智谱AI因GLM-5.2需求激增,探索自研AI芯片

Chubby♨️ · @kimmonismus · X·2026-07-07 21:13·55天前
AI 导读

智谱AI正探索定制AI芯片,以应对GLM模型带来的算力压力。据The Information,智谱已接触中国芯片设计公司,但尚未选定合作伙伴。其GLM-5.2模型在Vercel聚合器上增长最快,首周日token用量增长27倍。目前智谱依赖华为、其他国产及少量英伟达芯片,但美国出口管制、华为硬件限制及自身被列入美方黑名单,使其长期转向纯中国芯片路线,以加强国内算力独立。

Chubby♨️@kimmonismus
55AI 编辑部评分,满分 100

智谱AI因GLM-5.2需求激增,探索自研AI芯片

2026-07-07 21:13· 55天前
AI 导读

智谱AI正探索定制AI芯片,以应对GLM模型带来的算力压力。据The Information,智谱已接触中国芯片设计公司,但尚未选定合作伙伴。其GLM-5.2模型在Vercel聚合器上增长最快,首周日token用量增长27倍。目前智谱依赖华为、其他国产及少量英伟达芯片,但美国出口管制、华为硬件限制及自身被列入美方黑名单,使其长期转向纯中国芯片路线,以加强国内算力独立。

Zhipu AI is reportedly exploring a custom AI chip as demand for its GLM models starts to strain its compute supply.

The Information says the Beijing lab has made early inquiries with Chinese chip design houses, but has not selected a partner.

The trigger is clear: GLM-5.2 has reportedly been the fastest-growing model on Vercel’s aggregator since launch, with daily token usage rising as much as 27x in the first week.

Zhipu currently relies on Huawei chips, other Chinese chips and some Nvidia chips. But U.S. export controls, Huawei hardware constraints and Zhipu’s own U.S. blacklist status make a China-only chip path the obvious long-term move.

The path is clear: China and its Frontierlabs want to achieve even greater independence from Western chips and simultaneously strengthen the expansion of domestic computing.