News

Chinese AI chipmakers are raising prices sharply as a global shortage of high-bandwidth memory (HBM) pushes up costs and exposes a critical bottleneck in Beijing’s push for home-grown alternatives to Nvidia.

As per Astute Analytica, the High bandwidth memory market is poised for significant growth, with revenues expected to rise from US$ 501.0 million in 2024 to US$ 5,810.5 million by 2033, at a CAGR of 31.3% during the forecast period 2025-2033. 

Prices Jump 20–50% Across Huawei, Cambricon and Rivals

Huawei has lifted the indicative price of its Ascend 950DT accelerator card to above 250,000 yuan (about $37,255), a 20% to 50% increase from quotes given to customers just two months ago, depending on contract terms.

Cambricon has repriced its next‑generation chip, tentatively called the 690, at 20% to 30% above levels indicated two months earlier, while smaller domestic players MetaX and Iluvatar CoreX have made similar moves.

Older products are not spared: Huawei’s Ascend 950PR has climbed from around 60,000 yuan at the start of 2026 to more than 80,000 yuan, and the Ascend 910C has risen from roughly 90,000 yuan to over 110,000 yuan.

HBM Shortage Turns Global Squeeze into China-Specific Shock

The immediate trigger is a worldwide shortage of HBM, the stacked DRAM that feeds AI accelerators with data at very high speeds. HBM is produced by just three firms—SK Hynix, Samsung and Micron—and demand from AI data centers has outstripped supply, pushing spot prices for advanced modules to around $2,100, four to five times typical long‑term contract levels.

For China, the situation is compounded by U.S. export controls tightened in December 2024 that restrict access to the most advanced HBM. As a result, Chinese chipmakers face the same global supply crunch but with fewer legal channels to secure top-tier memory, forcing greater reliance on costlier gray‑market supplies and domestic alternatives that cannot yet fully meet volume needs.

Industry estimates suggest Huawei has been drawing down a stockpile of Samsung HBM acquired before the late‑2024 controls, while domestic HBM from ChangXin Memory (CXMT) will cover only a fraction of 2026 Ascend production targets. 

Strategic Implications for China’s AI Ambitions

The price hikes underscore that memory, not logic, is now the binding constraint on China’s AI accelerator rollout. Beijing has been pressing cloud providers, state-owned enterprises and AI labs to replace Nvidia GPUs with domestic chips such as Huawei’s Ascend series and Cambricon’s processors.

Higher prices and tighter HBM supply could slow the pace of this substitution, especially for cost-sensitive customers and smaller AI firms. At the same time, the squeeze is accelerating investment in domestic HBM capacity and advanced packaging, with CXMT reportedly beginning small-scale HBM3E production as it tries to close the gap with Korean leaders.

In the near term, however, Chinese AI chipmakers are passing memory cost inflation directly to customers, signaling that the dream of cheap, Nvidia-beating domestic accelerators will remain constrained by the realities of global memory markets and export policy.