Huawei Tells Customers to Find AI Chip Memory

Q: Why Is Huawei Asking Customers to Find Their Own AI Chip Memory?

Huawei's Chinese suppliers cannot produce enough high-bandwidth memory, while U.S. export rules block direct purchases from Samsung and SK Hynix. Huawei is therefore directing some customers to procure the memory independently for its new AI chips.

Q: How Much Can Memory for One AI Chip Cost?

Each AI chip requires six to eight HBM units. At the reported black-market price of around $4,000 per unit, the memory alone can cost up to $32,000.

Q: How Are Smugglers Bypassing U.S. Export Rules?

Smugglers pair HBM with low-end server processors that fall below the export threshold, ship the packages into China through Hong Kong and separate the components at workshops there. The process has a high error rate, according to the report.

Free Instant SEO Audit

Glowing AI memory module representing Huawei's HBM supply shortage

By NewSunSEO November 5, 2026

Digital agencies managing client websites can now assess how hardware shortages and sanctions may affect AI-related projects before clients commit to a platform or product launch. Huawei may be directing some customers to procure high-bandwidth memory independently for its new AI chips because Chinese suppliers cannot meet demand and U.S. export rules restrict direct purchases.

That approach may have contributed to a black market for smuggled memory. Prices for smuggled HBM3E units may have risen from a few hundred dollars to around $4,000 in one year, creating cost and compliance concerns for businesses that rely on AI infrastructure.

What Changed at Huawei?

Huawei plans to launch the Ascend 960DT early next year. According to The Information, as reported by The Decoder, the company says the chip will double the performance of its current chips.

The planned launch comes with a memory constraint. Huawei’s Chinese suppliers may not produce enough High-Bandwidth Memory, also known as HBM, to meet the company’s requirements. U.S. export rules also block direct purchases from Samsung and SK Hynix, which together are reported to control more than 80 percent of the global market.

Huawei’s response has been to tell some customers to find and purchase the memory themselves. That shifts part of the procurement burden from the chip supplier to buyers who may have little visibility into how the component was obtained.

How the Black Market Supply Route Works

Smugglers may have developed a workaround for the export restrictions.

The source says that, once the packages arrive, workshops separate the memory chips from the processors. The source describes this process as fragile, with a high error rate, although the large price gap has made the operation attractive to smugglers.

The market also includes supply from older Nvidia GPUs that have been stripped for parts. It also draws on excess inventory held by chip designers that abandoned projects or reduced orders. These channels give buyers more than one way to obtain scarce components outside standard distribution networks.

What the Memory Shortage Means for Clients

Each AI chip may require six to eight HBM units. At an estimated black-market price of around $4,000 per unit, the memory associated with one chip can cost as much as $32,000.

For agencies supporting clients with AI products, that cost changes the planning conversation. Hardware availability, component provenance and export restrictions can affect pricing, delivery schedules and product architecture before development begins. An agency looking after client websites should ask vendors for a clear breakdown of component supply and identify any features that depend on restricted hardware.

The wider enforcement picture adds another layer of uncertainty. According to South Korea’s customs authority, trade violations worth $560 million had been uncovered by May, exceeding the previous full-year record. According to South Korea’s customs authority, one-third of the total involved strategic goods such as chips exported without proper authorization or through false declarations.

Clients and agencies should document the source of critical components and avoid treating unusual prices or undocumented supply routes as routine procurement. The issue is especially relevant where a hardware supplier cannot explain where HBM came from or whether its acquisition complies with applicable export rules.

What to Do Now

Review AI hardware dependencies. Ask each relevant supplier which chips and memory components are required, and identify where alternatives exist.
Request supply-chain details. Seek confirmation of how HBM was acquired, including whether it came through an authorized distributor or an unverified secondary market.
Recheck rollout budgets. Calculate the cost of six to eight memory units for each planned chip and include the possibility of scarcity-related price changes.
Separate web delivery from hardware supply. Evaluate whether client websites can use existing computing capacity while constrained AI hardware is sourced and deployed.
Document compliance risk. Keep records of vendor assurances and avoid any procurement route involving false declarations or unauthorized exports.

How Agencies Can Protect Client Projects

An agency that manages client websites can turn this supply issue into a clearer planning process. Start with the services the client actually needs, then examine which AI functions depend on new hardware. This helps the agency distinguish a software availability problem from a component constraint and gives the client a more grounded view of costs.

For web projects, the useful question is whether the proposed experience can launch with current infrastructure while a preferred chip platform waits for stable memory supply. Where hardware provenance is unclear, the agency should require documentation before endorsing the deployment.

Cost estimates also need more context. A reported unit price of around $4,000 may not represent a normal catalog price. It reflects a constrained gray market in which extraction and separation can fail. Agencies can use the six-to-eight-unit requirement to identify where memory may dominate the cost of an AI deployment, then ask the supplier for an authorized alternative or a production timeline.

FAQ

Why Is Huawei Asking Customers to Find Their Own AI Chip Memory?

Huawei’s Chinese suppliers may not produce enough high-bandwidth memory, while U.S. export rules can block direct purchases from Samsung and SK Hynix. Huawei is therefore directing some customers to procure the memory independently for its new AI chips.

How Much Can Memory for One AI Chip Cost?

Each AI chip may require six to eight HBM units.

How Are Smugglers Bypassing U.S. Export Rules?

Smugglers may pair HBM with low-end server processors that fall below the export threshold, ship the packages into China through Hong Kong and separate the components at workshops there. The process has a high error rate, according to the report.

This article reports on publicly available information from the sources listed below. We have no reason to believe it is inaccurate, but we have not independently verified every claim, and details may change as the story develops.