Huawei pushes AI chip memory procurement to black market amid U.S. restrictions

Why Huawei is shifting memory procurement to customers

Huawei is advising customers to source High-Bandwidth Memory (HBM) for its Ascend 960DT AI chip independently, citing shortages driven by U.S. export restrictions. The company’s Chinese suppliers cannot meet demand for HBM, a critical component for AI chips, while direct purchases from dominant suppliers Samsung and SK Hynix are blocked under U.S. rules.

The Ascend 960DT, planned for launch early next year, is expected to double the performance of Huawei’s current chips. However, the inability to secure sufficient HBM through legal channels has pushed procurement into the black market, where costs have escalated sharply.

How smugglers bypass U.S. export controls

A smuggling network has emerged to circumvent U.S. export restrictions on HBM. Smugglers bundle memory chips with low-cost server processors—whose performance falls below export thresholds—and route shipments to China via Hong Kong. Once delivered, workshops in cities like Shenzhen separate the HBM from the processors, making the components available for AI chip production.

This process is described as fragile, with a high error rate, but it has become a key supply route for Chinese companies facing restrictions. The method exploits gaps in export controls, enabling the flow of critical components despite regulatory barriers.

Costs and trade violations surge

Prices for smuggled HBM3E memory have soared to $4,000 per unit, up from a few hundred dollars a year ago. Each AI chip requires six to eight units of HBM3E, meaning the memory alone can cost up to $32,000 per chip. This price surge reflects the risks and scarcity driven by U.S. restrictions, as well as the limited supply of alternative sources.

The black market is also fed by older Nvidia GPUs stripped for parts and excess inventory from chip designers, further complicating efforts to track and regulate the flow of components.

South Korea’s customs authority reported trade violations worth $560 million by May, with a third involving unauthorized chip exports or false declarations. The scale of these violations underscores the challenges regulators face in enforcing export controls amid high demand for AI hardware.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

For founders and operators in the AI hardware space, Huawei’s shift to customer-driven procurement highlights the growing risks and costs of navigating U.S. export restrictions. The emergence of a smuggling network—while a stopgap—demonstrates the lengths to which companies may go to secure critical components, as well as the unintended consequences of export controls.

The price surge for HBM3E memory also signals broader market distortions, where restricted supply chains drive up costs and create opportunities for intermediaries. For startups reliant on AI chips, this environment increases volatility in both pricing and availability, particularly for those without established relationships with suppliers like Samsung or SK Hynix.

Moreover, the trade violations reported in South Korea suggest heightened scrutiny on chip exports, which could lead to stricter enforcement or new regulatory measures. Companies in the AI hardware ecosystem should anticipate continued disruptions and consider diversifying supply chains where possible, though options remain limited under current restrictions.

Related on Lazyfounder

• Samsung
• SK Hynix
• Nvidia
• More technology news