AWS’s AI business is rapidly expanding across diverse industries, driven by widespread adoption of both training and inference workloads, with a current AI revenue run rate of $25 billion and significant investments in infrastructure and proprietary chips to meet growing demand. The company emphasizes openness and choice in AI models through its Bedrock platform, supporting innovation and a competitive ecosystem while securing long-term customer commitments and exploring future opportunities in chip sales.
The AI business at AWS is experiencing rapid growth driven by a broad range of customers across various industries, not just the major AI labs like OpenAI and Anthropic. This growth spans startups and enterprises in sectors such as financial services, healthcare, retail, and media, all leveraging AI to enhance their operations. AWS sees this widespread adoption as a positive sign, reflecting the pervasive impact of AI across different business sizes and industries.
AWS’s AI revenue run rate has reached $25 billion, encompassing both training and inference workloads. Large companies and startups alike use AWS for training AI models and running inference to automate processes, improve efficiencies, and create new customer experiences. While training remains significant, the trend is shifting increasingly towards inference, as companies integrate AI models into their daily operations to generate value more cost-effectively.
Capital expenditure (CapEx) at AWS is substantial, with $220 billion allocated this year, primarily driven by AI-related investments and higher memory pricing. AWS plans to increase CapEx further next year to meet the growing demand, which currently outstrips supply. The company has secured long-term commitments from customers through 2027 and 2028, reflecting strong confidence in continued growth and the need to expand infrastructure to support AI workloads.
AWS’s chip business, which includes renting out capacity based on proprietary chips like Trainium and Graviton, is also thriving. These chips offer customers cost savings and performance benefits, with some workloads seeing 20-30% reductions in inference costs. While AWS currently focuses on renting capacity rather than selling chips outright, it remains open to exploring chip sales in the future. The combination of AWS-designed chips and Nvidia GPUs provides customers with flexible, optimized options for their AI needs.
Finally, AWS emphasizes the importance of choice and openness in AI models, supporting both frontier models and open-weight models through its Bedrock platform. Signing the open weights letter reflects AWS’s commitment to fostering innovation and ensuring a level regulatory playing field for all AI models. AWS sees a future where open-weight model providers monetize their IP through licensing, benefiting from AWS’s broad customer base and platform, which enables a competitive and vibrant AI ecosystem.
