Amazon CEO Andy Jassy said on the company’s 29 October earnings call that inference “already counts for well over half of AWS’s AI revenue and growing”, according to a machine transcript of the call published by Gloom.
He also said AWS is “installing an AI accelerator about every 14 seconds”. By our arithmetic, that pace works out to about 6,170 accelerators a day.
A vertical news Short narrated from this piece’s headline and opening line.
Business Pill · Model Routing
An evergreen Business Engineer explainer on model routing and inference cost. It explains the general idea only and does not discuss Amazon.
The key insight: Per a machine transcript of the 29 October call, Jassy said inference already counts for well over half of AWS’s AI revenue, and Amazon’s release describes Bedrock, AWS’s inference service, as growing token volume nearly 17x year-over-year.
Per the Gloom transcript, Jassy said enterprises are “still very early in using inference of scale”, and that inference “already counts for well over half of AWS’s AI revenue and growing and we have a lot of training revenue as well”. The transcript is machine-made and unpunctuated; Amazon had not published its own transcript when we wrote this.
He described lowering the cost of inference as a major goal for AWS and one of the main reasons Amazon builds its own chips, according to the same transcript. He did not give a dollar figure for AWS’s AI revenue on the parts of the call we read.
Year-over-year growth multiples disclosed this week: Amazon Bedrock token volume nearly 17x and Fortune 500 Bedrock spending nearly 7x (Amazon Q3 2026 release); Google tokens processed per minute more than 4x (Pichai, Alphabet Q3 call). Each company measures something different. Sources: Amazon, Alphabet.
An Accelerator Every 14 Seconds
Per the transcript, Jassy said the backlog is “five hundred seventy eight billion dollars growing triple digits year over year and well diversified” and that “we’re installing an AI accelerator about every 14 seconds”. GeekWire also reported the $578 billion backlog from the call.
By our arithmetic, one accelerator every 14 seconds is about 6,170 a day, or roughly 568,000 over a 92-day quarter if that pace held throughout. Jassy did not say how long the pace has held or which chips it counts.
What the Release Says
Amazon’s third-quarter release says “AWS segment sales increased 39% year-over-year to $46.0 billion” and that AWS operating income was $17.8 billion, against $11.4 billion a year earlier. It describes Bedrock as “AWSโs inference service”, with “Fortune 500 companies spending nearly 7x more year-over-year and total Bedrock token volume growing nearly 17x year-over-year”. Our Bedrock piece has the detail.
What Google Reported
On Alphabet’s 28 October call, Sundar Pichai said Google was processing “nearly 32 billion tokens per minute, 40% more than last quarter and more than four times what we processed a year ago”. The two companies measure different things, so the figures do not compare directly. Our agent-platform piece sets the metrics side by side, and our Microsoft piece covers its token pricing.
As we read the transcript, Jassy tied two things together: inference as the larger share of AWS’s AI revenue, and lowering the cost of inference as a main reason Amazon builds its own chips.
As we read the sources, the demand figures this week came in different measures: Bedrock token volume in Amazon’s release and tokens per minute on Alphabet’s call, so they do not compare directly. Amazon’s release gives no dollar split of AWS revenue between inference and training.
By our arithmetic, an accelerator every 14 seconds is about 6,170 a day; Jassy gave no chip mix, and the figure is a pace, not a quarterly total Amazon reported.
Andy Jassy, Amazon Q3 call, 29 October 2026, per a Gloom machine transcript
“already counts for well over half of AWS’s AI revenue and growing”
Three Implications
FOR AWS CUSTOMERS Jassy said lowering the cost of inference is a major goal for AWS, according to the transcript; he gave no price figures.
FOR CHIP SUPPLIERS Jassy said AWS is installing an AI accelerator about every 14 seconds, per the transcript, without saying how many are Trainium and how many are from other suppliers.
FOR INVESTORS Amazon’s release gives no dollar split of AWS revenue between inference and training, and the inference line comes from a machine transcript, not an Amazon document.
The analysis puts it this way: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, Jassy’s remark that inference is over half of AWS’s AI revenue is the kind of shift the analysis describes. The analysis was published in March 2026 and does not discuss Amazon’s third quarter.
What Is Not Established
The call lines come from a machine transcript published by Gloom, not an Amazon document, and Amazon’s release gives no dollar split of AWS revenue between inference and training. The 6,170-a-day and 568,000-a-quarter figures are our arithmetic from Jassy’s 14-second remark.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Jassy said on the call that inference is already well over half of AWS’s AI revenue and that AWS installs an AI accelerator about every 14 seconds, according to a machine transcript. Amazon’s release put AWS sales at $46.0 billion, up 39%.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Jassy’s call remarks come from a machine transcript of the 29 October call published by Gloom; Amazon had not published its own transcript. AWS and Bedrock figures come from Amazon’s Q3 2026 results release (8-K Exhibit 99.1); the Google figure is from Alphabet’s 28 October call. The 6,170-a-day and 568,000-a-quarter figures are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
