DeepSeek造芯进展曝出!梁文锋3年前或已战略布局 - 智东西
Quick Answer
DeepSeek is developing its own AI inference chip to reduce reliance on Nvidia and Huawei, having initiated this project about a year ago.
Quick Take
DeepSeek is developing its own AI inference chip to reduce reliance on Nvidia and Huawei, having initiated this project about a year ago. The company is currently in early stages of development and is actively recruiting chip design engineers while preparing for the release of its DeepSeek V4 model in mid-July.
Key Points
- DeepSeek's chip development aims to enhance model efficiency and reduce external supply dependency.
- The company has increased hiring for chip design engineers in recent months.
- DeepSeek V4 model is set to launch in mid-July with a new pricing strategy.
- Current models utilize both Nvidia and Huawei chips, with a shift towards Huawei's Ascend chips.
- Self-developed chips are becoming a trend among leading AI model companies globally.
📖 Reader Mode
~1 min read
智东西(公众号:zhidxcom)
编译 | 陈骏达
编辑 | 李水青
智东西7月7日消息,今天,据路透社报道,三位知情人士透露,DeepSeek正在自研AI推理芯片,以减少对英伟达以及华为芯片的依赖。
知情人士称,DeepSeek自研芯片的工作大约在一年前启动,其研发仍处于早期阶段。目前,DeepSeek正在接触外部合作伙伴,并与芯片设计企业、晶圆代工厂以及存储企业展开讨论。
知情人士还透露,DeepSeek近几个月增加了芯片设计工程师的招聘数量,但相关招聘并未通过公开招聘平台发布,而是在私下进行。
DeepSeek方面拒绝了路透社的置评请求。
DeepSeek对算力与芯片的关注由来已久。在2023年、2024年两次接受暗涌采访时,DeepSeek创始人兼CEO梁文锋就曾说过:“我们真正的挑战从来不是资金,而是高端芯片的出口禁令”,“对研究员来说,对算力的渴求是永无止境的”,“我们会有意识地去部署尽可能多的算力”。
目前,DeepSeek正同时使用英伟达和华为芯片。英伟达H800曾是支撑不少DeepSeek模型训练的底座,但美国政府在2023年底禁止其出口。
此后,DeepSeek逐渐加大对昇腾芯片的使用。
今年4月,DeepSeek发布了适配昇腾芯片的V4模型。华为称,昇腾芯片参与了V4-Flash轻量版本模型的部分训练过程。路透社此前还曾报道,该模型发布后,中国科技企业对华为昇腾950芯片的订单大幅增加。
DeepSeek布局芯片之际,其新一代模型DeepSeek V4正式版也已经呼之欲出。上周,DeepSeek向API客户发送了邮件,称DeepSeek V4正式版计划于7月中旬发布,届时API将采用峰谷定价,高峰时段价格翻倍,非高峰时段维持原价。

在国内,有用户反馈DeepSeek V4正式版已经开始灰度测试,其编程能力似乎有明显提升。

▲疑似DeepSeek V4正式版生成的《我的世界》(图源:Bilibili@冬眠の松鼠_)
结语:自研芯片成全球头部大模型厂商共识
在当前AI应用落地不断加速、模型架构相对保持稳定的背景下,打造为推理任务专门设计的芯片正成为头部大模型厂商的共识。
在海外,OpenAI已经与博通合作开发首款自研推理芯片,Anthropic也被曝正在评估自研AI芯片的可能性。对于这些大模型公司而言,自研芯片的核心目的并非完全替代外部方案,而是希望针对自身模型和应用场景进行硬件优化,在推理阶段降低成本、提升能效,并增强供应链掌控能力。
DeepSeek此次布局自研推理芯片,本质上也是顺应这一趋势。通过软硬件协同优化,DeepSeek有望进一步降低对外部芯片供应的依赖,同时提升其模型的推理效率。
来源:路透社
— Originally published at m.zhidx.com
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from WebSearch (Tavily)
See more →Stop just chatting with AI. Learn to build production-ready software in ...
The 2026 Bootcamp offers hands-on training in building production-ready software using Generative AI, LLM applications, and AI agents, emphasizing practical skills over casual interaction with AI. Participants will learn to develop applications like Cursor AI, preparing them for real-world challenges in AI development.


