Diversion and resale: estimating compute smuggling to China

2026-01-28 · Source: Epoch AI · Field: Government & Public Sector — Public Policy & Governance, Regulatory & Compliance, International Relations & Diplomacy · Depth: Intermediate, medium

Summary

An analysis estimates that between 290,000 and 1.6 million Nvidia H100-equivalent (H100e) AI chips were smuggled into China by the end of 2025, with a median estimate of 660,000 H100e. This volume represents approximately 3% of the global compute stockpile. The study, based on Monte Carlo simulations, integrates evidence from indictments and investigative reporting, which initially points to nearly 300,000 H100e in alleged diversions by the end of 2025. This figure alone equals roughly a quarter of China's legally acquired or domestically produced compute. US export controls, initiated in October 2022 and tightened in October 2023 and April 2025, have incentivized this illicit trade. Evidence includes DOJ indictments, such as Supermicro employees accused of diverting \$2.5 billion in Nvidia chips, and reports of over \$1 billion in AI chips smuggled between April and July 2025. The true total is likely higher due to undetected smuggling.

Key takeaway

For policy makers evaluating the effectiveness of AI chip export controls, this analysis indicates that current measures are being circumvented at a substantial scale. You should re-evaluate enforcement strategies and enhance supply chain monitoring to mitigate the flow of advanced compute to restricted entities. The estimated 660,000 H100-equivalent chips smuggled by 2025 represent a significant challenge to national security objectives and technological parity. Consider strengthening international cooperation to disrupt these illicit networks.

Key insights

US export controls have driven substantial, difficult-to-quantify AI chip smuggling to China, impacting global compute distribution.

Principles

• Export controls create strong incentives for illicit trade.
• Smuggling estimates require diverse data sources.
• Undetected activities skew true market figures.

Method

Monte Carlo simulations combine "diversion" evidence (supply chain leakage) and "resale" evidence (illicit market transactions) to estimate total smuggled compute.

In practice

• Track supply chain anomalies for diversion.
• Monitor gray markets for resale activity.
• Cross-reference diversion and resale data.

Topics

• AI Chip Smuggling
• Export Controls
• NVIDIA H100-equivalent
• Supply Chain Diversion
• Monte Carlo Simulation
• National Security

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Policy Maker, Legal Professional, Consultant

Related on AIssential

• Taiwan Seeks to Detain Three in AI Chip Smuggling Crackdown — The provided content is a paywall or error page from Bloomberg.com, preventing access to the full article.
• How a Beijing Visit Could Reset AI’s Global Supply Lines — US-China tech diplomacy is shifting towards transactional engagement, potentially resetting AI supply lines.
• Super Micro Co-Founder Arrested for Smuggling $2.5 Billion in NVIDIA GPUs to China — A sophisticated scheme allegedly diverted billions in US AI servers to China, raising national security and export control enforcement concerns.
• Chip Smuggling — US export controls have fueled a significant, complex AI chip smuggling operation into China.
• Nvidia’s B300 systems fetch over $1 million on China’s underground market — US export controls and persistent Chinese demand have created a highly inflated black market for Nvidia's restricted AI chips.
• Nvidia Staff Detained by Taiwan in China Chip Smuggling Case — Nvidia staff detained in Taiwan over alleged China chip smuggling highlights export control enforcement.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Epoch AI.