Artificial Intelligence (AI) May Have a New Bear Thesis: Human Extinction

AI-sector safety concerns have intensified after former OpenAI and Anthropic researcher Jacob Coxon alleged frontier-model developers are racing toward self-improving superintelligence without acting responsibly; OpenAI CEO Sam Altman has said the company will not pursue an IPO in 2026, citing safety and extinction risks. Alphabet, Amazon, Microsoft, Meta and Tesla are expected to spend more than $750 billion on AI infrastructure this year, while S&P Global projects the four hyperscalers, Oracle and SpaceX will spend another $1.3 trillion in 2027. A genuine slowdown in AI model development or capex could weaken the AI-driven EPS narrative underpinning 30%-40% of S&P 500 value, though hyperscalers continue to raise spending guidance.

Analysis

The investable issue is not safety rhetoric itself but whether it converts into enforceable training, deployment, or power-permitting constraints. Until a developer cuts compute commitments or a hyperscaler lowers capex guidance, this is primarily a positioning risk: crowded AI infrastructure longs have elevated sensitivity to any evidence that utilization, model-release cadence, or enterprise monetization is slowing. Treat the reported security episode and private-company timing as unverified until corroborated by regulators, customers, or audited disclosures.

A genuine slowdown would not hit the stack uniformly. ORCL is more exposed than cash-rich hyperscalers because its AI-cloud expansion relies more heavily on externally financed capacity and rapid revenue ramping; a lower utilization outlook could pressure both EBITDA expectations and financing spreads. Conversely, MSFT, GOOG, AMZN, and META could see near-term multiple support if reduced spending restores free-cash-flow conversion, although their long-duration AI revenue narratives would weaken. Second-order downside is greatest for high-beta compute, networking, cooling, and power names—NVDA, AVGO, VRT, ETN, and CEG—where valuations embed sustained build rates rather than merely installed capacity.

Over the next days, headline-driven de-risking is plausible but unlikely to be durable without an earnings-call capex revision. The 1-3 month catalyst path is regulatory action, an AI-lab financing delay accompanied by supplier order changes, or cloud backlog/lease disclosures that imply demand is lagging capacity. Over 6-18 months, a safety regime could actually entrench incumbent clouds: compliance, secure deployment, and audit requirements raise fixed costs and make frontier-model development less accessible to smaller labs. The bearish thesis is falsified if Q3/Q4 capex guidance rises again while cloud growth, GPU utilization, and AI product monetization accelerate.

Key Decisions for Investors

• Do not initiate a broad AI short on rhetoric alone; set an alert for any 5%+ cut to aggregate MSFT/GOOG/AMZN/META capex guidance or a material reduction in ORCL cloud backlog. That would justify a 1-3 month underweight in AI infrastructure.
• Express relative risk via long MSFT / short ORCL over the next quarter: MSFT has balance-sheet capacity and enterprise distribution, while ORCL has greater execution and funding sensitivity to a delayed AI-capacity ramp. Reassess if ORCL reports accelerating cloud revenue with stable or improving free-cash-flow conversion.
• For existing NVDA, AVGO, VRT, ETN, and CEG longs, reduce gross or add 2-3 month downside hedges ahead of hyperscaler results; these names face asymmetric downside if capex moves from 'higher but scrutinized' to 'deferred.' Remove hedges if aggregate capex guidance remains intact and supplier lead times/utilization do not weaken.
• If regulation becomes concrete rather than rhetorical, favor large-platform compliance beneficiaries—MSFT and GOOG—over smaller AI application and infrastructure beta. The key confirmation is a rule requiring model audits, licensing, or deployment controls rather than voluntary safety statements.