Amazon Q2: AWS AI tops $25B run rate with $496B backlog as Bedrock catalogue expands

Amazon's Q2 results showed AWS AI revenue running above $25 billion annually, more than doubling year over year, with a $496 billion backlog and the fastest cloud growth since 2021. The print eased Wall Street concerns about Amazon's AI strategy and sent shares higher. CEO Andy Jassy made the strategic case explicit: Amazon can win the AI race through platform breadth — offering customers every major model — rather than by owning the single best frontier model.

That thesis is visible in Bedrock's expanding catalogue, which added more than ten managed foundation models including OpenAI's GPT-5.6, Anthropic's Claude Opus 5, Google DeepMind's Gemma 4, and xAI's Grok 4.3. On the infrastructure side, SageMaker HyperPod introduced Disaggregated Prefill and Decode (DPD) inference, which splits the compute-heavy prefill and memory-bound decode phases of LLM serving across dedicated GPU pools to improve throughput and efficiency.

The breadth strategy is a notable bet given that Amazon is simultaneously scaling back its own Nova model line and reportedly staffing a new frontier-research team — an implicit acknowledgment that its in-house models haven't led. It also contrasts with Microsoft's similar 11,000-model catalogue pitch, making 'be the neutral platform' the shared playbook of the two cloud giants.

SageMaker practitioners on LinkedIn praised the DPD inference addition and the $25B ARR figure as a genuine production-ops win rather than marketing. Skeptics will watch margins: hosting rivals' frontier models is lower-margin than owning them, and the $496B backlog's conversion timeline and pricing pressure from cheap Chinese models are the numbers to track next quarter.