Amazon's AI Business Surpasses $25 Billion Run Rate Amidst Growth Challenges

Amazon has reported that its AI business has crossed a $25 billion annualized revenue run rate, reaching this milestone in just three years. During the Q2 2026 earnings call, CEO Andy Jassy noted that this unit is growing “several hundred times faster” than AWS did in its early days.

Key Financial Metrics

In the earnings report, Amazon disclosed that both its AWS AI business and custom chip division surpassed the $25 billion run-rate mark, with both segments exhibiting triple-digit year-over-year growth. AWS revenue reached $42.2 billion, marking a 37% increase from the previous year, and achieving the fastest growth rate in the last 18 quarters.

Despite this growth, Amazon is facing challenges with a backlog of $496 billion and rising memory costs, which have contributed to a trailing twelve-month free cash flow outflow of $7.6 billion. The company has raised its 2026 capital expenditure guidance to approximately $220 billion, primarily for AI infrastructure and custom chips.

AI Cost Challenges and Market Potential

Andy Jassy also addressed the disparity in AI demand, describing it as a “barbell” with AI labs and enterprise cost-cutting at opposite ends. He emphasized the untapped market potential that remains due to high inference costs, which pose challenges for broader enterprise adoption of AI technologies.

Amazon's proprietary AI chip, Trainium, is a crucial component of its strategy, with significant agreements established with companies like Anthropic and OpenAI. The company’s managed AI service, Bedrock, has also seen increased usage, surpassing all previous quarters combined.

Looking Ahead

As Amazon navigates these complexities, key metrics to monitor include AWS growth rates, backlog conversion, operating margins, and the recovery of free cash flow. Additionally, developments regarding strategies to lower AI inference costs and any announcements from partners like Anthropic will be critical in shaping market perceptions and valuations.

New Developments in Amazon's AI Business

• Amazon has launched TorchNeuron, a native PyTorch backend for its Trainium2 and Trainium3 accelerators, aimed at simplifying research training.
• TorchNeuron entered public beta on October 19, 2026, and was open-sourced on October 20 during the PyTorch Conference North America.
• The new feature allows developers to move models onto Trainium with a single line of code, maintaining compatibility with standard PyTorch APIs.
• Supporting upgrades in Neuron 2.33 include Neuron Observer for hardware health detection, improved NKI functionality, and updated vLLM plugins.
• TorchNeuron currently supports only Trainium2 and Trainium3, excluding Trainium1 and the Inferentia chip family.
• More than 30 early customers, including companies in robotics, voice AI, and large language model development, are already using TorchNeuron.
• Notable users like Cartesia and Deepgram have reported significant performance improvements with the new configurations.
• The update is expected to broaden the research use cases for Trainium, including applications in drug discovery.