tomshardware.com via Reddit

Huawei builds 122TB SSD using homegrown packaging

chips china ai china sanctions storage chips

Key insights

  • Huawei's 122TB SSD uses proprietary NAND die-stacking packaging, avoiding sanctioned 3D NAND chips entirely.
  • The packaging approach lets China reach hyperscale storage capacity without requiring leading-edge fab access.
  • AI data center demand for dense local storage is the primary driver behind Huawei's development push.

Why this matters

US semiconductor export controls have largely targeted chip fabrication chokepoints, but Huawei's packaging workaround shows that system-level engineering can partially offset process-node restrictions, shortening the timeline before China fields competitive AI storage infrastructure. For AI founders and infrastructure leaders, this signals that supply chain diversification assumptions built around US-allied NAND dominance need revision, particularly for enterprise and hyperscale deployments. The move also sets a precedent: if packaging innovation closes the capacity gap in storage, similar approaches in compute packaging could erode the effectiveness of GPU export controls faster than current policy models assume.

Summary

Huawei has unveiled a 122TB SSD that sidesteps US export controls on 3D NAND chips entirely, using proprietary die-stacking packaging technology to achieve enterprise-grade density without touching sanctioned components. The drive stacks standard 2D NAND dies in a custom high-density footprint, a packaging approach Huawei developed in-house after US restrictions cut off access to advanced 3D NAND from suppliers like Micron and SK Hynix. The result is a capacity figure, 122TB, that rivals what Western hyperscalers deploy in AI data centers today. Essentially: Huawei engineers a hardware workaround that makes the export control architecture partially irrelevant for storage. - The breakthrough is in packaging, not the underlying NAND process node, meaning China doesn't need leading-edge fabs to compete on drive capacity. - AI infrastructure is the demand driver; large model training and inference require petabyte-scale local storage, and China is building that stack domestically. - Tom's Hardware flags this as a sign of accelerating Chinese capability to route around semiconductor restrictions at the system design layer. If packaging innovation can substitute for process-node access, the strategic leverage embedded in US chip export controls narrows faster than policymakers anticipated.

Potential risks and opportunities

Risks

  • Western NAND suppliers (Micron, SK Hynix, Kioxia) face accelerated share loss in Chinese enterprise and AI storage markets if Huawei's drive reaches volume production in 2026.
  • US Bureau of Industry and Security may face pressure to expand controls to packaging materials and advanced substrate technologies, creating new compliance exposure for global semiconductor equipment firms (Applied Materials, Lam Research) with Chinese customers.
  • Hyperscalers outside China that benchmark against Huawei's density specs could pressure their existing storage vendors on pricing and roadmap timelines within the next two quarters.

Opportunities

  • Advanced packaging specialists (Amkor Technology, ASE Group) with non-China customer bases can use this as a proof point to accelerate investment pitches for high-density die-stacking programs.
  • Western AI infrastructure vendors (Pure Storage, NetApp) have a window to lock in multi-year contracts with Asian hyperscalers before Huawei's drive reaches verified volume availability.
  • Policy-focused consultancies and export control law firms gain a concrete case study to drive retainer engagements with semiconductor and storage clients reassessing their China market and supply chain exposure.

What we don't know yet

  • Performance specs, specifically sequential read/write throughput and IOPS, have not been publicly disclosed for the 122TB drive.
  • Whether Huawei's proprietary packaging yield rates are competitive enough for volume production at hyperscale quantities remains unreported.
  • Which Chinese cloud or AI infrastructure customers are lined up as launch buyers, and on what timeline, is unconfirmed as of May 2026.