Brief IA
›

CXMT Launches HBM3E with 3 to 5 Years Delay and 25% Yield

Models & LLM·Tom Levy·

CXMT Launches HBM3E with 3 to 5 Years Delay and 25% Yield

CXMT Launches HBM3E with 3 to 5 Years Delay and 25% Yield
Key Takeaways
1CXMT is facing a technical delay of 3 to 5 years and estimated yields of 25% as of June.
2T-Head and Cambricon plan to integrate HBM3E into products starting in 2027.
3CXMT is producing HBM3E for the first time, while Samsung, SK Hynix, and Micron are already manufacturing HBM4.
Why it matters : Local production of HBM3E could help China circumvent U.S. restrictions on advanced HBM.

The Chinese manufacturer CXMT has produced HBM3E memory for the first time, which is essential for AI, but it remains behind global leaders and is facing production difficulties. Chinese designers are planning integration starting in 2027, while U.S. restrictions limit access to advanced HBM.

CXMT: Technical Delays, Low Yields, and Lack of Dedicated Funding

CXMT is reportedly facing a technological delay of three to five years and is encountering manufacturing issues, with yield rates estimated at nearly 25% in June according to some analyses. Upon its IPO in Shanghai, the company raised $8.6 billion, but the prospectus indicates that none of this amount is allocated to HBM.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Expected Adoption by T-Head and Cambricon Amid U.S. Restrictions

Alibaba's T-Head and Cambricon are currently testing this memory and plan to integrate it into products starting in 2027. U.S. export rules limit the purchase of advanced HBM by China, making local production strategic for domestic players.

Initial HBM3E Production at CXMT, a Generation Behind Leaders

CXMT is producing HBM3E in small quantities for the first time, according to two internal sources cited. This advancement places CXMT a generation behind Samsung, SK Hynix, and Micron, which are already mass-producing HBM4. HBM is a stacked memory positioned close to the processor, necessary for training and executing large AI models.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.