Andy Jassy 表示,AWS 有“潜力成为年收入达 1 万亿美元的业务”。1
亚马逊曾以为天花板是几千亿美元,如今它认为这项业务“至少会翻倍”。AWS 本季度按年化计算,营收运行率已达 1690 亿美元。
亚马逊的迅猛增长点燃了资本主义的梦想。AWS 增长了 36.7%,这是 18 个季度以来最快的增速,也是连续第五个季度加速增长。2 五个季度前,其增速为 17%;如今已达 37%。
在这张图表上,AWS 的曲线斜率位居第二陡峭,六个季度内加速了 20 个百分点,而 Google Cloud 则加速了 54 个百分点。相比之下,Azure 几乎持平,本季度之前连续三个季度恰好保持在 40%,仅增加了 10 个百分点。AWS 与 Azure 之间 16 个百分点的差距已缩小至 6 个百分点。
与微软对比是公平的,因为两者战略相似:都不拥有前沿模型,并且都在销售自研芯片,与 Nvidia 的利润率抗衡。
AWS 正以 37% 的复合增长率,从市场上最大的纯云业务基数出发,并且实现了盈利。营业利润增长 64%,达到 166 亿美元,利润率达到 39.4%,同比上升 650 个基点。21
前瞻性订单簿则讲述了一个不那么光鲜的故事。AWS 期末积压订单达 4960 亿美元,较三个月前的 3640 亿美元实现了三位数增长,但尽管 AWS 是三者中规模最大的业务,其积压订单却是三家中最小的。31
在这方面,亚马逊落后了,而前沿实验室正是原因所在。微软的 6780 亿美元积压订单规模最大,部分原因在于其中很大一部分来自 OpenAI,微软很早就锁定了 OpenAI 对 Azure 的承诺。4 亚马逊在这些交易上入场较晚:Anthropic 和 OpenAI 后来做出了多吉瓦级的 Trainium 承诺,但这些合同是在微软已经记入自己的订单之后才达成的。1
与 Google 一样,亚马逊正将每一美元的自由现金流都投入资本支出,并在此基础上举债,这与微软更为谨慎的路线形成分歧。过去十二个月的自由现金流从一年前的正 182 亿美元降至负 76 亿美元。5 Alphabet 在同一季度也转为负值,这是其自 2004 年上市以来的首次;微软则保持正值,为 196 亿美元。64 在被问及资金从何而来时,Olsavsky 指出,亚马逊今年一直在发行债券。1
亚马逊目前在基础设施上的支出,已超过整个公司所产生的现金总额。
微软面临同样的需求,支出了358亿美元的现金资本开支,若计入融资租赁则接近410亿美元,而艾米·胡德将这种短暂的份额表述为一种她可以调节的期权。亚马逊支出了531亿美元,并将其2026年计划从2000亿美元上调至2200亿美元,理由是内存价格上涨。微软做了对冲。亚马逊没有。
贾西对此的辩护异常具体。数据中心在产生任何收益之前,要先吸收资本长达两年,之后使用寿命可达30年或更久。服务器在部署前数月到位,不到三年即可回本,其使用寿命为五到六年,对应的是至少五年的AI产能合同。
因此,这万亿美元押注归结为一个问题,而这不是融资问题。贾西将AI采用描述为杠铃形,一端是消耗“大量、大量算力”的实验室,另一端是企业收获成本节省。“杠铃中间,”他说,“是所有当前的企业生产工作负载,”其中大多数尚未广泛使用推理能力。中间部分“将是最大的绝对细分市场”。
这一部分能否快速到来,就是这场赌注,而贾西主动承认了这一点:“我不知道杠铃中间部分的增长轨迹,是否会像我们目前看到的杠铃形AI实验室部分那样陡峭得惊人。”
如果他说对了,AWS将凭一己之力突破万亿美元营收大关。亚马逊的估值约为销售额的三倍,因此一家由AWS拉动前行的公司,其市值将接近10万亿美元。
-
亚马逊2026年第二季度财报电话会议记录,2026年7月30日;可通过亚马逊投资者关系网站获取回放。贾西谈及AWS有望成为“万亿美元营收业务”的潜力,以及从“数千亿美元”升级至“至少翻倍”的预期;4960亿美元的积压订单实现三位数增长;营收增长36.7%,连续第五个季度加速;单季新增营收超过46亿美元;1690亿美元的年化营收规模若作为独立公司可位列财富500强第24位;2026年预期现金资本支出从2000亿美元上调至2200亿美元,原因是内存成本上升;数据中心交付周期为两年,而使用寿命超过30年;服务器和网络设备盈亏平衡期不到三年,而使用寿命为五至六年;AI容量合同期限至少五年;“如果需求不存在,我们就不会投入资本”;2026年和2027年容量不足;2028年需求惊人;Anthropic和OpenAI对Trainium芯片的数十吉瓦级承诺;AI采用呈现杠铃形态,实验室消耗“海量计算资源”,中间的企业生产负载是“最大的绝对细分市场”,但他也警告称“我不确定杠铃中间部分的发展轨迹是否会像当前杠铃形态的AI实验室部分那样呈现同样陡峭的曲线”。奥尔斯夫斯基在回答科林·塞巴斯蒂安关于建设资金来源的提问时表示,亚马逊今年已发行债券,并保留了“大量选择”。奥尔斯夫斯基还提到,AWS利润率同比上升650个基点,剔除能源衍生品公允价值收益后为520个基点,第二季度现金资本支出为531亿美元。↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
亚马逊2026年第二季度财报:AWS净销售额422.32亿美元,同比增长37%;AWS营业利润166.21亿美元;公司整体营业利润274.61亿美元;滚动经营现金流1614.03亿美元;滚动物业及设备购置(扣除销售及激励收入)1690.07亿美元;滚动自由现金流为负76.04亿美元;第二季度物业及设备购置542.08亿美元,扣除11.32亿美元相关收入。AI与芯片业务年化营收均超过250亿美元。上年同期滚动数据为1029亿美元,作为可比期间。↩︎ ↩︎ ↩︎
-
亚马逊 2026 年第一季度 10-Q 财报:截至 2026 年 3 月 31 日,长期客户合同承诺(主要为 AWS)约为 3640 亿美元,加权平均剩余期限为 5.5 年。 ↩︎
-
微软 2026 财年第四季度财报:微软云收入 593 亿美元,同比增长 27%;Azure 及其他云服务收入同比增长 43%;智能云收入 393 亿美元;商业剩余履约义务(RPO)达 6780 亿美元;Azure 年收入突破 1000 亿美元。微软第四季度现金资本支出为 358 亿美元,含融资租赁在内约为 410 亿美元,数据来自公司 2026 财年第四季度材料和电话会议评论;同期经营现金流为 554 亿美元,自由现金流为 196 亿美元。Amy Hood 披露,剔除 OpenAI 后 RPO 增长 25%,而包含 OpenAI 时增长 84%;OpenAI 的隐含占比由这两个比率推算得出,并非单独披露。需要注意的是,亚马逊的 531 亿美元为固定资产净购置额,而微软的数字为现金资本支出,因此两者对比仅具参考性,并非严格同口径。 ↩︎ ↩︎ ↩︎
-
亚马逊 2026 年第二季度财报电话会议幻灯片:自由现金流从 2025 年第二季度的 182 亿美元降至 2026 年第二季度的负 76 亿美元;经营现金流同比增长 33%。 ↩︎
-
Alphabet 2026 年第二季度业绩:谷歌云收入 248 亿美元,同比增长 82%,营业利润 88 亿美元。Alphabet 第二季度资本支出为 449 亿美元,云积压订单为 5140 亿美元,数据来自公司第二季度报告及管理层评论。Alphabet 自由现金流转负:2026 年第二季度为负 59 亿美元,为 2004 年 IPO 以来首次出现负值季度。 ↩︎
-
这是作者自己的推算,并非公司指引。将 2026 财年第二季度的分部收入年化后,AWS 约为 1690 亿美元,核心业务约为 6340 亿美元,其中北美季度收入为 1162 亿美元,国际季度收入为 422 亿美元。核心业务当季综合增长 15.7%,其中北美增长 16%,国际增长 15%。若 AWS 保持 20% 的持续增长,其年收入将在 9.8 年后达到 1 万亿美元;核心业务同期按 15.7% 复利增长,将达到约 2.6 万亿美元,总收入接近 3.6 万亿美元,按三倍市销率计算市值约为 10.9 万亿美元。AWS 增长更快反而会拉低总估值,因为核心业务在更大基数上复利增长的年限更短:若 AWS 保持 37% 的持续增长,将在 5.7 年后达到 1 万亿美元,总收入约为 2.5 万亿美元,按相同倍数计算市值约为 7.4 万亿美元。核心业务在十年内保持 15.7% 的增长是较为激进的假设;若每年递减一个百分点至 6% 的底线,则在上述 AWS 增长路径下,总估值区间为 7.0 万亿至 8.5 万亿美元。Jassy 并未为万亿目标设定时间表,因此时间跨度是作者自己的假设。三倍市销率是对亚马逊当前交易水平的描述性指标,而非估值方法,而且万亿美元规模的 AWS 将带来比当前以零售为主的业务更高利润率的收入结构。 ↩︎
Andy Jassy said AWS has “the potential to be a $1 trillion revenue business.”1
Where Amazon once thought the ceiling was a few hundred billion, it now believes the business “will be at least double that.” AWS finished the quarter at a $169b annualized run rate.
Amazon’s torrid growth fuels the capitalist dream. AWS grew 36.7%, the fastest in 18 quarters & the fifth consecutive quarter of acceleration.2 Five quarters ago it grew 17%; today, 37%.
AWS traces the second steepest trajectory on that chart, 20 points of acceleration over six quarters against Google Cloud’s 54. Azure is close to flat by comparison, adding ten points & sitting at exactly 40% for three consecutive quarters before this one. A 16-point deficit to Azure has narrowed to six.
Microsoft is the fair comparison, because the two pursue similar strategies: neither owns the frontier model, & both sell custom silicon against Nvidia’s margin.
AWS is compounding at 37% from the largest pure-cloud base in the market, & doing it profitably. Operating income grew 64% to $16.6b & margin reached 39.4%, up 650 basis points year over year.21
The forward book tells a less flattering story. AWS closed with $496b of backlog, growing triple digits against $364b three months earlier, & it is the smallest of the triumvirate even though AWS is the biggest business.31
Here Amazon is behind, & the frontier labs are the reason. Microsoft’s $678b is the largest partly because a substantial share traces to OpenAI, whose Azure commitment it locked up early.4 Amazon came later to those deals: Anthropic & OpenAI have since made multi-gigawatt Trainium commitments, but the contracts arrived after Microsoft had already booked its own.1
Like Google, Amazon is funneling every dollar of free cash flow into capital expenditure & borrowing on top of it, diverging from the more cautious Microsoft route. Trailing free cash flow fell from positive $18.2b a year ago to negative $7.6b.5 Alphabet’s turned negative in the same quarter, its first since the 2004 listing; Microsoft’s stayed positive at $19.6b.64 Asked where the capital comes from, Olsavsky noted Amazon has been issuing debt this year.1
Amazon now spends more on infrastructure than the whole company generates in cash.2
Microsoft, facing the same demand, spent $35.8b of cash capex, nearer $41b once finance leases count, & Amy Hood framed the short-lived share as optionality she can throttle. Amazon spent $53.1b & raised its 2026 plan from $200b to $220b, blaming memory prices.41 Microsoft hedges. Amazon does not.
Jassy’s defense of that was unusually specific. Data centers absorb capital two years before earning anything, then last 30 years or more. Servers arrive months before deployment, break even in a little under three years, & carry five-to-six-year lives against AI capacity contracted for at least five years.1
So the trillion rests on one question, & it is not a financing question. Jassy describes AI adoption as barbelled, with the labs consuming “gobs and gobs of compute” at one end & enterprises harvesting cost savings at the other. “In the middle of the barbell,” he said, “is all of the current enterprise production workloads,” most of which do not yet use inference pervasively. That middle “will be the largest absolute segment.”1
Whether it arrives quickly is the wager, & Jassy conceded as much unprompted: “I don’t know if the trajectory of that middle part of the barbell will be the same wildly steep trajectory that we’ve seen with the current barbelled AI Labs piece.”1
If he is right, AWS pushes past the trillion-dollar revenue mark on its own. Amazon trades at about three times sales, so a company pulling AWS along behind it would approach a $10t market capitalization.7
-
Amazon Q2 2026 earnings call transcript, July 30, 2026; replay available via Amazon Investor Relations. Jassy on the potential for AWS to become “a $1 trillion revenue business” and the upgrade from “a few hundred billion” to “at least double that”; $496b of backlog growing triple digits; revenue growth of 36.7% accelerating for the fifth straight quarter; over $4.6b of sequential revenue added; the $169b run rate ranking 24th on the Fortune 500 as a standalone company; $220b of expected 2026 cash capex raised from $200b on higher memory costs; a two-year data center lead time against 30-plus-year useful lives; less-than-three-year server and networking breakevens against five-to-six-year useful lives; at-least-five-year AI capacity contracts; “if the demand isn’t there, we won’t spend the capital”; insufficient 2026 and 2027 capacity; striking 2028 demand; multi-gigawatt Trainium commitments from Anthropic and OpenAI; the barbell shape of AI adoption with the labs consuming “gobs and gobs of compute,” enterprise production workloads in the middle as “the largest absolute segment,” and his caveat that “I don’t know if the trajectory of that middle part of the barbell will be the same wildly steep trajectory that we’ve seen with the current barbelled AI Labs piece”. Olsavsky, asked by Colin Sebastian about sources of capital for the buildout, on Amazon having issued debt this year and retaining “a lot of options.” Olsavsky on AWS margin up 650 basis points year-over-year, 520 excluding the energy-derivative fair-value gain, and Q2 cash capex of $53.1b. ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
Amazon Q2 2026 earnings release: AWS net sales of $42.232b, up 37%; AWS operating income of $16.621b; company operating income of $27.461b; trailing operating cash flow of $161.403b; trailing purchases of property & equipment net of proceeds from sales & incentives of $169.007b; trailing free cash flow of negative $7.604b; Q2 purchases of property & equipment of $54.208b less $1.132b of proceeds; AI & chips businesses each above $25b run rates. The $102.9b prior-year trailing figure is the comparable period. ↩︎ ↩︎ ↩︎
-
Amazon Q1 2026 Form 10-Q: long-term customer-contract commitments, primarily AWS, of approximately $364b as of March 31, 2026, with weighted-average remaining life of 5.5 years. ↩︎
-
Microsoft FY26 Q4 earnings release: Microsoft Cloud revenue of $59.3b, up 27%; Azure & other cloud services revenue up 43%; Intelligent Cloud revenue of $39.3b; commercial RPO of $678b; Azure surpassing $100b of annual revenue. Microsoft Q4 cash capex of $35.8b and approximately $41b including finance leases are from the company’s FY26 Q4 materials and call commentary, alongside operating cash flow of $55.4b and free cash flow of $19.6b. Amy Hood disclosed RPO growth of 25% excluding OpenAI against 84% including it; the implied OpenAI share is derived from those two rates rather than separately disclosed. Note that Amazon’s $53.1b is net property & equipment purchases while Microsoft’s figure is cash capex, so the comparison is indicative rather than like-for-like. ↩︎ ↩︎ ↩︎
-
Amazon Q2 2026 conference call slides: free cash flow bridge from $18.2b in Q2 2025 to negative $7.6b in Q2 2026; operating cash flow up 33% year-over-year. ↩︎
-
Alphabet Q2 2026 results: Google Cloud revenue of $24.8b, up 82%, with operating income of $8.8b. Alphabet Q2 capex of $44.9b & $514b cloud backlog are from the company’s Q2 reporting and management commentary. Alphabet’s free cash flow turns negative: negative $5.9b in Q2 2026, its first negative quarter since the 2004 IPO. ↩︎
-
Author’s arithmetic, not company guidance. Annualizing Q2 2026 segment revenue gives AWS $169b & a core business of $634b, from North America of $116.2b & International of $42.2b per quarter. Core grew 15.7% blended in the quarter, weighting North America’s 16% & International’s 15%. At 20% sustained growth AWS reaches $1t of annual revenue in 9.8 years; core compounding at 15.7% over the same span reaches roughly $2.6t, for total revenue near $3.6t & a $10.9t capitalization at three times sales. Faster AWS growth produces a lower figure because core has fewer years to compound off the larger base: 37% sustained reaches $1t in 5.7 years, for roughly $2.5t of total revenue & $7.4t at the same multiple. Holding core at 15.7% for a decade is the aggressive assumption; fading it by a point a year to a 6% floor yields a $7.0t to $8.5t range across the same AWS paths. Jassy attached no timeline to the trillion, so the horizon is the author’s. Three times sales is descriptive of where Amazon trades rather than a valuation method, & a trillion-dollar AWS would carry a higher-margin revenue mix than today’s retail-weighted business. ↩︎