doug o'loughlin

T3 · host / generalist

SemiAnalysis analyst covering semiconductors, memory markets, AI infrastructure, and data center buildout dynamics. Returns to podcast after busy period working on deep-dive models and industry analysis.

5 calls·5 names·0% bull·last heard 13 days ago·SemiAnalysis
track record

no scored calls yet — needs a stated position or a categorical verdict, with a matured window vs SPY

top calls

highest conviction · one per company
1sthigh conviction
$OPENAIOpenAI

Final pre-train IPO funding round risks $5T capex vs $500B revenue timing mismatch

Frontier labs are racing to final pre-train models to IPO and raise final mega-rounds; but even if revenue hits $500B, a $5T cumulative capex implies 10-year payback on revenue (not profit), creating a narrow path where cash flow timing — not technology — becomes the existential risk.

2ndhigh conviction
$TSMTaiwan Semiconductor Manufacturing

TSMC employs 20% of Taiwan economy; labor and electrician constraints cap further doubling

TSMC's direct and indirect employment dominates Taiwan's economy; tripling output would require more workers than Taiwan has, and Arizona fabs face the same electrician shortage (flown in by Cessna), creating hard physical limits to semiconductor supply scaling.

3rdhigh conviction
$000660.KSSK Hynix

SK Hynix misses earnings on LTA shift as memory cycle second derivative slows

Memory prices 3x'd last year but rate of change is decelerating; SK Hynix shifted to long-term agreements showing conservative pricing, triggering levered retail margin calls in Korea and a 40% Kospi drawdown that mirrors historic Taiwanese bubble dynamics.

5 total
$OPENAI
OpenAI
HIGHdoug o'loughlin·SemiAnalysis·13 days ago
Final pre-train IPO funding round risks $5T capex vs $500B revenue timing mismatch
Frontier labs are racing to final pre-train models to IPO and raise final mega-rounds; but even if revenue hits $500B, a $5T cumulative capex implies 10-year payback on revenue (not profit), creating a narrow path where cash flow timing — not technology — becomes the existential risk.
"OpenAI and Anthropic believes in the final pre-train coming soon because they're going to IPO and have the final funding round... you built a house that you cannot pay for. And so…"
30:11
$TSM
···
Taiwan Semiconductor Manufacturing
HIGHdoug o'loughlin·SemiAnalysis·13 days ago
TSMC employs 20% of Taiwan economy; labor and electrician constraints cap further doubling
TSMC's direct and indirect employment dominates Taiwan's economy; tripling output would require more workers than Taiwan has, and Arizona fabs face the same electrician shortage (flown in by Cessna), creating hard physical limits to semiconductor supply scaling.
"TSMC employs almost directly and indirectly, I think it's 20% of the economy... if TSMC was to double again... you're going to run out of people in Taiwan to make the freaking chi…"
42:10
$000660.KS
···
SK Hynix
HIGHdoug o'loughlin·SemiAnalysis·13 days ago
SK Hynix misses earnings on LTA shift as memory cycle second derivative slows
Memory prices 3x'd last year but rate of change is decelerating; SK Hynix shifted to long-term agreements showing conservative pricing, triggering levered retail margin calls in Korea and a 40% Kospi drawdown that mirrors historic Taiwanese bubble dynamics.
"SK Hynix missed earnings today, missed consensus. Um and the reason why is cuz they shifted more to LTAs... memory prices, say DRAM and even NAND, right? Has probably 3x'd over la…"
4:55
$CXMT
ChangXin Memory Technologies
MEDdoug o'loughlin·SemiAnalysis·13 days ago
CXMT positioned as #4 DRAM player with province-backed capacity to disrupt pricing
Chinese memory makers compete for GDP output not margins; CXMT is incentivized to produce and already winning Apple orders as Micron is accused of price gouging, but structural demand still exceeds supply so CXMT may only 'ruin the party' partially.
"CXMT is clearly number four in the market... it's a shortage, so they're going to make a lot of money in the meantime. And Apple, for example, is down to using CXMT memory because…"
9:17
$NVDA
···
Nvidia
MEDdoug o'loughlin·SemiAnalysis·13 days ago
H100s risk obsolescence as model size outpaces Hopper; B200/B300 pricing divergence key signal
Next-gen models like Kimi K3 require B300/MI355X single-node inference; H100 clusters will need 100x chips per model making them uneconomic, so a pricing divergence between B200 and B300 would confirm generational obsolescence.
"the old chips will be worthless... it's going to take like 100 H100s to inference one of these models... the true confirmation of this trend is if there is a pricing divergence be…"
20:58