Amazon's AI capacity crunch and performance issues pushed customers to rivals including Google

This summer, Amazon's cloud division struggled to meet the explosive demand for artificial intelligence services, leading to significant revenue losses. Central to this challenge was the performance of Amazon Web Services' (AWS) Bedrock service, which allows developers to access advanced AI models like Anthropic's Claude and Meta's Llama. Internal documents revealed that Bedrock encountered "critical capacity constraints" during the busy summer months, prompting some users to explore alternatives from competitors such as Google Cloud. The resulting service shortages are estimated to have cost AWS tens of millions in potential revenue. For instance, Epic Games redirected a $10 million project for Fortnite to Google Cloud, citing insufficient availability from AWS. Additionally, oil trader Vitol considered shifting projects away from AWS, risking a $3.5 million revenue loss due to delays in quota approvals. Other companies, including Atlassian and GovTech Singapore, also faced obstacles, with at least $52.6 million in sales hindered by slow quota increase approvals. The capacity crunch has raised alarms among AWS executives, with CEO Andy Jassy emphasizing the urgent need to enhance cloud infrastructure, particularly regarding AI capabilities. However, sources indicate that the issue remained unresolved as of September, with AWS still grappling with capacity constraints. Despite these challenges, an Amazon spokesperson highlighted the rapid growth of Bedrock and the company's commitment to expanding its capacity to meet customer demand. They emphasized that feedback from users is essential for continuous improvement, underscoring Amazon's culture of self-criticism to enhance service offerings. As AWS races to expand its data center capabilities, it has reported substantial growth in power capacity, aiming to double it by 2027. Jassy mentioned during an October earnings call that AWS has added over 3.8 gigawatts of power in the past year, more than any other cloud provider. However, the competition remains fierce. Customers migrating from Bedrock include those seeking better performance and lower latency from Google's Gemini models, which have shown superior benchmarks compared to AWS's offerings. Companies like Thomson Reuters have already transitioned parts of their operations to Google Cloud, citing speed and compliance advantages. AWS's struggles reflect broader industry concerns about the sustainability of massive investments in AI technology, as investors worry about a potential market bubble. While the demand for AI services remains robust, the performance issues faced by AWS highlight the need for the company to refine its strategy to maintain a competitive edge in this rapidly evolving landscape.