NVIDIA 开启下一代 AI 浪潮：Rubin 架构亮相——六款全新芯片，打造卓越 AI 超算

NVIDIAIndustry:Technology⾏业：科技Website:⽹站：https://www.nvidia.com NVIDIA开启下⼀代AI浪潮：Rubin架构亮相——六款全新芯⽚，打造卓越AI超算 Press ReleaseActions新闻稿操作 Extreme Codesign Across NVIDIA VeraCPU, Rubin GPU, NVLink 6 Switch,ConnectX-9 SuperNIC, BlueField-4 DPUand Spectrum-6 Ethernet Switch SlashesTraining Time and Inference TokenGeneration Cost 涵盖NVIDIA Vera CPU、RubinGPU、NVLink 6交换机、ConnectX-9 SuperNIC、BlueField-4 DPU以及Spectrum-6以太⽹交换机的极致协同设计，⼤幅缩短训练时间并降低推理Token⽣成成本 January 05, 2026 17:20 ET| Source:NVIDIA Share 新闻摘要： News Summary: The Rubin platform harnesses extreme codesign across hardwareand software to deliver up to 10x reduction in inference token costand 4x reduction in number of GPUs to train MoE models,compared with the NVIDIA Blackwell platform. 与NVIDIA Blackwell平台相⽐，Rubin平台通过硬件和软件的深度协同设计，可将推理Token成本降低⾼达10倍，并将训练MoE（混合专家）模型所需的GPU数量减少4倍。 New NVIDIA Inference Context Memory Storage Platform withNVIDIA BlueField-4 storage processor to accelerate agentic AIreasoning. 全新的NVIDIA推理上下⽂内存存储平台搭载NVIDIABlueField-4存储处理器，旨在加速代理式AI（Agentic AI）推理。 Microsoft’s next-generation Fairwater AI superfactories — featuringNVIDIA Vera Rubin NVL72 rack-scale systems — will scale tohundreds of thousands of NVIDIA Vera Rubin Superchips. CoreWeave among first to offer NVIDIA Rubin, operated throughCoreWeave Mission Control for flexibility and performance. CoreWeave将成为⾸批提供NVIDIA Rubin的服务商之⼀，通过CoreWeave Mission Control进⾏运营，以实现灵活性和⾼性能。 与红帽（Red Hat）扩⼤合作，通过Red HatEnterprise Linux、Red Hat OpenShift和Red HatAI，为Rubin平台提供优化的完整AI堆栈。 LAS VEGAS, Jan.05, 2026(GLOBE NEWSWIRE) --CES --NVIDIAtoday kickstarted the next generation of AI with the launch of theNVIDIARubin platform, comprising six new chips designed to deliver oneincredible AI supercomputer. NVIDIA Rubin sets a new standard forbuilding, deploying and securing the world’s largest and most advancedAI systems at the lowest cost to accelerate mainstream AI adoption. 拉斯维加斯，2026年1⽉5⽇(GLOBE NEWSWIRE) --CES -- NVIDIA今⽇发布了NVIDIA Rubin平台，开启了下⼀代AI时代。该平台包含六款全新芯⽚，旨在打造⼀台卓越的AI超级计算机。NVIDIA Rubin为构建、部署和保护全球最⼤、最先进的AI系统设定了新标准，并以最低的成本加速主流AI的普及。 The Rubin platform uses extreme codesign across the six chips — theNVIDIA Vera CPU, NVIDIA Rubin GPU,NVIDIA NVLink™ 6 Switch,NVIDIA ConnectX®-9 SuperNIC,NVIDIA BlueField®-4 DPUandNVIDIASpectrum™-6 Ethernet Switch— to slash training time and inferencetoken costs. Rubin平台通过对六款芯⽚——NVIDIA Vera CPU、NVIDIA Rubin GPU、NVIDIA NVLink™ 6交换机、NVIDIA ConnectX®-9 SuperNIC、NVIDIA BlueField®-4DPU以及NVIDIA Spectrum™-6以太⽹交换机——进⾏极致的协同设计，从⽽⼤幅缩短训练时间并降低推理Token成本。 “Rubin arrives at exactly the right moment, as AI computing demand forboth training and inference is going through the roof,” said JensenHuang, founder and CEO of NVIDIA. “With our annual cadence ofdelivering a new generation of AI supercomputers — and extremecodesign across six new chips — Rubin takes a giant leap toward thenext frontier of AI.” “Rubin的到来正值关键时刻，因为AI计算在训练和推理⽅⾯的需求都在⻜速增⻓，”NVIDIA创始⼈兼⾸席执⾏官⻩仁勋表示。“随着我们以每年⼀代的速度交付新⼀代AI超级计算机，并对六款新芯⽚进⾏极致的协同设计，Rubin向AI的下⼀个前沿迈出了巨⼤的⼀步。” Named for Vera Florence Cooper Rubin — the trailblazing Americanastronomer whose discoveries transformed humanity’s understanding ofthe universe — the Rubin platform features theNVIDIA Vera RubinNVL72rack-scale solution and theNVIDIA HGX Rubin NVL8system. Rubin平台以先驱性的美国天⽂学家薇拉·弗洛伦斯·库珀·鲁宾（Vera Florence Cooper Rubin）的名字命名，她的发现改变了⼈类对宇宙的认知。该平台包含NVIDIA VeraRubin NVL72机架级解决⽅案和NVIDIA HGX RubinNVL8系统。 The Rubin platform introduces five innovations, including the latestgenerations of NVIDIA NVLink interconnect technology, TransformerEngine, Confidential Computing and RAS Engine, as well as the NVIDIAVera CPU. These breakthroughs will accelerate agentic AI, advancedreasoning and massive-scalemixture-of-experts(MoE) model inferenceat up to 10x lower cost per token of the NVIDIA Blackwell platform.Compared with its predecessor, the NVIDIA Rubin platform trains MoEmodels with 4x fewer GPUs to accelerate AI adoption. Rubin平台引⼊了五项创新技术，包括最新⼀代的NVIDIANVLink互连技术、Transformer引擎、机密计算和RAS引擎，以及NVIDIA Vera CPU。这些突破将加速代理型⼈⼯智能（agentic AI）、⾼级推理以及⼤规模混合专家（MoE）模型推理，其每Token成本⽐NVIDIA Blackwell平台降低了多达10倍。与前代产品相⽐，NVIDIA Rubin平台训练MoE模型所需的GPU数量减少了4倍，从⽽加速了⼈⼯智能的普及应⽤。 Broad Ecosystem Support⼴泛的⽣态系统⽀持 Among the world’s leading AI labs, cloud service providers, computermakers and startups expected to adopt Rubin are Amazon WebServices (AWS), Anthropic, Black Forest Labs, Cisco, Cohere,CoreWeave, Cursor,Dell Technologies, Google, Harvey, HPE, Lambda,Lenovo, Meta, Microsoft, Mistral AI, Nebius, Nscale, OpenAI, OpenEvidence, Oracle Cloud Infrastructure (OCI), Perplexity, Runway,Supermicro, Thinking Machines Lab and xAI. 在全球领先的AI实验室、云服务提供商、计算机制造商和初创公司中，预计采⽤Rubin的包括：亚⻢逊云科技(AWS)、Anthropic、Black Forest Labs、思科(Cisco)、Cohere、CoreWeave、Cursor、戴尔科技(DellTechnologies)、⾕歌(Google)、Harvey、HPE、Lambda、联想(Lenovo)、Meta、微软(Microsoft)、Mistral AI、Nebius、Nscale、OpenAI、OpenEvidence、甲⻣⽂云基础设施(OCI)、Perplexity、Runway、美超微(Supermicro)、Thinking MachinesLab以及xAI。 Sam Altman, CEO of OpenAI:“Intelligence scales with compute.