According to a New York Times report, Chinese intelligence allegedly ran an operation that cultivated a romantic relationship with NVIDIA CEO Jensen Huang to obtain information and influence the sale of advanced GPUs to China.
Allegations and key actors
The report names a woman who used aliases including “Alice Huang” and “Huang Le,” and identifies her as a former executive director at Singapore-based Megaspeed through at least 2023. US officials allege she cultivated a close personal relationship with Jensen Huang over several years. Authorities reportedly placed her under surveillance, and at least one US intelligence agency tracked her activity for more than a year.
Megaspeed is described in the reporting as having business ties to Chinese defense-related entities, including an investment in a cybersecurity subsidiary of China Electronics Technology Group (CETC). The company is also said to have engaged with Taiwan-based hardware firms and operated data centers in Southeast Asia using NVIDIA chips, with some capacity leased to Alibaba.
What intelligence agencies allege she was tasked to do
According to the reporting, Chinese operatives assigned two primary objectives: (1) collect inside information from Jensen Huang about NVIDIA’s technology and business strategies and (2) steer Huang’s advocacy to ease restrictions so certain NVIDIA chips could be sold into China. The operation, if true, combined classical intelligence collection (relationships, influence) with a commercial workaround to US export controls.
How the Megaspeed network reportedly sought to bypass export limits
The scheme described centers on intermediate entities and regional data-center deployments. Megaspeed allegedly built or contracted data-center capacity in Southeast Asia using NVIDIA accelerators, then leased capacity to large Chinese cloud providers such as Alibaba. That approach can create effective access to prohibited compute without directly shipping the highest-tier chips into mainland China, because cloud capacity in third countries can provide Chinese customers with computational power subject to export controls.
Timeline and claimed lobbying outcomes
US officials reportedly briefed President Trump last year about the relationship and associated concerns. The New York Times account states that in 2025 Jensen Huang lobbied the Trump administration and secured permission to sell limited quantities of NVIDIA’s second-strongest compute chip into China, while requests to export the top-tier Blackwell family were rejected. In public comments earlier, Huang had said Blackwell and its successor “Rubin” should not be exported to China.
Congressional response and upcoming trial
Following the reporting, US private sought answers. Congressional Republican members requested a closed-door session with Jensen Huang on November 18 to question him about his knowledge of the relationship, the timeline of events, and NVIDIA’s involvement in any third-party arrangements to route compute capacity to China.
NVIDIA and Chinese government responses
An NVIDIA spokesperson, John Rizzo, said the company does not share confidential customer information nor lobby customers to obtain export approvals. Rizzo emphasized that China tightly restricts imports of NVIDIA’s data-center products and that Jensen Huang routinely meets thousands of executives a year, characterizing reported interactions as not unusual.
The Chinese embassy dismissed the accusation as “baseless” and politically motivated. Neither the CIA nor the White House issued public statements on the specific individuals named in the reporting.
Practical implications for US export controls and cloud strategies
Export controls target hardware and sometimes cloud-provided services that deliver equivalent computational capability. If firms or state actors use third-country data centers to give restricted compute access to China, regulators face a jurisdictional challenge: should policy treat location of hardware, legal ownership, or end-user access as determinative?
Practical countermeasures and policy clarifications could include:
1) Tightening end-use and end-user rules. Explicitly cover cloud-access models in licensing criteria and add strict auditing of cross-border cloud contracts.
2) Enhanced disclosure requirements. Require companies to report third-party hosting arrangements that could deliver restricted compute to sanctioned customers.
3) Transaction screening and certification. Mandate export-compliance certifications from intermediaries that host restricted accelerators and create penalties for providing compute to prohibited users.
How companies can reduce risk
Enterprises that design, sell or host advanced accelerators should implement layered controls:
Due diligence: deeper screening of investors, partners and high-level external contacts for ties to state intelligence or defense-affiliated enterprises.
Contractual safeguards: clear clauses preventing subleasing, dual-use access, or covert partnerships that would enable sanctioned end users to obtain prohibited compute.
Technical controls: software-based access gating, telemetry that flags atypical multi-tenant usage patterns, and cryptographic attestation of hardware location and ownership.
Questions likely to appear in People Also Ask and featured snippets
To address common queries directly:
Who is the alleged Chinese agent? Reporting identifies a woman using the names “Alice Huang” and “Huang Le,” formerly an executive at Singapore-based Megaspeed.
Did she have a relationship with Jensen Huang? The New York Times reports a long-term personal relationship between the two; US officials monitored the woman for over a year according to the story.
Did NVIDIA export its best chips to China? According to the reporting, top-tier Blackwell chips were not approved for export, while limited quantities of a second-tier chip received permission in 2025.
What are the national security concerns? The concern is that foreign intelligence used personal influence to obtain proprietary information and to attempt to circumvent export rules that protect advanced compute capabilities.
What to watch next
Key developments to monitor include Jensen Huang’s closed hearing in Congress, any declassifications from US intelligence about the timeline or content of information shared, and regulatory clarifications regarding cloud-hosted compute. Investigative outcomes or further official statements from NVIDIA, US agencies, or the Chinese government could materially change the public understanding of the operation.
Data points and illustrative example: if a data center in Singapore hosts 1,000 accelerators that are nominally owned by a local entity but leased hourly to Chinese cloud customers, regulators could treat effective access (hours of GPU compute consumed by Chinese users) as an export equivalent. Monitoring and assigning compute-hours to end users is a practical enforcement lever that authorities are increasingly considering.
Be the first to comment