The Best AI Investor Is Betting Against NVIDIA

HQ — Linked to reporting on Leopold Aschenbrenner’s AI portfolio: reported large put/short exposure to NVIDIA and other infrastructure names, a private investment in Anthropic, and long exposure to memory, power, and data-center–related names.

Source reporting (Limitless FT podcast/newsletter) claims Leopold Aschenbrenner’s AI-focused fund has sizable short/put exposure—comments reference roughly $9 billion of puts across NVIDIA and other infrastructure names—and that the fund has deployed private capital into Anthropic (reported VC/private stake in March 2025). The coverage highlights NVIDIA’s recent capital activity (a reported $25 billion bond offering) and notes an $80 billion buyback authorization was mentioned in discussion. The hosts frame these moves as a strategic rotation: bearish or hedged positioning on crowded picks‑and‑shovels infrastructure (GPUs, some infrastructure providers) and long positions in memory, power, networking and data-center buildout companies that could benefit from the next infrastructure wave. These are reported claims from the podcast; amounts, timings, and positions should be independently verified before making investment decisions.

Podcast/newsletter episode that discusses Leopold Aschenbrenner’s reported AI portfolio, including his alleged short on NVIDIA, a large private investment in Anthropic, NVIDIA’s bond offering, and the idea that the AI infrastructure trade is rotating from chips toward power, networking, memory, and data‑center buildout. Episode includes timestamps and links to the Limitless newsletter and social channels.

Reporting that Anthropic released then quickly withdrew frontier AI models (noted examples 'Fable 5' and 'Mythos 5') following a U.S. export-control / national-security notice. Coverage claims AWS allegedly restricted access due to unresolved API origin/destination verification and reported jailbreak concerns. The market takeaway suggested is rising regulatory/compliance burden for frontier AI distribution, which could advantage large, regulated platforms and defense/compliance vendors while creating monetization and headline risks for AI labs and cloud AI services.

Episode discussing SpaceX’s potential IPO, Starlink growth, and AI data-center plans, alongside notes about OpenAI data‑center deals and Anthropic’s model releases. The episode frames hardware and data-center scale as critical advantages and discusses implications for valuation and future capital flows into AI infrastructure.

Thematic episode on ‘AI loops’ and longer-running autonomous AI workflows, emphasizing rising compute and cost constraints and the ongoing importance of human judgment—contextual background for why infrastructure and runtime economics matter.

Coverage of Anthropic’s new model (Claude/Fable 5), showing stronger capabilities alongside tighter safety controls, demos across visual reasoning and enterprise use cases, and discussion of pricing and compute limits—relevant to debates about model distribution, compliance and cloud monetization.

Episode summarizing Apple’s WWDC AI/assistant updates (Apple Intelligence, on‑device + encrypted cloud processing) and potential rollout timing/regulatory friction—served as additional context for edge and memory/compute considerations in AI infrastructure.

Discussion of Microsoft Build and NVIDIA announcements, contrasting Microsoft’s AI adoption narrative with NVIDIA’s hardware momentum and noting compute constraints at large AI providers—context for platform versus chip dynamics.

Report on an alleged exploit abusing Meta’s AI-driven account recovery flow to perform account takeovers; highlights AI security risks, likely increases in security spending, and regulatory scrutiny—relevant to the broader security/compliance theme affecting AI deployments.