US China Relations

The Trump administration has formally accused China of conducting industrial-scale theft of American AI intellectual property through tens of thousands of proxy accounts, signaling a hardening stance on technology competition that will reshape how institutional investors assess geopolitical risk in AI infrastructure. The disclosure comes as the administration simultaneously delays shipments of advanced Nvidia H200 chips to China, creating dual pressure points that institutional players must now factor into portfolio construction and supply chain due diligence.

• White House says China uses tens of thousands of proxy accounts to distill US frontier AI systems and jailbreak proprietary information
• Nvidia H200 chip sales to China remain blocked despite Trump administration approval in January, with no shipments yet completed
• Commerce Secretary signals H200 delays will persist while US and China remain at odds over trade terms and domestic investment priorities

• Tens of thousands of proxy accounts China allegedly uses to steal US AI intellectual property at industrial scale
• January when Trump administration formally approved H200 China sales, though conditions remain contested
• Zero H200 chip shipments to Chinese firms completed to date despite approval and heavy global demand

Michael Kratsios, director of the White House Office of Science and Technology Policy, issued a memo this week documenting what US officials characterize as a coordinated campaign by foreign entities based in China to extract proprietary information from American AI laboratories.

The accusation marks an escalation in Washington’s articulation of the technology competition threat and signals the administration intends to move beyond rhetorical criticism toward active enforcement.

Kratsios specified that the operations employ tens of thousands of proxy accounts designed to evade detection and deploy jailbreaking methods to expose information that American AI companies have classified as proprietary or trade-secret material.

For institutional investors, the memo represents a formal government acknowledgment that intellectual property theft in AI remains an active, unresolved vulnerability in the American AI supply chain. The scale cited, tens of thousands of accounts operating simultaneously, suggests the problem extends beyond isolated incidents to systematic, resource-intensive operations.

This finding carries direct implications for how investment committees should model technology risk, regulatory risk, and valuation assumptions for publicly traded AI infrastructure companies and software firms that depend on proprietary models as competitive moats.

White House pledges crackdown on unauthorized AI model distillation while alerting private sector

Kratsios signaled that the US government will take a three-part enforcement approach: first, alerting American AI companies to specific unauthorized distillation attempts targeting their frontier systems; second, considering steps to hold the actors accountable; and third, presumably working with relevant agencies to tighten access controls and monitoring.

The memo does not specify which agencies will lead enforcement or what tools the administration intends to deploy, leaving open the question of whether sanctions, criminal prosecution, or export controls will form the primary enforcement mechanism.

The White House announcement comes at a moment when American AI companies are rapidly scaling production and deploying models in cloud environments that require authentication and access monitoring.

The government’s warning will force these companies to undertake forensic analysis of their own user logs and authentication systems to identify whether proxy account campaigns have already succeeded in extracting training data or model weights.

For institutional investors holding positions in cloud infrastructure providers and AI software vendors, this creates a near-term disclosure risk: companies may be forced to publicly acknowledge successful intrusion attempts or implement visible security overhauls that could raise questions about prior control effectiveness.

The timing of the memo also signals a shift in how the Trump administration plans to coordinate between national security agencies and the private sector on AI competition. Unlike prior administrations that relied heavily on regulatory frameworks and interagency working groups, this approach emphasizes direct alert mechanisms and accountability.

Investors should monitor whether this translates into formal executive orders, Congressional legislation, or simply tighter vetting of export licenses for AI chips and software bound for jurisdictions deemed higher-risk.

Nvidia H200 sales gridlock between US conditions and Chinese government investment priorities

Commerce Secretary Howard Lutnick told a Senate hearing that Nvidia’s H200 chips have not been shipped to any Chinese firms despite the Trump administration’s formal approval of the sales in January.

Lutnick attributed the continued freeze to disagreement over sale terms and conditions in both jurisdictions, but emphasized a specific point: the Chinese central government has not yet permitted Chinese companies to purchase the chips because Beijing wants to prioritize domestic investment in its own AI chip industry rather than rely on imports.

The Chinese central government has not let them, as of yet, buy the chips, because they’re trying to keep their investment focused on their own domestic industry.

Howard Lutnick, Commerce Secretary

Lutnick’s statement reveals a more complex negotiation than a simple export ban. The Trump administration approved the sales conditionally in January, presumably with restrictions on end-use or technical specifications.

China’s response appears to be strategic restraint: by declining to purchase H200s even when given formal permission, the Chinese government signals both defiance toward US restrictions and confidence in its own domestic semiconductor road map.

For institutional investors tracking chip supply dynamics, this suggests that H200 demand will remain concentrated in North America, Europe, and allied markets, narrowing the addressable market for Nvidia’s highest-margin products and potentially prolonging delivery timelines in non-China regions as demand concentrates.

The delayed shipments also reveal fractures within Washington’s consensus on technology decoupling. Lutnick’s framing, that China itself is blocking the sales by refusing to permit domestic companies to purchase, contradicts the hardline position held by some Trump administration officials and Congressional China hawks who argue any sale of advanced chips strengthens Beijing’s AI capabilities.

By making public that China is rejecting the chips rather than the US is withholding them, Lutnick appears to be building a case that the administration has taken a cooperative stance while China has chosen confrontation. This rhetorical positioning matters because it affects how future sanctions or export controls are likely to be justified to Congress and international allies.

Affiliates rule delay signals trade deal negotiations trump unilateral tech restrictions

When pressed on whether the administration would restore an affiliates rule that would restrict US technology exports to entities with Chinese ownership stakes, Lutnick declined to commit.

He noted instead that the rule delay, extended for one year last November, forms part of a broader trade negotiation with China led by President Trump, Treasury Secretary Scott Bessent, and US Trade Representative Jamieson Greer.

Lutnick’s statement, that the rule “is a smart thing for the United States of America to consider, but it is part of the balance of that full trade agreement”, signaled that tech export policy remains subordinate to trade deal objectives.

This reveals a critical tension for institutional investors tracking technology policy. The administration has publicly accused China of industrial-scale AI theft while simultaneously signaling that comprehensive export restrictions on AI chips remain negotiable items within a larger trade framework.

Companies and investors cannot assume that the AI security concerns outlined in Kratsios’s memo will drive unilateral US policy action if those actions conflict with broader trade negotiation targets. Instead, technology access restrictions are likely to be traded off against tariff relief, agricultural concessions, or financial commitments.

For institutional allocators, this means the policy environment for US AI and chip companies will remain volatile and subject to quarterly trade negotiation updates rather than stable regulatory rules. An adverse trade negotiation outcome could rapidly expand Chinese access to American AI technology even after high-profile White House warnings about theft.

Conversely, a breakdown in trade talks could trigger immediate sanctions on AI exports. Investors should model both scenarios and monitor public statements from Bessent and Greer as leading indicators of which direction negotiations are trending.

The White House will need to issue specific guidance to American AI companies within weeks on how to identify and report proxy account campaigns, and the Commerce Department will clarify enforcement mechanisms for holding actors accountable, watch for formal interagency directives and whether any Chinese entities or individuals face public sanctions or criminal referrals in the next 60 days. Simultaneously, monitor Treasury and Trade Representative statements for signals on whether the January H200 approval will be expanded, restricted, or folded into a larger China trade agreement by mid-2025.