Following Jesus ✝️ | Family first | Navy veteran ⚓ | Investing | AI & data centers | Building Vestiq Finance | Learning AWS & Python | NFA
Joined January 2013
- Tweets64.7K
- Following1.2K
- Followers3.4K
- Likes75.4K
Even OpenAI can't get enough compute, So why is everyone so bearish ?? $IREN $NBIS xitter.gq/wallstengine/status/21…
Apple used to squeeze memory suppliers. Hard to do when supply's booked through 2029. xitter.gq/firstadopter/status/21…
Scummy moves by Apple. Geez.
WSJ: "The company had another hardball tactic: Avoid commitments. After ordering flash at rock-bottom prices, Apple would refuse delivery at times if flash prices were falling. That left suppliers holding piles of unsold chips, including ones custom-made for Apple. Kioxia executives joked they felt like Charlie Brown trying to kick the football—with Lucy always pulling it away." wsj.com/tech/apple-iphone-me…
I hope shipping is included
A 300-rack AI supercomputer powered by 9,600 Nvidia A100 GPUs is listed on eBay for $59.7M.
That’s 1,200 GPU compute servers with 384TB of combined GPU memory and roughly 1.4 petabytes of system RAM.
Each compute server connects eight A100 GPUs through $NVDA's NVSwitch technology.
More than six years after A100 launched, The seller is offering these systems refurbished for new AI training and inference deployments.
If Marvell buys Lumentum, as Bloomberg reports it's in advanced talks to, Marvell brings its laser supply in-house. That's an ~$83B optics company joining a ~$221B chip company, and remember Nvidia put $2B into Lumentum in March.
But now the question is, will customers who don't want to buy from Marvell shift orders to $COHR and $AAOI?
Memory is not cyclical. Look at the profits.
SK hynix just did ₩83.3T of operating profit at an 80% margin. Samsung's chip division did ₩106.5T, up from about ₩7T a year ago.
Samsung still fell as much as 6.2% today. How long can the stocks ignore this? $DRAM
China's answer to the HBM shortage reportedly fails spec on most of what it makes.
CXMT plans limited HBM3E production within ~2 months at a reported 20-30% yield vs ~70% at SK Hynix, and its samples tested closer to HBM3.
Q2 HBM revenue share is still SK Hynix 50%, Samsung 33%, Micron 18%. $MU $SKHY $DRAM
TSMC is paying NT$38.8B for AUO's Taichung display plant, about 429,000 square meters with cleanrooms already in place.
Buying a finished shell skips years of permits and construction, which says TSMC wants new capacity sooner than it can build it. $TSM
More on this:
• HBM4 revenue increased more than 3× QoQ
• HBM4 reached 60% of total HBM revenue
• 2027 HBM supply is sold out at Samsung’s target prices
• Customers are still requesting additional volume
• 2nm foundry wins are expected to more than double this year
Samsung is converting demand into higher HBM sales, stronger pricing, and more advanced-node orders.
$DRAM $SKHY
Samsung is locking up memory capacity through the end of the decade.
In July, it had signed long-term supply agreements with 5 major customers and was negotiating with 5 more.
Tonight’s update:
• 10 customers signed
• Another 10 in negotiations
• 60–70% of total production capacity could be covered by LTAs through 2030
Samsung also expects the memory shortage to worsen in 2027 and remain tight through 2028.
Customers aren’t waiting for spot supply. They’re reserving capacity years in advance.
$MU $SKHY $DRAM 🔥
Samsung is locking up memory capacity through the end of the decade.
In July, it had signed long-term supply agreements with 5 major customers and was negotiating with 5 more.
Tonight’s update:
• 10 customers signed
• Another 10 in negotiations
• 60–70% of total production capacity could be covered by LTAs through 2030
Samsung also expects the memory shortage to worsen in 2027 and remain tight through 2028.
Customers aren’t waiting for spot supply. They’re reserving capacity years in advance.
$MU $SKHY $DRAM 🔥
Google started using third-party compute at the end of Q3 and will ramp it in Q4 due to supply constraints, bullish for neoclouds. $IREN
$GOOGL on call: "Given the supply constraint environment, we started utilizing third party compute capacity at the end of Q3 and will ramp usage in Q4. We expect margin headwinds in the near term as we utilize this capacity. Moving to investments, our full year 2026 CapEx guidance range of 195 to $205 billion remain unchanged. We continue to expect our CapEx to increase significantly in 2027, and will provide further details on our Q4 call."
Microsoft's light capex quarter looks like timing, not demand.
$45.2B vs ~$50.7B expected, but Q2 is guided to over $50B and the calendar year to ~$170B. H2 accelerates on CPUs (including higher memory and storage costs), GPUs, and 3P capacity that slipped.
$CRWV and $NBIS fell ~5% after hours, $IREN ~3% most likely because of the capex news. Does the delayed capacity actually land in H2?
MICROSOFT $MSFT just reported earnings.
Earnings per share reached $4.89 🟢.
Revenue totaled $92.3B 🟢.
Zeek | Insights 📊 retweeted
Remember what Elon said exactly “Debt markets have almost Entirely shutoff”
Seems like $IREN will be just fine.
This is why secured power is becoming one of the most important parts of the AI trade.
The companies that can secure and deliver power may have a real edge as the grid struggles to keep up.
$IREN $CIFR $WULF are some companies to keep an eye out for.
Why a Marvell x Lumentum deal makes sense:
AI can’t scale on compute alone. It needs faster, more power-efficient connections between chips and optics is becoming a critical part of that infrastructure.
Marvell brings high-speed connectivity silicon. Lumentum brings optical components and laser technology, with demand already running ahead of capacity.
Together, they could give Marvell a more complete AI networking offering and put it closer to a supply-constrained part of the buildout. $MRVL $LITE
Alphabet reported $30.7B of Google Cloud revenue, up 103% YoY.
TPU system sales are now a meaningful contributor, while quarterly capex reached $51.4B and free cash flow fell to negative $9.1B
Google is no longer building TPUs only for itself. It is turning them into an external infrastructure business while spending aggressively to add more capacity.
Equinix reported the same demand from the data-center side: record bookings and nearly 40% more cabinet deliveries than originally planned for 2026.
AI infrastructure demand is still running ahead of available capacity. $GOOGL $EQIX
Oracle expects positive free cash flow in H2 FY28 and $30B+ in FY29, while planning up to $20B in FY27 funding.
That’s a huge AI infrastructure buildout. The next question is who’s adding the compute to meet that demand and whether they can make money doing it.
$NBIS reported $236M in Q2 adjusted EBITDA, with AI Cloud at about $286M and a ~50% margin.
$IREN reported $246M in FY26 adjusted EBITDA. Its AI revenue is still ramping as it brings capacity online.
Oracle, Nebius and IREN have different business models, but they’re all tied to the same AI compute buildout. $NBIS $IREN 👀
