英伟达这次把六家供应链伙伴全锁定了,连京元电子都拿到 11 亿美元预付款,想看 AI 芯片产能瓶颈怎么解的可以细读。
NVIDIA英伟达
The main supplier of GPUs for AI training and inference
What they do
Headquartered in Santa Clara, California, NVIDIA's GPUs and CUDA software stack are the main source of compute for training and serving large models; it also sells data-centre networking and platforms for autonomous driving and robotics.
Latest coverage
Stories in the feed that mention this companyPerplexity 和 NVIDIA 搞了个叫 JOLT 的训练方法,一个模型当老师又当学生,在 MATH 上准度追平 GRPO,训练 token 少了 6.5 倍。
数据中心公司 DayOne 拿 Nvidia 的 AI 工厂设计直接套用,建机房更快、跨市场标准统一,搞算力基建的可以看看。
163 MB 的语音转写模型,M 系列芯片上半秒转完 5 分钟音频,错字率还比 Whisper 低,Clean 版连“嗯啊”都帮你删了。
Siemens 把 Nvidia Cosmos 接进自家数字孪生软件,生产前就能做完电路板光学检测准备,省掉不少产线等待时间。
黄仁勋说 GPU 能用 8 年,马斯克说 H100 租金比新时还贵,几个人从不同角度聊算力为什么越老越值钱。
英伟达把检索三件套 Parse、Embed、Rerank 全开源了,扫描 PDF 和图表都能解析,准确率上去了成本还降了 14%,做 RAG 的可以试试。
Nvidia 和 Vast Data 弄了个 OpenShell 沙盒,专门给 AI 智能体跑隔离执行,还能跨会话保存状态,做智能体开发的可以看看。
Pre-configured reference hardware could cut weeks of environment setup to days, lowering the barrier for labs that want to experiment with general-purpose robots rather than build their own stacks. Research teams at universities and corporate labs are the target buyers, with Stanford and ETH Zurich already expressing interest. Regional availability and pricing for Southeast Asian buyers have not been announced, so early access likely runs through Agile Robots' existing research channels.
西门子医疗和英伟达要把智能体 AI 塞进放疗流程了,用数字孪生在治疗前先模拟方案,医生能边治疗边实时调整计划。