这段文字是与英伟达 (NVIDIA) 首席执行官黄仁勋 (Jensen Huang) 的一次对话记录，重点讨论了英伟达的发展历程、技术创新以及人工智能和加速计算的广泛影响。它涉及英伟达从一家以游戏为中心的 GPU 公司向数据中心行业提供各种硬件和软件的供应商的演变。 黄仁勋首先讨论了英伟达的愿景，即用一种新的计算形式（特别是图形处理）来增强通用计算。该公司战略性地将其应用扩展到计算机图形之外，包括物理模拟和科学计算。一个关键的决定是保持跨代架构的兼容性，从而培养庞大的安装基础并激励围绕英伟达 CUDA 平台的生态系统，确保长期的软件开发者支持。 随后，讨论转向通用计算机和加速计算机之间的核心差异。与 CPU 不同，加速计算需要仔细选择要加速的内容。英伟达通过创建针对特定领域的丰富库来解决这个问题，例如自动驾驶汽车、机器人技术和气候科学。英伟达的方法是识别占运行时 5-10% 的代码并对其进行优化，从而显著提高速度。 黄仁勋随后强调了英伟达对数据中心市场的关注。他批评了充满空气的低效传统数据中心，并提倡密度和能源效率。他强调了计算能力呈指数级增长及其对各个行业的影响。他设想人工智能将扩展到数据中心之外，进入日常技能，例如自动驾驶、机器人和虚拟助手，并将它们称为“数字人”。 在讨论人工智能基础设施中客户的投资回报率时，黄仁勋承认了关于商业模式转变和采用周期的争论。他指出，英伟达已经从只与主要参与者合作转变为赋能较小的公司和团体。 谈话转向创新和竞争的话题。黄仁勋坚称英伟达致力于快速创新，并计划每年向市场推出一个新的 AI 超级集群。他讨论了英伟达 Blackwell 架构的技术优势，强调它将在降低成本的同时提高性能吞吐量。他还提到了英伟达提供高性能和高性价比解决方案的策略，以满足广泛的客户需求。 黄仁勋还谈到了地缘政治方面的考量，强调了设计多样性、冗余和供应链敏捷性的重要性。他强调了英伟达的合作伙伴台积电 (TSMC) 在扩大生产规模方面的作用，并保证英伟达拥有在必要时转移制造的应急计划。 黄仁勋随后详细介绍了英伟达与各个行业和组织的广泛合作，包括游戏、云服务、计算机制造商和研究机构。他承认作为关键技术提供商所带来的责任，并理解许多客户对英伟达成功的情感投入。 在回应关于扩大规模和满足需求的担忧时，黄仁勋表达了效仿美国工业先驱亨利·福特 (Henry Ford) 和安德鲁·卡耐基 (Andrew Carnegie) 的愿望，旨在扩大生产规模以满足客户的需求。他表示，对其产品的巨大需求可能会产生令人难以置信的意图。 黄仁勋最后表达了他对人工智能在各个领域潜力的兴奋，并表示这是他的一部分，其他人正在利用它来产生更多创新设计并解决各种复杂问题。他强调看到那些机器人走来走去真是不可思议。 讨论还提到了本杰明·格雷厄姆 (Benjamin Graham) 及其在所在领域的成功。他还说，他可以再花 1/2 的资金，必须停止。非常感谢您的时间和出席。

This is a transcript of a conversation with Jensen Huang, the CEO of NVIDIA, focusing on the company's journey, its technological innovations, and the broader implications of AI and accelerated computing. It touches upon NVIDIA's evolution from a gaming-centric GPU company to a broad-range hardware and software provider for the data center industry. Huang begins by discussing NVIDIA's vision of augmenting general-purpose computing with a new form of computing, specifically graphics processing. The company strategically expanded its applications beyond computer graphics to physics simulation and scientific computing. A pivotal decision involved maintaining architectural compatibility across generations, fostering a large installed base and incentivizing an ecosystem around NVIDIA's CUDA platform, ensuring long-term software developer support. The discussion then shifts to the core differences between general-purpose and accelerated computers. Unlike CPUs, accelerated computing requires carefully selecting what to accelerate. NVIDIA addressed this by creating rich libraries tailored for specific domains like self-driving cars, robotics, and climate science. NVIDIA's approach is to identify the 5-10% of code that dominates runtime and optimize it, leading to significant speed improvements. Huang then emphasizes NVIDIA's focus on the data center market. He criticizes inefficient traditional data centers filled with air and promotes densification and energy efficiency. He highlights the exponential increase in computing power and its impact on various industries. He envisions AI expanding beyond data centers into everyday skills, such as autonomous driving, robotics, and virtual assistants, referring to them as "digital humans." Discussing customer ROI in the AI infrastructure, Huang acknowledges the debate on business model shifts and adoption cycles. He points out that NVIDIA has transitioned from solely working with major players to enabling smaller companies and groups. The conversation moves to the topic of innovation and competition. Huang asserts NVIDIA's commitment to rapid innovation, with plans to release a new AI supercluster to market every year. He discusses the technological advantages of NVIDIA's Blackwell architecture, emphasizing that it would enhance performance throughput while reducing cost. He also mentions NVIDIA's strategy of offering both high-performance and cost-effective solutions, catering to a wide range of customer needs. Huang also addresses geopolitical considerations, emphasizing the importance of design diversity, redundancy, and supply chain agility. He highlights the role of NVIDIA's partner, TSMC, in scaling up production, and assures that NVIDIA has contingency plans to shift fabrication if necessary. Huang then details NVIDIA's extensive collaborations with various industries and organizations, including gaming, cloud services, computer manufacturers, and research institutions. He acknowledges the responsibility that comes with being a crucial technology provider and understands the emotional investment that many customers have in NVIDIA's success. In response to concerns about scaling and fulfilling demand, Huang expresses a desire to emulate American industrial pioneers like Henry Ford and Andrew Carnegie, aiming to scale production to meet the demands of its customers. He states that the extraordinary demand for its products will potentially generate incredible intentions. Huang concludes by expressing his excitement about the potential of AI in various fields, and that is his part and others are using it to produce more innovative designs and solve various complex problems. He highlighted that it is incredible to see those Robots walking around. The discussion touches on Benjamin Graham and his success in his area. He also said that he could spend another 1/2 of our funds that have got to stop. Thank you very much for your time and attendance.