Amazon is attempting to own the AI supply chain the way the great industrial combinations once owned ore, furnaces, and rail. Its position rests on two linked assets: it sells the cloud infrastructure in which models run, and it holds an investment in Anthropic, which is also an AWS customer 62,72. That combination gives Amazon exposure to rising compute demand through AWS revenue and the value of its holdings. But the combination also makes the investment case harder to read: approximately $200 billion in annual AI-infrastructure spending consumes cash, while changes in Anthropic’s valuation can move reported earnings without producing cash 5,47,72,74.
AWS and the Cost of Expansion
AWS has genuine scale, though the supplied measures are not directly comparable: one account puts recent annual revenue above $90 billion 58, another describes a $128.7 billion scale 58, and a later assessment cites $169 billion annualized 80. The direction of travel matters more than any single figure. A reported 37% revenue increase and a $4.6 billion sequential gain point to momentum 72, while AWS AI and chip revenue were each reported above $25 billion with triple-digit growth 72. A longer-term projection of $250–300 billion in annual AWS revenue by 2030, with market leadership retained, has two-source support but remains a forecast 58. Google Cloud is narrowing the share gap, and Microsoft and Google are described as growing faster in cloud infrastructure 59,63.
Capturing that demand requires unusually large commitments. Amazon’s 2026 capital expenditure was reported at $220 billion 61. A more recent assessment says annual AI data-center spending above $200 billion strains free cash flow 74. Amazon is not relying on general-purpose compute alone: it is described as building a $25 billion chip business 65, alongside its Bedrock and Trainium infrastructure strategy 71. There is also evidence of outside capital being brought into the buildout, through reports of a vehicle to finance roughly $8 billion of Nvidia chips 78. The status of that financing is unresolved: other accounts describe the same $8 billion transaction as a completed sale or an established vehicle 64,68,77.
Anthropic as Anchor Customer and Distribution Rail
Anthropic links demand, distribution and valuation for Amazon. The commercial tie is material: Anthropic has an agreement for up to 5 GW of AWS capacity and more than $100 billion of spending over ten years, according to a single report 72. Claude reaches customers through Amazon Bedrock, including an India offering that keeps eligible inference requests in AWS’s Mumbai and Hyderabad regions 73. Claude Haiku 5.5’s availability through Bedrock, as well as Anthropic’s own API and Google Cloud Vertex AI, illustrates both AWS’s distribution opportunity and the fact that Amazon does not have exclusive access to the model 79.
The revenue base that travels through those channels is fast-growing but operationally narrow. Anthropic’s reported 2025 revenue rose roughly twelvefold to about $4.6 billion, a figure supported by 13 sources 49,50,52,54,55,56,60. Four sources report a $42 billion net loss 45,68,77, and a separate prospectus account distinguishes an $8.06 billion operating loss from that net loss 82. Revenue is heavily usage-dependent: approximately $3.8 billion came from Claude usage versus $789 million from subscriptions 60. Amazon and Google marketplaces supplied 47% of 2025 revenue 60, while five sources place roughly a quarter of revenue with two customers 50,51,53,56,82. Those channels aid distribution, but usage-sensitive sales and customer concentration make demand less predictable against $54.6 billion of reported non-cancelable compute and hosting commitments at the end of 2025 60. An estimate of approximately $518 billion in longer-term compute spending is not a locked-in obligation; its source expressly qualifies it on that point 75.
Valuation Marks and the Earnings Mirage
Amazon’s investment adds a separate valuation channel that can dominate reported results. June filing figures cited in the material put Anthropic preferred stock at $92.5 billion and convertible notes at $97.9 billion—approximately $190 billion together—plus $28.7 billion in OpenAI investment 59. A separate claim totals those three holdings at approximately $240.4 billion 59, an aggregate that does not reconcile with the stated components and should not be used without clarification. Amazon’s reported Q2 2026 net income of $62.6 billion included $53.4 billion of pretax other income primarily linked to Anthropic 59, which helps explain why headline earnings and a reported 20.2-times P/E can overstate the contribution of recurring operations 59.
A potential Anthropic IPO would bring that distinction into focus. Reports of an IPO filing are widely corroborated 1,2,3,6,7,8,9,10,11,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,34,35,38,40,41,42,44,46,57,69, and a $2 trillion proposed valuation appears across seven sources 37,40,43,70. Timing and valuation remain unsettled: some material suggested an offering as early as October 4,12,36,66, while a report said marketing could begin the week of November 9 64; another claim posits a $1 trillion target 33,39,69,82. The thesis that an IPO would revalue Amazon’s stake and its AWS, Trainium and Bedrock relationships is an interpretation, not an established market outcome 75. Regulatory exposure adds further uncertainty: the FTC is investigating Anthropic 48,67,70, and the Pentagon’s supply-chain-risk designation could impede defense adoption 69,76,81.
The Strategic Test Ahead
The strategic reading is therefore threefold. AWS demand has evidence behind it, but returns remain the test: reported cloud momentum must ultimately justify Amazon’s roughly $200 billion annual AI-infrastructure spending. Anthropic creates commercial upside and concentration risk: its AWS commitments and Bedrock distribution sit alongside usage-sensitive revenue, concentrated customers and fixed compute commitments. Finally, Amazon’s operations must be valued separately from investment marks, because Anthropic-linked gains have materially affected reported income while the proposed IPO’s timing, valuation and effect on Amazon’s shares remain uncertain.
The long-term question is an old one in new form: whether Amazon can convert enormous fixed investment into durable, recurring industrial returns before the competition and the accounting catch up with it.
The cloud and GPU infrastructure sector has moved from a demand story into a capacity, capital, and governance test. AWS remains the foundational backbone of enterprise IT and generative AI
Amazon is best understood as a multi-layer flywheel rather than as a retail company with ancillary businesses. Its online retail and marketplace franchise remains dominant 1,32,35, but
The problem of inquiry is the divergence between Amazon's accelerating high-margin profit engines and its capital-intensive infrastructure cycle. The supplied material, current as of early-to-