Skip to main content
Live
Main content

China's CXMT to start HBM3E production despite 20-30% yields

China's largest DRAM maker has sampled HBM3E to domestic AI chip designers and plans limited output within two months.

Jaeden Schafer
Editor in Chief · · 4 min read
China's CXMT to start HBM3E production despite 20-30% yields

ChangXin Memory Technologies, China's largest DRAM maker, plans to begin producing HBM3E high-bandwidth memory for AI processors even though its yields sit at roughly 20% to 30%, according to two people with knowledge of the plans. SK hynix, the HBM market leader, is estimated to achieve yields of about 70% on the same class of chip.

CXMT expects to begin limited production within the next two months, one of the people said. The company has already sent HBM3E samples to several Chinese AI chip designers in two versions: a standard design and an updated one intended to improve yields.

The push targets a supply bottleneck. Chinese AI chipmakers have struggled to secure enough advanced memory to scale output, and Washington tightened export controls on certain advanced HBM products bound for China in December 2024. Chinese buyers reportedly pay several times more for HBM obtained through unofficial channels.

Key facts

  • 01CXMT plans to start HBM3E production with yields of roughly 20% to 30%, per two people familiar with the plans.
  • 02SK hynix is estimated to achieve HBM3E yields of about 70%.
  • 03Limited CXMT production is expected within the next two months, according to one person.
  • 04CXMT held 10% of global DRAM revenue in Q2, ranking fourth, per Counterpoint Research.
  • 05SK hynix held 50% of Q2 HBM revenue, Samsung 33% and Micron 18%.

Low yields carry a cost. With as little as 20% of its chips usable, CXMT must discard or rework many of them, and its HBM3E is expected to be priced above comparable overseas products. In testing, CXMT's HBM3E performed closer to SK hynix's previous-generation HBM3 than to SK hynix's HBM3E, the people said.

Memory costs are already moving prices across China's AI chip sector. Huawei, Cambricon, MetaX and Iluvatar CoreX raised prices on current and next-generation AI chips by 20% to 50%, citing memory costs, according to reporting last month.

CXMT began developing HBM internally as early as 2024. In conventional DRAM, it held 10% of global revenue in the second quarter, making it the fourth-largest supplier, according to Counterpoint Research.

The HBM market remains concentrated among three suppliers. SK hynix held 50% of global HBM revenue in the second quarter, Samsung 33% and Micron 18%, with most of that revenue coming from HBM3E, Counterpoint said. Samsung said in July that it had expanded HBM4 sales and shipped HBM4E samples to major customers, which leaves CXMT at least one generation behind in commercial HBM production.

CXMT did not respond to a request for comment, and SK hynix did not immediately respond to a request for comment on the yield estimate.

Related · from this week
HPE argues enterprises should own AI capacity as production workloads scale
Jaeden Schafer · 4 min read →
ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Business

HPE argues enterprises should own AI capacity as production workloads scale
Business

HPE argues enterprises should own AI capacity as production workloads scale

Deloitte data shows the share of firms with 40% of AI projects in production is set to double within six months, changing the cost math.

Jaeden Schafer4 min read
Nvidia logo
Business

China weighs letting ByteDance and Alibaba buy new Nvidia chips

A policy reversal would reopen one of Nvidia's largest export markets after months of Beijing steering domestic buyers toward local silicon.

Jaeden Schafer4 min read
Blue Cross Blue Shield says AI hospital coding added $942M in costs
Business

Blue Cross Blue Shield says AI hospital coding added $942M in costs

A BCBSA analysis blames AI-assisted claims tools for a two-year spending surge, with no matching change in care delivered.

Jaeden Schafer4 min read