Ep. 031 - EMERGENCY EPISODE: Are We Doomed? | Jordan Nanos, Doug O'Laughlin, Max Kan, Joey Brookhart

Watch on YouTube ↗  |  September 16, 2026 at 02:30  |  1:02:10  |  SemiAnalysis
Speakers
Joey Brookhart — Substack author, SemiAnalysis
Max Kan — Substack author, SemiAnalysis
Doug Olaflin — Analyst, SemiAnalysis
Jordan Nanos — Member of Technical Staff at SemiAnalysis

Summary

The episode is an emergency roundtable on Dario Amodei's 'We Must Pace the Frontier' blog post and responses from Sam Altman and David Sacks. The panel debates whether pacing frontier AI means less compute spending or more, arguing safety, alignment, and monitoring are compute-intensive and that demand still outstrips supply. They also discuss the OpenAI/Hugging Face security incident, Moonshot serving Claude, cybersecurity risks, AI regulation, and near-term predictions. Market implications center on AI compute demand, semiconductor dependence on OpenAI/Anthropic ARR, and risks to legacy cybersecurity stocks.

  • Panel reviews Dario Amodei's 'We Must Pace the Frontier' proposal for embedded evaluators and global coordination.
  • Debate centers on whether safety controls reduce or increase compute demand; consensus leans toward higher compute spend.
  • Near-term compute remains scarce, with GB300 pricing showing backwardation and premiums for immediate capacity.
  • OpenAI/Hugging Face incident raises concerns about AI agents, security infrastructure, and monitoring.
  • Moonshot allegedly served Claude rather than Kimi and harvested exchanges, raising data-security issues.
  • Prediction segment covers small bank hacks, bioweapons, election security, and cyber incidents.
  • Discussion of AI regulation under Trump and the stalled RASA export-control bill.
  • Final hot take: short a basket of legacy cybersecurity names such as Okta, CrowdStrike, Palo Alto, and Zscaler.
Ideas
Joey Brookhart Substack author, SemiAnalysis 2:11
Semis trade hinges on lab ARR.
The entire semis trade now depends on expectations for OpenAI and Anthropic net new ARR. These two ARR figures and the pace versus expectations are the key variables for AI infrastructure demand and semiconductor market cap, so investors should monitor them closely.
SMH MED
Max Kan Substack author, SemiAnalysis 5:36
Available AI compute capacity is highly valuable.
Near-term GPU capacity is scarce, and pricing shows backwardation: GB300 quotes for immediate starts are very high, while deferred starts are much cheaper. Providers with available capacity can hold out for premium prices, and buyers may pay a 20-30% premium to start two to three months earlier because they can monetize immediately. This makes freely available gigawatts extremely valuable.
Neoclouds HIGH
Max Kan Substack author, SemiAnalysis 5:36
Available AI compute capacity is highly valuable.
People interpreted Dario's pacing proposal as a signal that Anthropic and OpenAI will slow compute purchases, but safety, alignment, monitoring, and operational excellence are highly compute-intensive. Chain-of-thought monitoring already consumes around 20% of rollout compute, and QA on RL environments plus interpretability research require large compute budgets. With physical capacity still below demand, compute demand should continue to outstrip supply.
AI compute infrastructure SMH HIGH
Joey Brookhart Substack author, SemiAnalysis 40:48
AI ARR grows without better models.
AI adoption is still early, so ARR can keep growing exponentially even if model training stops. No better models are needed to penetrate the remaining 5,000 US banks and other laggard industries, so the current models are sufficient to drive broader enterprise adoption.
AI software MED
Doug Olaflin Analyst, SemiAnalysis 58:45
Short legacy cybersecurity names on AI risk.
Legacy network-effect cybersecurity companies that touch endpoint, identity, and network security are vulnerable to AI-driven attacks because adversaries can rewrite and exploit code at scale. A high-profile incident at one of Okta, CrowdStrike, Palo Alto Networks, or Zscaler is likely, and the same mechanism exists across the group, so short the basket.
OKTA PANW CRWD ZS CIBR HIGH
Up Next
59:21
US Treasury yields finally broke through 5%—where are the undervalued zones to watch now? | Yeo Do-eun, Heo Jae-mu, Park Ji-hun [Morning N Investment]
3PRO TV (삼프로TV) · Sep 16
28:46
If everyone has moved to defensive stocks, that may be exactly when we're at the bottom | Park Se-ik, Executive Director & Chesley Investment Advisory [Morning Brief / 26.09.16.Wed]
Chesley Investment Advisory (체슬리투자자문) · Sep 16
26:03
美 금리·유가 충격에도 반도체주는 버텼다… 필라델피아 반도체 반등의 비밀 / 돈은 원자재로 몰린다ㅣ 이주연 대표
815 Money Talk (815머니톡) · Sep 16
10:03
CONSIGNADO CLT É SEGURO? O QUE ACONTECE COM A DÍVIDA SE VOCÊ PERDER O EMPREGO
Market Makers · Sep 16
27:35
Even if Memory Prices Fall, Materials/Equipment/Components Companies' Earnings Will Rise" Why on Earth? | Kwon Tae-woo, Research Fellow at Hana Securities Research Center [Double Up]
3PRO TV (삼프로TV) · Sep 16

This SemiAnalysis video, published September 16, 2026, features Joey Brookhart, Max Kan, Doug Olaflin discussing SMH, Neoclouds, AI compute infrastructure, AI software, OKTA, PANW, CRWD, ZS, CIBR. 5 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Joey Brookhart, Max Kan, Doug Olaflin  · Tickers: SMH, Neoclouds, AI compute infrastructure, AI software, OKTA, PANW, CRWD, ZS, CIBR