy4. The price floor is really getting pushed lower and lower. $TENCENT With prices like these, will the way you build things change? I’ll definitely recalculate the costs: I can now comfortably put things in the context that I wouldn’t have dared to before.
That drop in U.S. stocks last night was brutal. $NVDA is pulling this again right before earnings. The Nasdaq plunged 2.5% in one fell swoop, putting it back in correction territory. The news of Jensen Huang stepping down hasn’t even been digested yet, and now the U.S. government has announced new restrictions on China’s access to the most advanced AI semiconductors. The Philadelphia Semiconductor Index promptly crashed 5.3%. $NVDA fell 4.1%, $INTC and $MU both dropped more than 7%, and Dell was down 7.0% too—a rout across the board. With Nvidia’s earnings on the 17th and Anthropic’s massive IPO on the 19th just ahead, investors are heading for the exits. Risk aversion is through the roof.
I took a look at the gainers list: $ASTER was up 5.2x, $PIEVERSE was up 3.9x, $FF was up 3.8x, and several other altcoins were strengthening in tandem. But unusual price moves accompanied by a surge in volume are only a starting point, not a signal that’s been validated. Whether the rally can last and whether the data is reliable are what really matter going forward.
Hy4 flash is so cheap now that you barely need to worry about the price, but you shouldn’t skip validation. Tencent’s listed prices are $0.07 per million input tokens and $0.281 per million output tokens. For 1,000 tasks, with 20K input tokens and 2K output tokens each, flash costs $1.96, while full Hy4 costs $21.68. My approach is to run 50 of my own real tasks through both first, and switch to flash only for cases where its output passes. With the price difference of $TENCENT right there, the savings come from validation.
U.S. tech stocks got hammered right out of the gate at the start of the week, and the whole picture looked brutal to me: the S&P 500 Information Technology Index fell 2.32%, while semiconductors took an even harder hit, leading the declines with a 3.75% drop. $MRVL -7.6%, $MU -6.6%, $NVDA -2.0%—the chip sector was really hurting today. Hardware fell 1.44%, software dropped 1.01%, and many stocks followed suit. My take is straightforward: this isn’t an isolated crash; the entire tech sector is cooling off, with semiconductors taking the hardest hit first.
I took a look around and sorted them by relative volume. ADRs above $3 really aren't seeing much volume right now—not a single one has relative volume above 1x. $MSTR and $COIN are leading the declines, and there's selling pressure in storage stocks too. For $NVDA , we'll have to see how it opens tomorrow. I won't go through the rest of the stocks one by one. Overall, they're much the same; I haven't seen any that can get volume going first.
Let me pour some cold water on this: don't get swept up in the first-day hype ahead of Anthropic's IPO on Thursday. A massive day-one surge is very possible—that's how this year's $SPCX and $SKHY IPOs played out. But the subsequent drops are just as real, and losing value is no joke.
I saw $SKHYNIX lose more than $1B in market cap within minutes, while the $KOSPI slid as much as 3% and $SKHYNIX itself fell 5.8% compared to Samsung’s 3.8% drop. Foreign investors sold over ₩700B by 10:27 am KST, and the pressure was driven by liquidity concerns ahead of Anthropic’s IPO.
I think the easiest way to misread $MSTR being dropped from MSCI is as a crypto sell-off. Passive funds sell its stock; the company doesn’t have to move a single coin. What changes is the premium, not its holdings.
Buybacks can't save the ending. $SNDK and Hynix's buybacks are cautionary tales. When a company starts buying back shares, what it really means is that it has no new projects to work on, isn't investing further, sees no growth ahead, and has nothing else going for it. With AI so desperate for cash right now, instead of using that money to keep the revolving credit going, you're spending it on buybacks? That's just like drinking poison to quench your thirst—it'll only speed up your demise. I'm sure experienced Wall Street analysts hear news of a buyback and hit the sell button immediately.
$SKR I took a look at the 1-hour chart. The price has just broken above 0.010184 on a surge in volume—the candle’s volume was 4.48 times the usual level—and it’s now approaching the 0.011361 supply zone. Although the daily structure is still bearish, I’m bullish on this 1-hour impulse move.
Looking at $MRVL , Marvell makes chips that process data in optical connectivity, as well as custom chips for hyperscale cloud providers. Every time optical connectivity gets faster, signal processing needs new chips, and Marvell is well positioned to benefit from this upgrade cycle.
That’s the thing about publicly listed companies: one wrong step, and it’s hard to get back on your feet. I think the biggest problem with $ZHIPU , Minimax, Kimi, and the others isn’t how they’re doing right now, but what they’d have to fall back on if things go wrong. That 1GW data center from $ZHIPU is, frankly, still just a “project.” Have they actually secured the GPU allocation or not? That hasn’t been made clear. DeepSeek, on the other hand, has real computing power allocations, so it’s on much firmer footing than they are.