Jim Cramer questions Anthropic CEO's slowdown manifesto and puts forth an action plan

Potential self-imposed slowing of AI data-center construction by OpenAI, xAI and Anthropic is raising downside risk for Nvidia and other infrastructure beneficiaries, with the author warning that this risk is not yet reflected in share prices. The commentary argues Anthropic's call to "pace the frontier" could support stricter regulation and create competitive barriers, while reducing capital-expenditure intensity ahead of a potential IPO. The author has already reduced data-center exposure, including selling Corning, and plans to consider repurchasing favored AI-linked names only after a fuller market overreaction.

Analysis

The relevant transmission mechanism is not a modest reduction in AI demand, but a lower utilization and financing case for incremental data-center capacity. NVDA retains the strongest near-term earnings insulation because its backlog and accelerator replacement cycle extend beyond a single customer’s capex decision; the more vulnerable exposures are second-derivative infrastructure suppliers whose valuation assumes sustained greenfield construction. GLW faces greater downside asymmetry than NVDA if optical-networking orders are deferred, while ORCL’s multiple is vulnerable if contracted cloud capacity requires slower customer ramp or incremental financing.

A coordinated "pace" narrative would also alter competitive dynamics: incumbent frontier labs could preserve access to scarce compute while reducing the urgency of broad capacity deployment, hurting smaller model developers, GPU-cloud lessors, and networking vendors before it materially damages chip demand. Conversely, a regulatory framework that raises compliance costs can increase the value of hyperscaler distribution, proprietary data, and balance-sheet capacity—supporting MSFT, GOOGL and AMZN relative to pure-play AI infrastructure. The market should distinguish a delay in training-cluster construction from an outright reduction in inference demand; the latter is the condition required for a durable NVDA estimate reset.

Over the next days, headline-driven de-risking can pressure the crowded AI complex, but the 1-3 month catalyst is whether hyperscaler capex guidance, GPU lead times, and ORCL remaining-performance-obligation conversion weaken together. A single laboratory’s rhetoric is not independently verifiable demand destruction and is insufficient alone to underwrite a structural short. The bearish thesis is falsified by stable accelerator lead times, unchanged cloud capacity commitments, and continued optical component order growth through the next earnings cycle; it strengthens materially if multiple buyers cite power availability, regulation, or model-training economics as causes of deferral.

The contrarian read is that a regulatory moat may shift spend rather than eliminate it. If frontier-model development becomes more concentrated, approved operators may buy larger, more standardized clusters and capture higher returns, ultimately favoring the best-capitalized compute vendors. That makes a broad AI short unattractive after a sharp drawdown; relative-value positioning between construction-sensitive suppliers and diversified platform beneficiaries offers cleaner risk control.

Key Decisions for Investors

• Do not initiate a standalone NVDA short on policy rhetoric. Instead, place an alert around the next hyperscaler and NVDA guidance cycle: reduce long exposure only if aggregate capex guidance or NVDA data-center revenue outlook is cut, rather than merely delayed; absent that evidence, selloffs driven solely by commentary are potential staged long-entry opportunities over 1-3 months.
• Express near-term construction-delay risk via a 1-3 month pair: short GLW versus long MSFT or AMZN in dollar-neutral size. GLW has higher sensitivity to incremental network buildouts, while the platforms can benefit from regulatory concentration; cover if GLW reports resilient optical order growth or if a platform materially cuts AI capex.
• Maintain ORCL as a watch rather than a fresh long until customer conversion and funding assumptions are independently validated. A bearish position becomes actionable only if remaining-performance-obligation growth decelerates alongside cloud margin pressure or higher financing needs; without those data, the risk/reward is headline-dependent.
• Increase diversification through CAH and BNY rather than treating them as direct AI hedges. Their earnings drivers are less correlated with data-center capex, but the thesis is portfolio beta reduction over 6-18 months, not a catalyst trade; reassess if rate volatility or credit stress changes the financials’ earnings outlook.