Trump says working with Xi on AI would hurt US companies

President Trump rejected working with China on AI governance, prioritizing U.S. technological leadership ahead of a White House meeting with executives from Anthropic, Meta, Google, Nvidia, Palantir and OpenAI. The meeting follows reports of AI agents accessing customer systems without instructions, while 73% of Americans in a September Reuters/Ipsos poll said AI companies have not done enough to prevent serious societal harm. Policymakers face increasing pressure to balance oversight with innovation, but Trump and House Speaker Mike Johnson oppose aggressive regulation over concerns it could weaken U.S. competitiveness against China.

Analysis

The investable read-through is a likely compliance moat rather than an immediate capex slowdown. GOOG and META can absorb pre-release testing, logging, model-access controls and liability reserves across diversified cash flows; smaller application-layer companies may face slower product cycles and higher customer-acquisition friction if enterprise buyers demand auditability. PLTR is positioned to monetize this through government and regulated-enterprise deployments, while cybersecurity vendors such as PANW and CRWD gain if autonomous-agent permissions become a board-level control problem.

Over the next 1-3 months, the key catalyst is whether the administration converts voluntary review into procurement standards, incident-reporting requirements, or liability guidance. Binding rules would initially pressure frontier-model release velocity, but would be more damaging to private, venture-backed agent vendors than to hyperscalers; the latter can turn compliance into a distribution advantage. A U.S.-China split in AI governance also raises the probability that compute/export policy, rather than domestic safety regulation, becomes the operative risk for NVDA.

The consensus risk is treating safety headlines as categorically bearish for AI. Political incentives still favor domestic compute expansion and commercialization, making a broad development pause unlikely; the more probable outcome is targeted controls around autonomous system access, cyber capabilities and sensitive data. That favors long-duration platform owners and trusted government suppliers, while making richly valued hardware assemblers such as SMCI vulnerable if customers concentrate spend with vertically integrated cloud providers.

Key Decisions for Investors

• Initiate a 3-6 month pair: long GOOG and META, short SMCI, sized dollar-neutral. Compliance-driven enterprise preference for integrated cloud/model stacks supports platform multiples, while incremental AI infrastructure spend may consolidate at hyperscalers; target 10-15% pair return. Exit if SMCI demonstrates sustained gross-margin expansion or hyperscaler AI capex guidance accelerates materially above expectations.
• Accumulate PLTR on weakness ahead of the next federal procurement and earnings cycle; use a 6-12 month horizon. A formal government AI assurance framework would increase the value of deployment, permissions and audit layers, but limit position size given valuation sensitivity. Falsify on material deceleration in U.S. government revenue growth or evidence that standards remain purely voluntary.
• Buy a modest PANW or CRWD basket for a 6-18 month second-order exposure to agent identity, access-control and incident-response spending. The thesis requires disclosed enterprise security-budget reallocation toward AI-agent governance; absent that evidence by the next two earnings cycles, close the position rather than underwriting a generic AI-security premium.
• Maintain NVDA exposure but hedge China-policy risk through a defined-risk collar into the next export-control or Commerce Department communication. Domestic competitive policy remains supportive for accelerator demand, but a further tightening on China-bound products could create abrupt revenue and inventory-reset risk; remove the hedge if policy language explicitly preserves current product licensing.