AI capex strength keeps NVIDIA and chip suppliers in focus
📈 Growth Stock Playbook
-
Growth stocks are best owned when they look expensive and the thesis is still doubted. When everyone starts calling them cheap, the easy alpha is usually gone. 1
-
Bottom-calling is a trap. Trying to pretend the exact low is knowable leads to bad execution; the smarter play is to avoid overfitting entries around a fake floor. 2
🤖 AI Infrastructure & Semiconductors
-
NVIDIA still looks underpriced after a long period of mispricing. The next wave of Big Tech earnings should be a tailwind, because the likely message is continued capex, and most of that spend still flows toward NVIDIA. 3
-
Not all LLM demand is equal. The key capex/RPO engine for major cloud vendors is still OpenAI and Anthropic. If revenue growth at either one hesitates, high-capex AI infrastructure names can get hit hard again. 4
-
Cheap semis are everywhere, but cash is being reserved for higher-momentum names. Valuation alone is not enough when stronger tape leadership exists elsewhere. 5
-
MRVL should not be compared with its tariff-war-era fundamentals. Back then EPS was roughly 1-2 dollars; current next-12-month consensus EPS is above 6 dollars. The setup is no longer the same business profile. 6
-
The AI capex cycle still has legs. TSMC and ASML raised guidance and kept leaning into capex; Meta lifted capex to nearly 2x last year; even Buffett built a large Google position for AI infrastructure exposure. 7
💾 Memory Cycle
- 3Q DRAM ASP is expected to rise 21% QoQ, matching channel checks and sitting above TrendForce assumptions of 13-18% QoQ for conventional DRAM, or 8-13% including HBM. The memory cycle looks stronger than consensus framing. 8
🛢️ Geopolitics & Energy Risk
- The US-Iran conflict should not be treated as a quick “TACO” headline. Energy-market risk is stickier than the market wants to price. 9