How is the global high bandwidth memory (HBM) shortage—intensified for Chinese buyers by U.S. export controls since December 2024 and reliance on costly grey market supplies—driving price increases and allocation changes among Chinese AI chip makers such as Huawei, Cambricon, Met

The shortage is raising the cost and limiting the availability of the memory required for Chinese AI accelerators, turning HBM into a bottleneck for China’s domestic alternatives to Nvidia. U.S. controls added in December 2024 specifically restrict covered HBM for China, compounding the global supply squeeze. 12

Reported price and allocation effects

• Huawei’s indicated price for the forthcoming Ascend 950DT has risen to more than RMB 250,000 (about $37,000), reportedly 20%–50% above customer quotes from two months earlier, depending on contract terms. 3452

• Cambricon reportedly raised the indicative price of its next-generation accelerator, tentatively called the 690, by 20%–30%. The available evidence does not provide a reliable absolute RMB price. 52

• Reuters reports that Huawei, Cambricon, MetaX and Iluvatar CoreX have raised prices for current or planned AI processors as HBM costs rise and supplies tighten. 34

• The supplied evidence does not substantiate exact current prices or percentage increases for Huawei’s 950PR and 910C, nor does it substantiate the specific claim that Iluvatar has prioritized a defined share or volume of shipments to ByteDance. Those details should therefore be treated as unverified here.

• Huawei had planned roughly 750,000 950PR shipments in 2026; Reuters reported that samples went to customers in January, mass production was expected to begin the following month, and full shipments were expected in the second half of 2026. 7

• The available evidence does not establish a precise launch date for the 950DT. It does indicate that broader Ascend 950 supernode supply was expected to improve only once products began shipping at scale, with constraints persisting during the production ramp. 6

What this means strategically

• China’s replacement of Nvidia is not simply a GPU-design problem: constrained and more expensive HBM raises accelerator prices, reduces deliverable volumes, and makes domestic systems less economical for Chinese cloud firms. That slows substitution even when a Chinese logic chip is available. 234

• Reliance on costly nonstandard or grey-market supply channels would logically worsen this disadvantage, but the supplied evidence does not independently quantify those purchases or their premium.

• For Korean memory leaders, persistent scarcity can sustain pricing power and reinforce the strategic value of HBM. SK Hynix is identified as Nvidia’s main HBM supplier, illustrating why Korean memory capacity remains central to the AI supply chain. 5

• The opposite longer-term risk for Korean firms is that Chinese customers have a strong incentive to qualify domestic memory and accelerator supply. Thus, the shortage protects near-term Korean leadership and margins, while intensifying China’s motivation to build competing memory capability.

Likely duration

• There is insufficient evidence in the supplied material to attribute a specific warning to former Bank of Korea governor Changyong Rhee or to assess his forecast.

• The evidence does support a prolonged, rather than brief, constraint: AI demand is still outpacing supply, and SK Group chairman Chey Tae-won said the broader chip-wafer shortage could last until 2030. That is not a precise HBM forecast, but it supports the conclusion that memory tightness may endure for years as capacity additions lag AI infrastructure demand. 5