The market has underestimated Amazon AWS's "AI potential": The "deeply integrated" Claude, API business has surpassed OpenAI

The AI growth potential of Amazon Web Services (AWS) is severely underestimated, as the API business of Anthropic, which has a deep partnership with AWS, is significantly contributing to its revenue.

On September 3rd, Barclays' latest analysis report indicated that the deep collaboration between Anthropic and Amazon AWS is providing significant growth momentum for the cloud service giant, but the market has not fully recognized the potential of this AI-driven growth. If AWS can maintain its collaboration on training workloads with Anthropic, the company is expected to achieve unexpected revenue growth in the fourth quarter.

Barclays analysts estimate that Anthropic currently contributes about 1% of AWS's growth (in the second quarter of 2025), but with the dual boost from Claude 5 training and existing inference revenue, this contribution could rise to 4% per quarter. The key point is that Anthropic's API business has already surpassed OpenAI in scale and is growing at a faster pace.

The report states that Anthropic is expected to bring approximately $1.6 billion in inference revenue to AWS in 2025, and its annual recurring revenue (ARR) is projected to leap from $1 billion at the beginning of the year to $9 billion by the end of the year. However, there have been complaints in the industry regarding the limitations on accessing Anthropic's models through the AWS Bedrock platform, indicating that the partnership between the two companies may face some challenges.

Barclays maintains an "Overweight" rating on Amazon, with a target price of $275, representing a 21.7% upside from the current stock price.

Anthropic's API Business Has Surpassed OpenAI

According to Barclays' analysis, Anthropic has established a significant advantage over OpenAI in the API business sector.

Data shows that 90% of Anthropic's revenue comes from its API business, while only 26% of OpenAI's revenue comes from APIs, with its main revenue still relying on the ChatGPT consumer product.

Specifically, Anthropic's API business is expected to generate $512 million in revenue in 2024, and is projected to skyrocket to $3.907 billion in 2025, a year-on-year growth of 662%. In contrast, OpenAI's API business is expected to generate $1 billion in revenue in 2024, growing to $1.8 billion in 2025, an increase of 80%

This difference mainly stems from the explosive growth of AI integrated development environment (IDE) applications.

AI programming tools like Cursor and Lovable obtain model authorization through Anthropic's Direct API, paying per million tokens. Barclays estimates that the average Cursor Pro user contributes about $5 in revenue to AWS each month.

The launch of models such as Claude 3.5 (released in June 2024) and Claude 3.7 (released in February 2025), particularly the reasoning and thinking chain capabilities integrated into the latter, has propelled Anthropic's leading position in the API field.

The report indicates that the AI integrated development environment as a category is expected to exceed $1 billion in ARR by 2025, while this figure was nearly zero in 2024.

AWS Expected to Achieve Exceeding Growth in Q4

Barclays predicts that if Amazon's cloud service AWS can maintain its partnership with Anthropic, revenue growth in the fourth quarter is expected to exceed market expectations by about 2%.

Currently, the market consensus expects AWS's fourth-quarter revenue to grow by 18%, but Anthropic's contribution may significantly push the actual growth rate beyond this expectation.

Analysts point out that Anthropic may begin pre-training Claude 5 in the fourth quarter, which will contribute about 1.5% growth to AWS. Combined with reasoning revenue, Anthropic may contribute a total of 4% to AWS growth.

The research report states that Barclays' model assumes AWS's non-AI workloads maintain a year-on-year growth trend of 14%-16%, consistent with performance over the past few quarters. On this basis, Anthropic's additional contribution will significantly enhance the overall growth rate.

AI Capacity Expansion Supports Long-term Growth Prospects

To support the rapid growth of AI business, AWS is significantly expanding its AI computing capacity.

Barclays estimates that by the end of 2025, AWS may have over 1 million H100 equivalent AI capacities, thanks to the rollout of Blackwell GPUs and 400,000 Trainium "Project Rainier" chips.

However, there is still debate over whether AWS has enough AI capacity to support all of Anthropic's growth activities. During the second-quarter earnings call, Amazon's management continued to express the standard statement that demand exceeds supply, without providing much new confidence.

Barclays' analysis shows that the new AI capacity added since the launch of ChatGPT is expected to exceed 1 million H100 equivalent computing power by the end of 2025. This capacity expansion is crucial for supporting the rapid growth of partners like Anthropic.

Deep Cooperation Relationship Faces Potential Challenges

Barclays stated that while the partnership between Anthropic and AWS brings significant benefits, the relationship also faces some tests.

According to previous media reports, there have been some complaints in the industry about accessing Anthropic's models through Amazon AWS Bedrock, indicating that there may be some degree of rift in the AWS/Anthropic relationship.

More concerning is that important clients like Cursor (one of Anthropic's largest IDE clients) are beginning to turn to OpenAI's GPT-5 API as their default choice. While users can manually switch back to Anthropic, this stickiness is far weaker than the default settings of white-label paid search in the consumer space.

Despite some uncertainties, Barclays remains optimistic about AWS's long-term prospects in the AI field.

The bank's analysts pointed out that while the "broad expansion" phase of AI deployment in 2025 has not truly arrived, a few large AI laboratories are currently generating most of the AI revenue for hyperscale cloud service providers, and AWS's deep cooperation with Anthropic places it at the core of this trend.

It is worth noting that Barclays' analysis is based on the assumption that 70% of Anthropic's revenue is hosted on AWS, with the remaining 30% on Google Cloud Platform (GCP).

Anthropic is the only AI laboratory with three different infrastructure teams managing training and inference on three different architectures: GPU, TPU, and Trainium, providing it with flexibility in terms of cost and availability