🧔♂️ A friendly human may check it before it goes live. More news here
Amazon Q3 sales rise 13% to $180.2b, AWS revenue grows 20%
Amazon reported Q3 2025 net sales of US$180.2 billion, up 13% from a year earlier.
North America sales rose 11% to US$106.3 billion, while international sales rose 14% to US$40.9 billion.
Amazon Web Services (AWS) revenue increased 20% year-on-year to US$33 billion.
Operating income was US$17.4 billion, flat compared to Q3 2024, and included a US$2.5 billion legal settlement and US$1.8 billion in severance costs.
Net income rose to US$21.2 billion, up from US$15.3 billion last year, with gains from investments in Anthropic.
Operating cash flow for the trailing twelve months increased 16% to US$130.7 billion, while free cash flow dropped to US$14.8 billion due to increased capital expenditures.
Amazon also expanded AWS capacity, added new AI chips and services, and signed new cloud agreements with companies including Delta Air Lines and Volkswagen Group.
The company reported growth in its same-day delivery network, and continued investments in AI-powered retail and cloud products.
🔗 Source: Amazon
🧠 Food for thought
Implications, context, and why it matters.
Look at AWS margins to judge GenAI’s real contribution
- AWS posted $11.4 billion in Q3 operating income, up ~10% on 20% revenue growth, for a 34.6% margin 1. A falling margin amid rising sales likely comes from heavy infrastructure spend, and CapEx reached $34.2 billion in the quarter with 2025 outlays at $125 billion 2.
- Backlog reached $200 billion at Q3 end, and October deals topped the quarter’s total 2. It means signed long-term commitments not yet booked as revenue. Without a split between generative AI and core services such as storage or databases, AI’s share stays unclear.
- Amazon plans to double AWS capacity by 2027 2. It covers compute, storage plus network resources.
Trainium2 pricing opens room for AI cost tools
- Trainium2 is Amazon’s second-gen AI training chip on AWS that rents for about $4.80 per hour 3, while Nvidia H100 instances average $5.76 to $6.80 in the U.S. 3. That gap and 150% quarter over quarter growth into a multibillion dollar line 2 have spurred demand for cheaper AI compute.
- Vendors can ship migration and benchmarking tools to help enterprises compare Trainium2 with Nvidia instances 4. Clear data on arithmetic intensity, or compute per unit of memory access, can pinpoint where Trainium2 fits Mixture of Experts models better than dense transformers 4. Anthropic’s Project Rainier is a large training cluster scaling toward more than 1 million Trainium2 chips by year end, which raises demand for workload profiling and multi cloud cost management 5.
Recent Amazon developments
Stay updated on the go with our mobile app.
Get latest insights with smoother, more personalized experience through TIA mobile app.




