Tracking the AI buildout through chips, GPU demand, HBM, data-center capex and earnings. Numbers over narratives. No stock calls.
- Tweets 596
- Following 1,194
- Followers 462
- Likes 0
Would you trade 2-5% pricier homes for 100-200 jobs?
1,500 data centers, no local wage gain: Survey says host communities bore power and subsidy costs without measurable wage upside. AI capex favors asset owners over workers. Would your town accept it?
3
5
32
Why aren't cheaper AI tokens crushing old GPU values?
Token costs fell from $2+ to ~$1/M, while H100 residuals rebounded toward $30K and A100s held near $18K. Demand is cascading down the GPU stack—not exiting. Still valuing GPUs like servers?
2
4
26
Would you pay nearly 90x earnings for Marvell's AI ramp?
Over 50% of Marvell’s revenue now comes from data centers. Its valuation prices in an on-time H2 FY2027 custom-silicon ramp; even one customer delay could raise execution risk and compress the multiple.
1
22
What if Nvidia's strongest moat is measured in watts?
$80B run rate by October: Morgan Stanley’s Nvidia Rubin estimate. The bet is agentic AI; power-constrained data centers should favor more compute per GW. Is that edge durable?
16
What would justify 2026's $800B AI capex spree?
AI services may require spending equal to 9% of GDP—about what the economy spends on food and 2x energy. If customer demand drives AI infrastructure buildout, is that end-market scale plausible?
15
What does an $820B AI buildout hit first?
$415B-$820B in hyperscaler capex vs $800B-$1.4T+ for US utilities: Blackstone’s Jon Gray says available power may constrain AI first. The infrastructure question: generation or transmission?
1
12
What if AI's real bottleneck isn't chips, but transformers?
AI's hidden grid play: High-voltage unit lead times are 104–156 weeks. Atlanta Electricals has 63,060 MVA capacity at 30–35% utilization; more orders could lift margins without new capex.
1
2
17
Where does every $100 of AI buildout money actually go?
2027 data-center capex is projected at $1.3T. Only $50 of every $100 goes to chips; the rest funds power, networking, cooling, buildings and land. AI infrastructure economics reach beyond semis.
10
Could AI make your next phone ship with less RAM?
HBM could consume 30% of global DRAM wafer capacity by 2027, up from ~20% today, Samsung says. AI data-center demand may squeeze mobile-memory supply, making 16GB phones pricier and favoring 12GB.
7
How much of America's investment-grade debt is now an AI bet?
$600B: KKR says AI-linked debt is 6.3% of US investment-grade credit and could near 20% by 2030. Guarantees and leases mask exposure. Rational buildout or concentration risk?
2
2
15
What's the AI bottleneck already sold out for 2027?
2027 HBM capacity is nearly sold out, Micron says, at prices far above last year. With new cleanrooms not online until late 2028, AI capex faces a memory bottleneck it can't fix. Who's pricing that in?
1
2
15
What could stall the AI capex boom before demand does?
5% Treasury yields mean 7–8% project funding costs, topping 10% for weaker borrowers. Every data center must clear that bar. GPU orders aren't enough—watch ROIC. Which AI projects still pencil out?
1
2
16
What does Modine look like after cutting $1.1B of revenue?
$MOD data-center cooling grew 90% last quarter vs. 22% for commercial HVAC. The spin-off sheds an auto/industrial unit shrinking 3%, sharpening AI exposure—and sensitivity to peak capex.
3
4
28
What matters more than Micron's 379% revenue surge?
$12.7B in customer prepayments secured future memory supply this year. Buyers fund capacity pre-delivery as AI demand hits constraints, signaling supplier capex. Durable demand or peak-cycle caution?
3
7
41
Why did an unchanged GPU become 3.5x more expensive?
NVIDIA B200 pricing jumped from $2.31 to $8.11/GPU-hour in 7 months. HBM shortages and strong inference demand made capacity scarce—AI pricing may hinge more on allocation than speed. B300 next?
3
4
34
What if India's biggest AI winners never build a model?
20 of Goldman’s 42 AI enablers span power generation/transmission. Key point: grid capacity is the bottleneck—data centers need reliable megawatts before earning a rupee. Who has the strongest moat?
1
1
26
Where will AI find $4.2 trillion in new annual revenue?
AI needs $6T in annual revenue to support $1.5T/year in infrastructure by 2031, Bain says. Consumer and enterprise AI may bring in only $1.2T–$1.8T—a $4.2T+ gap. What economic value closes it?
4
4
41
Could Big Tech's AI capex already be covered by booked demand?
Goldman: Amazon, Microsoft & Google need ~$1T in 2028–30 revenue to earn 15% ROIC on AI capex. Their ~$1.7T backlog is 1.7x the hurdle. What breaks first: conversion or margins?
5
6
33
What can fill AI's $4.2T revenue hole by 2031?
$6T/year: Bain’s revenue target to justify current data-center spending, up from $2T last year. Existing AI could deliver $1.8T; robotics, drug discovery and other bets must close the $4.2T gap. What scales?
3
4
26