英伟达在 2026 年 CES 上放大招：AI 不再只在屏幕里，它要活过来了！
NVIDIA’s Big Move at CES 2026: AI Is No Longer Just on Screens, It’s Coming to Life!
科技界的朋友们，准备好迎接一场思维的巨浪了吗？英伟达创始人黄仁勋在 2026 年 CES 上的主题演讲，如同一颗划破夜空的流星，照亮了人工智能发展的新方向。他掷地有声地宣告，AI 正经历一场深刻的变革，从局限于屏幕之内智能，演变为能够感知、理解并操控物理世界的“物理智能（Physical AI）”。这绝非空中楼阁式的畅想，而是英伟达依托于芯片革新与生态构建的宏伟蓝图。今天，让我们一同拨开迷雾，探寻这场科技革命背后的深层逻辑。
Friends in the tech world, are you ready for a tidal wave of new ideas? NVIDIA founder Jensen Huang’s keynote speech at CES 2026 was like a shooting star piercing the night sky, illuminating a new direction for AI development. He emphatically declared that AI is undergoing a profound transformation, evolving from intelligence confined within screens to “Physical AI” capable of perceiving, understanding, and manipulating the physical world. This is not a fanciful notion but NVIDIA’s grand blueprint, built upon chip innovation and ecosystem development. Today, let’s peel back the layers and explore the deeper logic behind this technological revolution.
一、 “物理智能”：AI 发展的必然之路？
I. “Physical AI”: The Inevitable Path for AI Development?
长久以来，我们习惯于将 AI 视为一种虚拟的存在，它存在于算法之中，服务于屏幕之后。然而，“物理智能”的概念打破了这一固有认知，它预示着 AI 将拥有“身体”，能够与真实世界进行直接的互动。设想一下，未来的机器人不再是按照预设程序运行的机器，而是拥有自主思考能力，能够像人类一样学习、适应环境的智能体。
For a long time, we have been accustomed to viewing AI as a virtual entity, existing within algorithms and serving behind screens. However, the concept of “Physical AI” shatters this inherent perception, foreshadowing that AI will possess a “body,” capable of direct interaction with the real world. Imagine future robots that are no longer machines operating on preset programs, but intelligent entities with autonomous thinking capabilities, able to learn and adapt to environments just like humans.
1. 从虚拟到现实：AI 进化方向的共识
1. From Virtual to Real: The Consensus on AI’s Evolutionary Direction
无论是学术界的严谨定义，还是工业界的实践探索，亦或是媒体的广泛报道，对“物理智能”的理解已趋于一致：它是一种将 AI 算法与物理系统（如机器人、传感器）紧密结合的智能形态，赋予机器在真实环境中进行实时决策，乃至像人类一样学习和行动的能力。这种共识并非偶然，它反映了 AI 发展的内在需求。
Whether it’s rigorous academic definitions, practical explorations in industry, or widespread media reports, the understanding of “Physical AI” has converged: it is an intelligent form that closely integrates AI algorithms with physical systems (such as robots, sensors), empowering machines with the ability to make real-time decisions and even learn and act like humans in real environments. This consensus is not accidental; it reflects the intrinsic needs of AI development.
2. 解决现实难题：物理 AI 的价值所在
2. Solving Real-World Problems: The Value of Physical AI
物理 AI 的出现，并非为了追逐新概念，而是为了解决传统 AI 难以逾越的现实难题。以自动驾驶为例，仅仅依靠模拟数据进行训练是远远不够的，AI 必须在真实的道路上进行试错、学习，才能真正应对复杂的交通状况。这种“脚踏实地”的 AI，将在制造业、物流、医疗等领域引发颠覆性的变革，推动社会生产力的巨大飞跃。
The emergence of Physical AI is not merely about chasing new concepts, but about solving real-world problems that traditional AI struggles to overcome. Taking autonomous driving as an example, training solely on simulated data is far from enough; AI must undergo trial and error and learn on real roads to truly cope with complex traffic conditions. This “down-to-earth” AI will trigger disruptive changes in manufacturing, logistics, healthcare, and other fields, driving a significant leap in social productivity.
二、 英伟达的“肌肉”：Vera Rubin 架构解析
II. NVIDIA’s “Muscle”: Dissecting the Vera Rubin Architecture
要赋予 AI 以“身体”，强大的算力是不可或缺的基石。英伟达推出的 Vera Rubin 架构，堪称专为物理 AI 量身打造的“超跑引擎”，其核心在于“极端协同设计”（Extreme Co-design）。这并非单一芯片的升级，而是由六大芯片组成的“梦之队”，共同构建起 AI 算力的强大支撑。
To give AI a “body,” powerful computing power is an indispensable cornerstone. NVIDIA’s Vera Rubin architecture, an “engine like a supercar” tailor-made for Physical AI, is centered on “Extreme Co-design.” This is not an upgrade of a single chip but a “dream team” composed of six major chips, collectively building a robust foundation for AI computing power.
1. 性能怪兽：芯片家族的核心奥秘
1. Performance Monster: The Core Secret of the Chip Family
- Vera CPU：专为 AI 工作负载设计的 CPU，拥有 88 个自研核心，能够高效地进行数据处理，为 GPU 提供充足的“养分”。
Vera CPU: A CPU designed specifically for AI workloads, featuring 88 proprietary cores, capable of efficient data processing to provide ample “nourishment” for the GPU.
- Rubin GPU：AI 算力的核心引擎，其推理性能相较于上一代 Blackwell 架构提升了 5 倍，训练性能提升了 3.5 倍。这种性能的飞跃，将极大地加速 AI 模型的开发和部署。
Rubin GPU: The core engine for AI computing power, its inference performance has increased by 5 times and training performance by 3.5 times compared to the previous Blackwell architecture. This leap in performance will significantly accelerate the development and deployment of AI models.
- ConnectX-9 网卡：每颗 GPU 独享高达 1.6 TB/s 的高速通道，确保数据传输畅通无阻，避免“交通堵塞”。
ConnectX-9 Network Card: Each GPU enjoys an exclusive high-speed channel of up to 1.6 TB/s, ensuring smooth data transmission and avoiding “traffic jams.”
- BlueField-4 DPU：扮演着“大管家”的角色，负责网络、安全、存储等关键任务，同时也是无限上下文存储的核心组成部分。
BlueField-4 DPU: Acts as a “steward,” responsible for critical tasks such as networking, security, and storage, while also being a core component of infinite context storage.
- NVLink Switch（第六代）：实现机架内部 72 颗 GPU 的协同工作，如同一个超级大脑，共同处理复杂的 AI 任务。
NVLink Switch (6th Gen): Enables 72 GPUs within a single rack to work collaboratively, like a super brain, jointly processing complex AI tasks.
- Spectrum-X 以太网交换机：连接多个机架，支持数千个机架组成的 AI 工厂的调度，实现大规模的 AI 计算。
Spectrum-X Ethernet Switch: Connects multiple racks, supporting the orchestration of AI factories composed of thousands of racks to enable large-scale AI computing.
2. 绿色节能：液冷技术的哲学思考
2. Green Energy Saving: The Philosophy Behind Liquid Cooling Technology
英伟达此次还大胆采用了 100% 液冷系统，取消了所有风扇和电缆，利用 45°C 温水进行冷却。此举不仅能够节省数据中心 6% 的电力，降低运营成本，更体现了英伟达对可持续发展的深刻思考。
NVIDIA also boldly adopted a 100% liquid cooling system, eliminating all fans and cables, using 45°C warm water for cooling. This move not only saves 6% of data center electricity and reduces operating costs but also reflects NVIDIA’s deep consideration for sustainable development.
三、 AI 的“灵魂”：Cosmos 世界模型构建
III. AI’s “Soul”: Building the Cosmos World Model
仅仅拥有强大的“身体”是远远不够的，AI 还需要一个能够理解物理规律的“大脑”。英伟达的 Cosmos 世界模型，正是赋予 AI “常识”的关键所在。
Merely possessing a powerful “body” is far from enough; AI also needs a “brain” capable of understanding physical laws. NVIDIA’s Cosmos World Model is precisely the key to endowing AI with “common sense.”
1. 合成数据：AI 训练的新范式
1. Synthetic Data: A New Paradigm for AI Training
Cosmos 的核心逻辑在于，如果真实数据不足，那就创造合成数据！通过 “训练机-模拟机（Omniverse）-推理机” 的三机协作模式，让 AI 在符合物理规律的虚拟世界中进行数十亿次的试错。这相当于为 AI 打造了一个永不疲倦、无所不能的“驾校”，使其能够快速掌握各种技能。
The core logic of Cosmos is: if real data is insufficient, then create synthetic data! Through a “Trainer – Simulator (Omniverse) – Inferencer” three-machine collaboration model, AI conducts billions of trials and errors in a virtual world that adheres to physical laws. This is equivalent to building a tireless, omnipotent “driving school” for AI, enabling it to quickly master various skills.
2. 构建 AI 的世界观：理解重力、因果
2. Building AI’s Worldview: Understanding Gravity, Causality
Cosmos 不仅能够让 AI 感知世界，还能够让其理解世界的运作规律，例如重力、物体恒存性和因果关系。这意味着 AI 将不再仅仅是 “看”，更能够 “懂”，从而做出更加智能的决策。
Cosmos not only enables AI to perceive the world but also to understand how the world operates, such as gravity, object permanence, and causality. This means AI will no longer merely “see” but also “understand,” leading to more intelligent decisions.
四、 落地应用：AI 正在改变真实世界
IV. Real-World Applications: AI is Changing the Real World
英伟达的物理 AI 并非纸上谈兵，而是已经开始在真实世界中落地生根。
NVIDIA’s Physical AI is not just theoretical; it has already begun to take root in the real world.
1. Alpha Mio 自动驾驶：会“思考”的汽车
1. Alpha Mio Autonomous Driving: Cars That “Think”
自动驾驶是物理 AI 最先落地的场景之一。英伟达发布了全球首款具备 推理能力 的自动驾驶系统 Alpamayo (Alpha Mio)，并将于 2026 年搭载在梅赛德斯-奔驰 CLA 车型上。
Autonomous driving is one of the first scenarios where Physical AI is being implemented. NVIDIA released the world’s first autonomous driving system with reasoning capabilities, Alpamayo (Alpha Mio), which will be integrated into Mercedes-Benz CLA models in 2026.
- 端到端：完全由数据驱动，无需预设规则，更加灵活和智能。
End-to-End: Completely data-driven, no pre-set rules required, making it more flexible and intelligent.
- System 2 Thinking：像人类一样进行 “思考” 和预判，解决传统自动驾驶难以处理的 “长尾难题”。
System 2 Thinking: “Thinks” and anticipates like humans, solving “long-tail problems” that are difficult for traditional autonomous driving to handle.
- 双栈冗余设计：智能驾驶和传统安全系统双重保障，确保安全系数达到极致。
Dual-Stack Redundancy Design: Dual safeguards of intelligent driving and traditional safety systems ensure maximum safety.
2. 机器人与工业元宇宙：未来工厂的雏形
2. Robotics and the Industrial Metaverse: The Blueprint for Future Factories
从人形机器人到手术机器人，再到工业机械臂，都将基于 NVIDIA Isaac 平台和 Cosmos 模型进行训练。英伟达还与西门子深度合作，通过 Omniverse 模拟制造数字孪生，以应对全球劳动力短缺的挑战，利用 AI 机器人实现自动化生产。
From humanoid robots to surgical robots, and industrial robotic arms, all will be trained based on the NVIDIA Isaac platform and the Cosmos model. NVIDIA is also deeply collaborating with Siemens, using Omniverse to simulate digital twins for manufacturing, addressing the global labor shortage challenge by leveraging AI robots for automated production.
五、 突破瓶颈：无限上下文存储平台
V. Breaking Bottlenecks: Infinite Context Storage Platform
AI 常常面临 “记不住” 或 “记起来太贵” 的问题，这限制了其处理复杂、长期任务的能力。英伟达的无限上下文存储平台，正是为了解决这一难题。
AI often faces the problem of “not remembering” or “remembering being too expensive,” which limits its ability to handle complex, long-term tasks. NVIDIA’s infinite context storage platform is designed to solve this very problem.
1. BlueField-4 DPU：AI 的长效记忆
1. BlueField-4 DPU: AI’s Long-Term Memory
通过 BlueField-4 DPU，在 GPU 显存和传统硬盘之间构建了第三层存储，其速度接近显存，容量却如同大海般广阔。
Through the BlueField-4 DPU, a third layer of storage is built between GPU memory and traditional hard drives, with speeds approaching that of VRAM but capacity as vast as the ocean.
2. 从对话工具到智能 Agent
2. From Conversational Tools to Intelligent Agents
这意味着 AI 将拥有长期的 “工作记忆”，不再是 “一次性对话工具”，而是能够长期跟踪项目、持续思考的 “智能 Agent”。
This means AI will possess long-term “working memory,” no longer being a “one-off conversational tool,” but rather an “intelligent agent” capable of tracking projects long-term and continuous thinking.
六、 生态战略：开源与赋能
VI. Ecosystem Strategy: Open Source and Empowerment
英伟达不仅致力于打造自身的 AI 巨擘地位，还积极拥抱开源，开放大量的预训练模型（NIMs）和数据集，涵盖生物学、气象学（Earth-2）和机器人（Project GR00T）等领域。英伟达正从一家芯片供应商转型为 模型构建者和生态赋能者，鼓励更多的开发者参与到物理 AI 的建设中来。
NVIDIA is not only committed to establishing its own giant status in AI but also actively embraces open source, opening up a large number of pre-trained models (NIMs) and datasets, covering fields such as biology, meteorology (Earth-2), and robotics (Project GR00T). NVIDIA is transforming from a chip supplier into a model builder and ecosystem enabler, encouraging more developers to participate in the construction of Physical AI.
七、 争议与挑战：高速发展下的冷静思考
VII. Controversies and Challenges: Sober Reflection Amidst Rapid Development
英伟达在物理 AI 领域的高速发展也伴随着一些争议和挑战：
NVIDIA’s rapid development in the field of Physical AI also comes with several controversies and challenges:
1. 伦理与安全：AI 决策的“黑箱”问题
1. Ethics and Safety: The “Black Box” Problem of AI Decisions
随着 AI 系统在物理世界中扮演着越来越重要的角色，其决策的透明度、公平性以及潜在的偏见问题日益凸显。自动驾驶的安全问题、AI 在医疗领域的应用伦理，都需要我们进行审慎的思考。
As AI systems play increasingly crucial roles in the physical world, issues of decision transparency, fairness, and potential biases are becoming more prominent. The safety concerns of autonomous driving and the ethical implications of AI applications in healthcare all require careful consideration.
2. 市场垄断与竞争：平衡的艺术
2. Market Monopoly and Competition: The Art of Balance
英伟达在 AI 芯片市场占据着主导地位，引发了关于垄断和反竞争行为的担忧。高昂的 AI GPU 价格，以及其专有的 CUDA 生态系统，都给竞争对手和客户带来了压力。如何维护市场的公平竞争，避免一家独大的局面，是摆在我们面前的重要课题。
NVIDIA’s dominant position in the AI chip market has raised concerns about monopoly and anti-competitive practices. The high prices of AI GPUs and its proprietary CUDA ecosystem put pressure on competitors and customers alike. Maintaining fair competition in the market and avoiding a monopolistic situation is a significant challenge before us.
3. 技术鸿沟与能源消耗：可持续发展的远见
3. Technological Divide and Energy Consumption: A Vision for Sustainable Development
训练和运行物理 AI 需要巨大的计算能力和能源，如何降低能耗、实现可持续发展是未来的重要课题。同时，全球劳动力市场可能因为自动化而面临转型，如何应对技术鸿沟带来的社会影响也值得我们深思。
Training and running Physical AI require enormous computing power and energy. How to reduce energy consumption and achieve sustainable development is a crucial issue for the future. Meanwhile, the global labor market may face transformation due to automation, and how to address the social impact of the technological divide also warrants deep consideration.
八、 未来展望：物理 AI 的无限可能
VIII. Future Outlook: The Infinite Possibilities of Physical AI
英伟达的 CES 2026 演讲，无疑开启了 “物理 AI” 的新篇章。未来，随着 Vera Rubin 架构的成熟、Cosmos 模型对物理世界的更深理解，以及无限上下文存储的普及，我们将会看到：
NVIDIA’s CES 2026 presentation undoubtedly ushered in a new chapter for “Physical AI.” In the future, with the maturation of the Vera Rubin architecture, a deeper understanding of the physical world by the Cosmos model, and the popularization of infinite context storage, we will see:
- 更智能的自动系统：自动驾驶汽车将更加安全、可靠，机器人将能处理更复杂的任务，在工厂、仓库甚至家庭中扮演更重要的角色。
Smarter Autonomous Systems: Autonomous vehicles will become safer and more reliable, and robots will be able to handle more complex tasks, playing more significant roles in factories, warehouses, and even homes.
- 工业元宇宙的普及：数字孪生将与 AI 深度融合，实现更高效、更灵活的智能制造和运营。
Popularization of the Industrial Metaverse: Digital twins will be deeply integrated with AI, enabling more efficient and flexible intelligent manufacturing and operations.
- 人机协作新模式：AI 将成为人类的智能助手，帮助我们解决难题，提升生活品质。
New Human-Machine Collaboration Models: AI will become an intelligent assistant to humans, helping us solve problems and improve the quality of life.
正如黄仁勋所说，这正是 “机器人的 ChatGPT 时刻”。AI 不再是屏幕里的一个虚无概念，它正在拥有 “身体” 和 “灵魂”，走向真实世界，并将彻底改变我们的生活和工作方式。让我们拭目以待，物理 AI 将如何重塑我们的未来！
As Jensen Huang stated, this is “the ChatGPT moment for robotics.” AI is no longer a vague concept on a screen; it is acquiring a “body” and “soul,” moving into the real world, and will profoundly change the way we live and work. Let us await to see how Physical AI will reshape our future!
Leave a Reply