Amazon (AMZN 1.27%) has been in the news recently, mainly for its $38 billion deal with OpenAI. This partnership, which involves OpenAI running artificial intelligence (AI) training and inference workloads on AWS’ cloud infrastructure, has put Amazon at the forefront of the AI cloud discussion.

Image source: Amazon.

Over the past few years, Microsoft‘s (MSFT +1.37%) Azure and Alphabet‘s (GOOG 0.77%) (GOOGL 0.86%) Google Cloud have been quickly gaining the attention of developers, start-ups, and businesses. While Azure has been benefiting from its deep integration with OpenAI’s models, Google Cloud has leveraged Gemini AI models and other AI tools to drive enterprise adoption. On the other hand, Amazon seemed to be lagging as it was known mainly for leveraging AI to improve its e-commerce, cloud, and advertising offerings, rather than leading AI innovation.

However, that perception is now shifting, particularly following the highly publicized AWS deal. So, the real question now is whether AWS is quietly winning the 2025 AI Cloud race, or if it is only playing defense to avoid losing to its AI competitors.

I think that while Amazon has not yet won the generative AI-powered cloud race, it has definitely made a solid move in that direction.

Cloud infrastructure leadership

AWS exited the third quarter of 2025 with a 29% share of the global cloud infrastructure market, higher than Azure’s and Google Cloud’s share of 20% and 13%, respectively.

Amazon

Today’s Change

(-1.27%) $-3.02

Current Price

$234.56

In the third quarter, AWS generated $33 billion in revenue, up 20.2% year over year, marking the fastest growth since 2022. AWS’ operating income also climbed 9.6% year over year to $11.4 billion, translating into a strong operating margin of over 34%. This profitability is impressive, primarily since the company is aggressively investing in expanding data center capacity and developing custom silicon.

With an annualized run rate of $132 billion and a backlog of $200 billion at the end of the third quarter, AWS also boasts a massive scale and high near-term revenue visibility. CEO Andrew Jassy has even pointed out that the backlog number does not include several deals from October 2025, which together are higher than the total deal volume in the third quarter.

In the most recent quarter, Azure and Google Cloud revenues were up year over year by 39% and 34%, respectively. While both of these competitors are growing faster, AWS has also started to pick up the pace.

AI capacity

Amazon has added more than 3.8 gigawatts of data center power capacity in the past 12 months, more than any of its competitors. The company expects to add at least one additional gigawatt of power capacity in the fourth quarter. While the current capacity is already double that of 2022, the company aims to again double it by 2027. This rapid expansion will enable AWS to capture a significant share of the surging AI training and inferencing demand, secure even more long-term AI deals, and monetize new capacity as it comes online.

Custom silicon advantage

Amazon has also created custom chips, such as Trainium for training AI models and Inferentia for AI inference, which offer better price-performance than many other AI chips. Trainium2 is already a multibillion-dollar business, with revenue growing 150% quarter over quarter in the third quarter. Trainium2 chips are already fully subscribed. In fact, Amazon’s Project Rainier has built a massive AI compute cluster across multiple data centers in the U.S., comprising approximately 500,000 Trainium2 chips. Amazon expects to expand the cluster to 1 million Trainium2 chips by the end of 2025. Anthropic is using this AI cluster to build and deploy its Claude AI model.

While Trainium2 is already positioned as 30% to 40% better in price-performance than many other GPU options, the company is now gearing up for the preview of the Trainium3 chip at the end of 2025 and its volume availability in early 2026. The company expects the new chip to deliver 40% better price-performance than the Trainium2 chip. The higher price-performance can become a competitive edge, especially as customers try to control explosive costs while transitioning AI workloads from training to production environments.

Despite this, AWS is also offering its customers multiple chip options from Nvidia, Advanced Micro Devices, and Intel. This reduces the risk of vendor lock-in for clients.

Complete AI stack

Besides hardware, AWS offers platform services such as SageMaker and Bedrock. SageMaker enables clients to build and deploy their custom foundational models, while Bedrock provides them with access to a range of foundational models for deploying their AI inference workloads. This model-agnostic approach, coupled with a focus on high price-performance, can help expand its enterprise customer base in the coming quarters.

Amazon is also focusing on the agentic AI opportunity and has offered an open-source capability called Strands on AWS. This helps clients build agents based on any foundational models. AWS has also introduced a set of infrastructure building blocks called AgentCore, to enable clients to deploy agents securely and scalably in production. AWS has already built agents such as Kiro, an agentic coding IDE, and the Transform agent to demonstrate the real-time utility of its agentic services to enterprises.

Is AWS a winner?

AWS’ rapid capacity expansion, custom silicon development, and focus on AI platform services demonstrate its increasing momentum in the AI cloud race. While Microsoft and Alphabet are growing at a faster rate, AWS is also well positioned to emerge as one of the biggest winners of the current AI cloud boom.