WHY MARKET IS COMPLETELY MISREADING $AMZN EARNINGS
I think Amazon's selloff is confusing the cost of building AWS with the economics of the business being built. AWS revenue grew 39% to $46B, but the more revealing number is that operating profit increased $6.3B on roughly $13B of additional revenue, implying nearly 49% incremental operating margins.
That doesn't establish returns on new AI capacity because depreciation is still catching up and management acknowledges AI margins remain below mature cloud margins. However, it challenges the idea that Amazon is adding revenue without adding earnings. The $578B backlog and Jassy's expectation of growth above 40% in both 2027 and 2028 provide remarkable visibility, although neither represents guaranteed near-term profit.
What makes this opportunity more interesting is that inference potentially improves AWS economics beyond simply renting more GPUs. Bedrock token volume increased 17x while inference already represents more than half of AWS AI revenue, and management says custom chips saved customers over $2B in the quarter.
The question is how much of those savings Amazon retains rather than passes through in lower prices. Older Trainium capacity renting at a premium proves scarcity today, not a longer useful life.
More importantly selling Trainium externally beginning in 2027 creates immediate hardware revenue, whereas deploying those same chips inside AWS could generate years of recurring revenue and larger cumulative cash flows.
An AI agent also consumes CPUs for tool calls alongside storage and networking, allowing AWS to capture more spending per workload than the accelerator bill alone which makes chip allocation and utilization just as important as chip performance.
I also understand the selloff because trailing operating cash flow reached $168B while free cash flow fell to negative $42B, with $220B in cash CapEx planned this year and more expected in 2027. As an illustrative hurdle, every additional $100B of productive capital needs approximately $12B in sustainable annual operating profit to earn a 12% pretax return, and AWS doesn't disclose returns by infrastructure cohort.
Yet the softer retail headlines don't establish deterioration either. Moving Prime Day into Q2 reduced reported Q3 growth, while North American operating income rose from $4.8B to $8.7B and International profit improved. Advertising's 12% growth also faced an unusual comparison, although agentic commerce could eventually weaken Amazon's direct customer relationship.
Over the next few reports I'll watch AWS incremental profit as new depreciation arrives, Trainium's contribution to inference economics and whether operating cash flow begins catching up with CapEx. If those converge, the market may be penalizing Amazon for building an asset whose long-term earning power it's underestimating. If they don't, the discount is deserved.