Calls for Slower AI Development Have Created a Compelling Dip for AI Stocks

The article argues that a recent selloff in AI-related stocks following safety-driven calls to slow AI development is overdone, asserting that the Trump administration is unlikely to impose meaningful restrictions. It highlights Broadcom's maintained long-term revenue targets and Iren's view that existing AI capabilities alone require more compute capacity than the world can build for years. The bullish thesis is that profitable hyperscalers such as Meta, Microsoft and Alphabet will continue AI capex, supporting chipmakers and data-center infrastructure despite potential constraints faced by cash-burning OpenAI, Anthropic and SpaceX.

Analysis

The relevant distinction is not AI-safety rhetoric but whether it changes hyperscaler committed capex, chip purchase orders, or utility interconnection schedules. A pause in frontier-model training would likely shift spend toward inference deployment and lower-cost custom silicon rather than eliminate compute demand; that is relatively supportive of AVGO’s ASIC/networking mix and the cloud platforms monetizing distribution, while it is less cleanly supportive of the highest-duration GPU and data-center-development equities. The article provides no evidence of canceled orders, revised capex, or regulatory text, so any selloff attributed solely to commentary should be treated as positioning-driven rather than fundamental.

The more consequential 1-3 month catalyst is upcoming hyperscaler guidance: META, MSFT and GOOG need to demonstrate that incremental AI capex is translating into cloud, advertising, or engagement monetization rather than merely extending depreciation and power costs. A sustained infrastructure build also tightens power, transformers, cooling and networking capacity; this can preserve pricing for incumbent operators but raises execution and financing risk for levered data-center developers such as IREN. IREN’s equity is a high-beta proxy for both AI demand and capital-market access, not a pure beneficiary of secular compute demand.

Contrarian risk: a pro-build policy backdrop can be negative at the margin if it accelerates supply additions faster than AI workloads monetize, compressing returns on invested capital for hyperscalers and hosting providers. The thesis is falsified by a material reduction in aggregate 2027 capex guidance, evidence that GPU/ASIC lead times are easing alongside falling utilization, or a policy action restricting data-center power access/export markets—not by nonbinding executive statements.

Key Decisions for Investors

• Buy AVGO on broad AI-risk-off weakness, with a 6-12 month horizon; prefer it to NVDA where the objective is exposure to inference/custom-silicon mix and networking content. Size only after confirming no reduction in management’s AI revenue outlook; exit on a material guide-down or customer concentration deterioration.
• Pair long AVGO / short IREN over the next 1-3 months if AI infrastructure beta continues to re-rate higher without financing clarity. The trade isolates durable component demand from data-center funding, power-price and execution risk; cover the IREN short if it secures long-duration contracted capacity at attractive economics or if AVGO’s AI order commentary weakens.
• Maintain META and GOOG as higher-quality AI capex exposures rather than adding indiscriminately to data-center developers. Add only around earnings if management quantifies monetization against capex; the key risk is another step-up in spending without a corresponding revenue or margin bridge.
• Set an event alert around each major hyperscaler earnings release for aggregate capex guidance, disclosed useful-life/depreciation assumptions, and AI revenue commentary. Do not treat the article’s claimed regulatory backdrop as a trading catalyst until an actual executive order, permitting rule, or export-control action is published.