Amazon · one lens
Robert Shiller on Amazon
Amazon has a testable AI infrastructure business, but its deals and coverage do not establish the price-to-story-to-buyer loop in Shiller’s framework.
Shiller’s feedback-loop test asks whether a rising price recruited a new-era explanation, then buyers, then another rise. Amazon supplies a plausible explanation, not proof of that sequence. Per the TEXXR coverage record, Amazon raised its 2026 capital-spending guide to $220 billion from $200 billion after AWS reported quarterly revenue up 37%, to $42.2 billion from roughly $30.8 billion a year earlier; the prior-year figure is derived from the reported growth rate. AMZN rose 15.32% on July 31, to $271.58 from $235.50. The supplied market data put its September 29 close at $246.67, below that July close. Neither price establishes why buyers acted.
Amazon’s OpenAI agreement shows why Shiller separates a precipitating event from narrative amplification. Amazon committed $15 billion upfront against another $35 billion contingent on an IPO or AGI milestone, with roughly 2 gigawatts of Trainium capacity attached. Amazon then agreed to invest up to $25 billion more in Anthropic, above $8 billion already committed; Anthropic agreed to use up to 5 gigawatts of Trainium, versus OpenAI’s roughly 2 gigawatts. Amazon’s July AGI-lab closure and reported Nova deprecation complicate a story that treats its model development and cloud partnerships as the same bet. Shiller’s new-era thinking test asks whether coverage turns these distinct commitments into a blanket justification for price.
Anthropic’s IPO filing, as reported by Reuters, offers a narrower check. Anthropic routed 47% of its 2025 sales, about $2.16 billion, through Amazon and Google combined. Reuters estimated that Anthropic paid about $351 million in distribution fees against those $2.16 billion in partner-routed sales. Reuters did not separate Amazon’s portion of either figure. Amazon’s investment commitment therefore cannot be read as a disclosed return from that channel.
TEXXR counted 151 Amazon articles in July–September 2026, versus an average of 101.9 articles per quarter over the preceding eight quarters. TEXXR measures coverage output here, not buyer recruitment or AI use. On September 30, Digiday covered Amazon’s joined-up advertising workflows while Bloomberg covered a new Fire TV device. Those reports also test Shiller’s media mechanism: the record shows which Amazon stories received attention, but not whether any story recruited the next stock buyer.
An editorial application of Robert Shiller's published framework — not their actual view, affiliation, or advice.
Where this lens comes from
Robert Shiller's framework is set out at the lens page, drawn from Irrational Exuberance. This page applies it to Amazon and nothing else — the company's full record is in the dossier.