Huawei's HBM Shortage Forces Customers to Black Market Procurement

Huawei is effectively outsourcing the procurement of critical memory components for its AI chips to its customers, a move that pushes buyers into a volatile and expensive black market. According to The Information, as reported by The Decoder, the company cannot secure sufficient High-Bandwidth Memory (HBM) from domestic suppliers and is blocked from direct purchases from major global vendors by U.S. export controls.

What happened

The Decoder reports that Huawei plans to launch the Ascend 960DT chip early next year, which the company claims will double the performance of its current hardware. However, Chinese suppliers lack the capacity to produce enough HBM, and U.S. regulations prevent Huawei from buying directly from Samsung and SK Hynix, which together control over 80 percent of the global HBM market. To circumvent these barriers, a sophisticated smuggling network has emerged. Smugglers bundle HBM with low-performance server processors that fall below U.S. export thresholds, shipping them to China via Hong Kong. Once in China, workshops separate the memory chips from the processors. While this process is fragile with high error rates, the profit margins remain attractive. A broker in Shenzhen noted that prices for smuggled HBM3E memory have surged from a few hundred dollars to approximately $4,000 per unit in just one year. Since each AI chip requires six to eight units, the memory alone can cost up to $32,000 per chip. Additionally, South Korea’s customs authority uncovered trade violations worth $560 million by May, surpassing previous annual records, with a third involving strategic goods like chips exported without proper authorization.

Why it matters for small and mid-sized tech teams

This development signals a structural fracture in the global AI hardware supply chain that directly impacts procurement strategy for companies relying on third-party AI infrastructure. For small and mid-sized companies, the risk is not just about price, but about compliance and reliability. If you are building AI products or services, the assumption that hardware supply chains are stable and transparent is no longer valid. The fact that a major player like Huawei is offloading procurement risk to its customers indicates that even large enterprises are struggling to secure compliant, high-performance components. For smaller teams, this means that any dependency on AI hardware with opaque sourcing or potential sanction violations carries significant legal and operational risk. Furthermore, the dramatic price inflation of HBM in the gray market suggests that the cost of AI inference and training will remain high and volatile for the foreseeable future, squeezing margins for companies that do not have long-term, secured supply contracts with major OEMs. This is a warning that the "easy" era of buying off-the-shelf AI hardware is ending, replaced by a complex landscape where supply chain integrity is a competitive differentiator.

What to do this week

1. Audit your current AI hardware vendors to verify their sourcing compliance and ensure they are not relying on gray-market or smuggled components that could expose your company to legal risk.
2. Review your AI infrastructure costs to identify dependencies on specific chip models that may be subject to supply shocks, and begin exploring alternative hardware configurations or cloud-based alternatives to hedge against price volatility.
3. Update your vendor risk management policies to include specific clauses regarding supply chain transparency and compliance with international export controls, ensuring your contracts protect you from downstream procurement failures.

Sources

• Huawei tells customers to find their own AI chip memory, effectively outsourcing procurement to the black market — The Decoder