Amazon CEO Andy Jassy on the $220 Billion AI Buildout

在这期于2026年10月由本主持的播客节目中，亚马逊首席执行官安迪·贾西探讨了亚马逊庞大的资本支出投资、对人工智能的爆发式需求，以及该公司长期的战略雄心。 回顾自2021年接任首席执行官以来的五年，贾西指出，亚马逊的整体营收大约高出80%，AWS业务规模几乎扩大了三倍，广告收入增长了大约两倍半。早期，在经历了大流行期间物流网络规模翻倍后所带来的服务成本挑战时，亚马逊将其美国网络进行了区域化改造。这一意外之喜释放了配送速度和零售需求。不久之后，生成式人工智能浪潮加速到来，推动AWS的年度营收运行率达到1690亿美元，同比增长36.7%。 对话的一个核心主题是亚马逊庞大的资本支出。贾西解释说，亚马逊预计今年将在资本支出上花费2200亿美元，相当于每天6亿美元。尽管有这笔非同寻常的投资，亚马逊仍然留下了可观的未满足需求，2026年和2027年的产能基本售罄，2028年的相当一部分也已被预定。贾西回应了金融市场对是否有足够资金来资助这些基础设施建设的担忧。他辩称，国家和世界承受不起不建造数据中心的代价，他将这种规模比作20世纪50年代的高速公路和桥梁建设，并警告说，落后于中国等其他国家并不是一个可选的选项。 在谈到人工智能需求时，贾西将市场描述为“哑铃型”的。一端是消耗大量计算资源的人工智能实验室以及像ChatGPT和Claude这样失控的应用；另一端是将人工智能用于生产力和成本节约的企业。哑铃的中间部分——使用推理的当前企业生产工作负载——仍然占据未来增长的最大绝对部分，尽管广泛的企业采用在很大程度上依赖于降低推理成本。 贾西强调了在2015年收购的安纳普尔纳实验室在开发定制芯片（如Graviton CPU和Tranium人工智能加速器）方面所起的关键作用。这些芯片提供了卓越的性价比，并有助于缓解供应链限制。他还谈到了亚马逊对Anthropic和OpenAI等公司的战略合作伙伴关系和数十亿美元的投资，并指出这些团队正在构建该领域中最成功且发展最快的技术。 在组织设计和未来工作方面，贾西分享了人工智能工具（如Bedrock团队在76天内重建其推理引擎）如何缩短团队规模并加速原型设计。他预测，虽然会有新的工作岗位出现，但组织在传统任务上将使用更少的人类，并且公司将需要重写其代码库以进行智能体化运作，特别是为了对抗即时的网络安全威胁。此外，贾西强调，现有产品——如Alexa和亚马逊的核心零售体验——必须从头开始被彻底重新构想，而不仅仅是进行渐进式的改进。 最后，贾西讨论了亚马逊涵盖机器人技术、低地球轨道卫星（亚马逊Leo）、医疗健康和自动驾驶（Zoox）的多元化雄心。通过由重投资支持的去中心化单一职责团队进行运营，亚马逊旨在利用我们一生中最大的三次技术变革——人工智能、机器人技术和太空发展——来不断让客户的生活变得更好、更轻松。

In this October 2026 podcast episode hosted by Ben, Amazon CEO Andy Jassy discusses Amazon’s massive CapEx investments, the explosive demand for AI, and the company's long-term strategic ambitions. Reflecting on his five years as CEO since taking over in 2021, Jassy notes that Amazon’s overall top line is about 80% higher, the AWS business is almost three times bigger, and ads revenue has grown about two and a half times. Early on, facing a cost-to-serve challenge after doubling the fulfillment network during the pandemic, Amazon regionalized its U.S. network. This unexpected blessing unleashed delivery speed and retail demand. Soon after, the generative AI wave accelerated, driving AWS to a $169 billion annual revenue run rate growing at 36.7% year-over-year. A central theme of the conversation is Amazon's massive capital expenditure. Jassy explains that Amazon expects to spend $220 billion in CapEx this year, equating to $600 million a day. Despite this extraordinary investment, Amazon is still leaving meaningful demand on the table, with 2026 and 2027 largely sold out and good chunks of 2028 already reserved. Jassy addresses concerns from the financial markets about whether there is enough money to finance these infrastructure builds. He argues that the country and the world cannot afford *not* to build data centers, comparing the scale to the 1950s highway and bridge build-out, warning that falling behind other nations like China is not an option. Addressing AI demand, Jassy describes the market as "barbelled." On one end are AI labs consuming massive amounts of compute alongside runaway apps like ChatGPT and Claude; on the other end are enterprises using AI for productivity and cost avoidance. The middle of the barbell—current enterprise production workloads using inference—still holds the largest absolute segment of future growth, though widespread enterprise adoption relies heavily on driving down the cost of inference. Jassy highlights the critical role of Annapurna Labs, acquired in 2015, in developing custom silicon like Graviton CPUs and Tranium AI accelerators. These chips provide superior price-performance and help alleviate supply chain constraints. He also touches on Amazon's strategic partnerships and multi-billion-dollar investments in companies like Anthropic and OpenAI, noting that these teams are building some of the most successful and rapidly growing technologies in the space. On organizational design and the future of work, Jassy shares how AI tools (like the Bedrock team rebuilding their inference engine in 76 days) are shrinking teams and accelerating prototyping. He predicts that while there will be new jobs, organizations will utilize fewer humans for traditional tasks, and companies will need to rewrite their codebases to operate agentically, especially to combat immediate cybersecurity threats. Furthermore, Jassy emphasizes that existing products—such as Alexa and Amazon's core retail experiences—must be completely reimagined from scratch rather than just incrementally improved. Finally, Jassy discusses Amazon's diverse ambitions spanning robotics, low-earth orbit satellites (Amazon Leo), healthcare, and autonomous driving (Zoox). Operating via decentralized single-threaded teams backed by heavy investment, Amazon aims to leverage three of the biggest technological transformations of our lifetime—AI, robotics, and space development—to continually make customers' lives better and easier.