Urgent calls from OpenAI, Anthropic for an AI slowdown fall on deaf ears with Trump, Xi ahead of next week’s meeting

U.S. and Chinese leaders Donald Trump and Xi Jinping are expected to discuss AI safety at their Sept. 24 meeting, but both countries have rejected calls from OpenAI and Anthropic for a coordinated slowdown in frontier AI development. Trump said the U.S. needs no AI "guardrails," while China characterized U.S. safety rhetoric as fear-mongering and views restrictions on advanced AI-chip exports as an effort to preserve America’s technological lead. A potential agreement against AI weaponization or misuse by non-state actors may be discussed, but policy experts say a broader safety accord remains far off given major disagreements over model testing, government oversight, open versus closed models, and export controls.

Analysis

The investable implication is not an AI-safety outcome but a lower probability of coordinated capacity restraint. That supports the 6-18 month accelerator, networking, power, and cooling capex cycle: NVDA, AVGO, VRT, GEV and CEG retain upside to utilization and order-duration assumptions if hyperscalers treat geopolitical competition as justification to defend spend. The second-order beneficiary is physical infrastructure, where grid interconnection and transformer scarcity constrain deployment more than model availability; this shifts economic rents toward power equipment and contracted generation rather than application-layer AI vendors.

The near-term risk for NVDA is that intensified strategic competition raises the odds of tighter U.S. export enforcement even if broad AI investment accelerates. A bilateral statement that includes model-security cooperation but no semiconductor concession would be modestly positive for sentiment; a renewed restriction on China-specific accelerators would pressure NVDA's China revenue expectations while benefiting domestic Chinese substitution efforts and potentially compressing its multiple. The relevant 1-3 month catalyst is the Sept. 24 meeting and any subsequent Commerce Department language, not safety rhetoric itself.

Consensus may be underestimating how open-weight, lower-cost models can decouple inference growth from frontier-model monetization. Cheaper models expand compute demand, but they also weaken pricing power for closed-model providers and reduce the probability that enterprise AI software captures outsized margins quickly. This favors picks-and-shovels exposure over unlisted frontier labs and argues against extrapolating GPU demand into every AI application software multiple.

The thesis is falsified if hyperscaler capex guidance turns materially lower, NVDA lead times normalize without offsetting inference demand, or export rules expand to products currently designed to comply with China limits. Conversely, evidence of rising data-center power procurement, sustained networking attach rates, or a policy outcome that preserves legal China shipments would extend the infrastructure trade.

Key Decisions for Investors

• Maintain an overweight in NVDA versus broad software through the next two earnings cycles; add on policy-driven weakness around the Sept. 24 meeting only if no new China product restrictions are announced. Target a 10-15% upside from continued capex revisions; exit/reduce on explicit restrictions covering compliant China-oriented accelerators or a material hyperscaler capex cut.
• Initiate a 6-12 month pair: long VRT / short IGV. Data-center cooling and power-density bottlenecks have clearer revenue conversion than AI software monetization, while IGV remains exposed to compression if low-cost/open models erode application pricing. Reassess if VRT backlog conversion or gross-margin guidance disappoints.
• Add a smaller long GEV or CEG position over 6-18 months as an AI-power constraint hedge. Contracted power demand and grid equipment lead times should benefit even if the GPU cycle becomes more volatile; key risk is delayed data-center construction or easing power-price expectations.
• Avoid treating a diplomatic safety communiqué as a de-risking event for Chinese semiconductor exposure. Set an alert for Commerce Department follow-on action rather than trading headlines; the missing decision-useful data are product-level export thresholds and NVDA's disclosed China shipment mix.