Geopolitics, AI (including agentic AI), trade tensions and market hype reshape capital allocation
Key Questions
What is Anthropic's Project Glasswing?
Anthropic announced Project Glasswing, involving its powerful unreleased cyber model called Mythos or Claude Mythos, which has discovered thousands of zero-day vulnerabilities. The company partnered with AWS, Google, and Nvidia to patch these issues. It is considered too dangerous to release publicly due to its capabilities.
Why is Anthropic withholding its most powerful AI cyber model?
Anthropic's Mythos model is so powerful that it finds thousands of zero-days, making it too dangerous for public release. Instead, they built Project Glasswing to collaborate with partners like AWS, Google, and Nvidia on patching vulnerabilities. Security experts note that these flaws existed before AI but are now discovered faster.
What threats face OpenAI's Stargate data center?
OpenAI's $30B Stargate data center in UAE (1GW capacity) faces bomb threats from Iran's IRGC, including a video threatening 'complete and utter annihilation'. It also contends with Nvidia and Oracle supply risks amid geopolitics. This highlights geo-threats in AI infrastructure.
What is Anthropic's current revenue run-rate?
Anthropic has reached a $30B annual run-rate (ARR), topping OpenAI. This growth is fueled by deals like expanded chip agreements with Broadcom for 3.5GW TPUs and partnerships with Google. It underscores booming AI demand.
What is Meta's Muse Spark model?
Meta debuted Muse Spark, its first multimodal AI model developed under Alexandr Wang. It represents Meta's push into advanced AI alongside open-source efforts and natural gas infrastructure. Rumors suggest a massive contract with Anthropic, possibly the largest in software history.
How is Japan supporting AI development?
Japan is relaxing privacy laws to become the 'easiest country to develop AI'. This eases data usage for training models. It aims to boost competitiveness amid global AI races.
What is Nvidia-backed Firmus planning?
Nvidia-backed Firmus, valued at $5.5B, secured $505M equity and $10B debt from Blackstone for 1.6GW AI data centers in Australia. It targets a $2B ASX IPO. This reflects massive infrastructure financing.
What risks does Micron face despite AI demand?
Micron's HBM3E for 2026 is sold out due to AI boom, but rising capex erodes FCF and highlights allocation risks. Broader concerns include hyperscalers' $720B AI capex, Big Tech layoffs amid spending, and needs for ROIC gates, supply committees, and governance on agentic AI, cyber threats, and power/debt.
Hyperscalers' $720B AI capex + OpenAI $122B raise/Stargate UAE DC faces Iran IRGC bomb threats ($30B/1GW, NVDA/ORCL supply risks) + Anthropic $30B ARR/Broadcom 3.5GW TPU/Google + Anthropic Glasswing/Claude Mythos (unreleased cyber model finds thousands zero-days, partners AWS/Google/Nvidia patching; too dangerous to release publicly) + Meta Muse Spark multimodal model + rumored Meta-Anthropic mega contract + agentic bottlenecks (web/UI access) + pmarca AI security fixes + Japan privacy easing for AI dev + Nvidia-backed Firmus $5.5B val/$10B debt (1.6GW Aussie) + Meta nat gas/open-source + NA VC $252.6B Q1 (87% AI) + Big Tech layoffs vs spend + Nvidia $40B buybacks + Micron HBM3E 2026 sold out but capex/FCF erosion highlight risks; boards need ROIC/burn/infra gates, agent controls/red-teaming, supply committees, power/debt/cyber/policy/geo-threat tests, CEO AI reinvention playbooks, superintelligence auditing. Agentic AI governance trends emphasize value creation, ROI/observability/guardrails.