# AI semiconductor stocks — X 热门讨论 (2026-09-15 15:32 UTC)
## @nezih_gokay (Nezih Gökay) · 09-15 15:06 · ♥86 ↻4 💬0 A MAJOR CRASH IN AI CHIP STOCKS
The chip giants driving the AI infrastructure have been rocked by a severe blow. @DyamondBre is providing real-time updates on why Marvell took the hardest hit in this massive market crash and what critical moves it will make in the coming period.
If you’re investing in AI infrastructure or want to stay in the game, follow this account or you’ll be left behind. > 引用 @DyamondBre: $MRVL's 7.32% drop came from a much wider reset across high-multiple chip stocks.
The semiconductor ETF fell more than 5% as investors reacted to the AI pacing debate and another Fed decision week. $AVGO and $NVDA fell with the group, while Marvell had more accumulated gains to protect.
There was no fresh company-specific announcement behind the move. I am watching whether sector selling eases and whether custom-silicon orders keep converting into revenue. https://x.com/nezih_gokay/status/2099877440986267939
## @RedDogT3 (Scott Redler) · 09-15 12:50 · ♥36 ↻7 💬7 📺 RATES ARE AT 2007 HIGHS — SO WHY AREN’T STOCKS BREAKING? + $NVDA HAS A NEW TRADE + $MU GAP-FILL SETUP + KEY LEVELS TO WATCH INTO THE FED
The market is heading into the Fed decision with plenty of reasons to break lower — yet it keeps refusing to do it.
$SPY continues to hold above last week’s lows around $756.64-$757 and remains around its 50-day despite an increasingly difficult macro backdrop.
Rates are at their highest levels since 2007, $TLT remains under pressure, oil is near its summer highs, and AI sentiment has deteriorated. Yet traders who tried to aggressively press shorts yesterday weren’t rewarded.
That resilience matters. The macro picture may look bearish, but price still hasn’t provided decisive downside confirmation.
*
Remember the last Fed day? $SPY broke the lows, everyone thought the 200-day was next, and traders piled into shorts. Then there was no downside follow-through. The market regrouped and produced a fast and furious move higher over the next two sessions.
That’s the lesson heading into this Fed: don’t marry the first move. Let the market show you what it wants to do.
*
$NVDA is a good example of how quickly a trade can change in the current environment.
The previous setup was a potential move toward new highs, but the break below $229.63 confirmed that the chart had changed. Now this isn’t an all-time-high trade — it’s potentially an oversold bounce/gap-fill trade.
$NVDA is now around $212 and needs to get above and stay above $212.77. If it can, that could trigger a tactical gap trade toward the 50-day and potentially fill more of the overhead gap.
The key is always adjusting your expectations: this is a cash-flow trade against support, not necessarily the beginning of another major leg higher.
*
$MU is setting up similarly.
It’s bouncing and approaching an overhead gap, with roughly $931-$932 as the key trigger.
Get above $932 and stay there, and there could be a quick trade toward $940 as shorts get squeezed and algorithms chase the gap.
Get rejected at $932 and leave the gap open, and that tells you something about the underlying weakness.
*
I would watch $NVDA and $MU today together. If both begin reclaiming their gap levels, that would provide better confirmation for a tactical semiconductor/memory bounce.
*
$QQQ is another one to watch closely. It bounced from the $702.70 area, giving traders a clearly defined level against which to manage risk.
Key levels into the Fed: – $SPY: $756.64-$757 support – $QQQ: $702.70 support
*
So, this is not the environment to force big predictions or oversized directional bets. Bring expectations down, trade the levels, protect your capital and don’t jump over your skis ahead of the Fed.
If the first post-Fed move breaks down but gets no downside follow-through, be willing to readjust. If support genuinely fails and sellers finally take control, respect that too.
Rates, oil and AI headlines say stocks should be under much more pressure. The fact that they aren’t is information in itself.
Let price confirm the next move.
*
If you found this helpful, please ❤️like and 🔁retweet https://x.com/RedDogT3/status/2099843269345567222
## @MelvinInvests (Melvin) · 09-15 14:15 · ♥33 ↻6 💬6 This chart presents one of the clearest bull cases for the AI infrastructure trade (Save this).
Global operational data center capacity increased from 82 GW in September 2024 to 133 GW in June 2026.
Capacity under construction more than doubled from 20 GW to 45 GW, while planned capacity nearly tripled from 78 GW to 231 GW.
These numbers suggest that the buildout remains in its early stages.
Every additional gigawatt requires power generation, transformers, switchgear, cooling systems, networking equipment and specialized construction.
Now here are some of the stocks that will benefit from this.
Eaton should benefit from rising demand for electrical distribution and power management equipment.
Its data center orders increased approximately 85% year over year in the second quarter of 2026.
GE Vernova should benefit from the need for additional electricity generation and grid infrastructure.
The company received more than $5 billion in data center orders for its Electrification segment during the first half of 2026.
Vertiv and nVent should benefit because increasingly dense AI systems require more advanced power delivery and cooling.
Vertiv’s quarterly revenue increased approximately 24%, while nVent reported 47% organic growth.
EMCOR should benefit from installing the electrical and mechanical systems inside these facilities.
Its contracted backlog reached a record $17.14 billion, supported heavily by data center demand.
Broadcom should also benefit from demand for custom accelerators and the networking chips required to connect large computing clusters.
Its AI semiconductor revenue increased 221% year over year in its latest reported quarter.
Higher risk capacity owners such as IREN could benefit if they convert secured power into operating data centers and long term AI contracts.
However, planned capacity is not guaranteed capacity because many projects still need power, financing, equipment, permits and customers.
The strongest investments are likely to be the companies controlling the bottlenecks required to make these projects operational.
Eaton, GE Vernova, Vertiv, nVent and EMCOR are the clearest physical infrastructure beneficiaries, while Broadcom provides exposure to custom compute and networking.
Even if only part of the planned pipeline gets built, the AI infrastructure spending cycle could continue for years.
If you enjoyed reading this, make sure to follow @MelvinInvests for more AI infrastructure insights and if you want to see exactly what stocks I'm buying to play this buildout, you can join Milk Road Pro for just $1 using the link below.
https://t.co/MMkdjfqLsf https://x.com/MelvinInvests/status/2099864785806590267
## @0xmarginman (MarginMan) · 09-15 11:50 · ♥32 ↻5 💬4 🚨BREAKING: Trump phoned into Jensen Huang's live interview at the All-In Summit to declare the entire AI-safety debate "a hoax."
Monday, Los Angeles. Huang is mid-interview when his phone rings on stage. He puts it on speaker.
"The whole thing is a hoax." Trump, live, to the room.
Huang's answer seconds later: safety is critically important, but it isn't a reason to slow down.
The market didn't wait for the nuance.
NVIDIA closed down 3.4% that day. Semiconductor stocks fell across the board. The Global X Cybersecurity ETF, the trade for "AI goes wrong," jumped more than 10% in a single session.
Trump called it a hoax on a live mic. Traders priced it as a risk anyway. > 引用 @0xmarginman: 2017, Tokyo: Softbank puts $700 million into Nvidia.
January 2019: Masayoshi Son sells the entire stake for $3.3 billion. calls it a win. books the profit.
I checked what that decision actually cost before writing this. By 2024, that same 4.9% Nvidia position would have been worth north of $150 billion — not a rounding error, a nine-figure multiple of the money he actually took off the table. Son addressed it directly at a Tokyo shareholder meeting: "the fish that got away was big."
What makes this case different from a normal missed trade is the size of the fund doing the missing. SoftBank's Vision Fund exists specifically to make outsized, early bets on exactly this kind of company. Nvidia was already inside the fund. The fund had already won. It just closed the position roughly five years before the actual payoff arrived, trading a genuine 4.7x return for what turned out to be a 200x-plus one.
The actual lesson here isn't about Nvidia. It's that finding the right company is only half the trade. The other half — knowing when not to sell — is the part almost nobody talks about, including the investors built specifically to get the first half right.
↓ https://x.com/0xmarginman/status/2099828127299940357