AWS Seeks Enhanced Control over AI Models and Cost Management
In a significant announcement made to investors, Amazon has revealed its ambitious plans to shift its focus towards developing custom artificial intelligence (AI) models. This move, according to the company, aims to enhance cost efficiency and specialization, which will ultimately set it apart from competitors in the AI space. In a landscape where AI capabilities are increasingly comparable across the board, Amazon asserts that differentiating factors will become vital for success.
On July 31, 2026, Amazon’s CEO, Andy Jassy, highlighted the company’s strategy in a detailed statement, emphasizing the potential benefits of utilizing niche models designed for specific audiences. This tailored approach is intended to keep prices lower for customers compared to relying on third-party developed models. As part of this strategy, Amazon has made the decision to redirect both engineering talent and computing resources to the creation of its own frontier AI model while gradually phasing out existing flagship models such as Nova AI.
"Our approach provides us with greater control over costs, not only for our consumer applications but also drives down costs for our customers," Jassy explained. He also emphasized the importance of prioritizing model development based on the needs of both external clients and internal stakeholders.
The financial performance of Amazon Web Services (AWS) reflects the impact of these efforts, as the company reported annual revenue exceeding $25 billion—an increase of more than 100% compared to the previous year. In the latest earnings report, it was noted that a rapidly growing division within AWS, known as Amazon Bedrock, now offers an array of AI models, recently incorporating cutting-edge AI offerings like OpenAI’s GPT-5.6, Anthropic’s Claude Opus 5, Google DeepMind’s Gemma 4, and SpaceXAI’s Grok 4.3.
Bedrock is engineered to facilitate the building, testing, and deployment of new AI agents while guiding users through intricate processes related to security, data management, memory, and identity. Jassy commented on the technical proficiency of companies in creating their own foundational models, pointing out that many firms will focus on developing smaller models tailored to leverage proprietary datasets.
Echoing this sentiment, Jassy stated, "AWS and Amazon can maintain a robust business without needing a proprietary frontier model. This is because the market won’t have a singular AI model dominating the landscape." Despite this reality, the company recognizes the potential competitive edge that could come from introducing a custom-built model, as the AI industry evolves towards competition based on factors beyond mere capability.
With various models reaching similar levels of performance, Jassy affirmed that the competitive landscape will pivot towards price, mirroring AWS’s long-standing strategy within computing infrastructure. Notably, AWS has developed its own CPU chip called Graviton, which delivers 30% to 40% better price-performance ratios compared to competing options. This chip is utilized by clients for post-training reinforcement learning and agent tool application, showcasing AWS’s commitment to cost-effectiveness.
"In a setting where a player like us is consistently focused on reducing price, performance, and associated costs for customers, it adds value and keeps AI models more cost-effective," Jassy noted. Furthermore, the introduction of a dedicated frontier model will enable Amazon to hone its focus on performance metrics that matter most to its clientele.
Most companies will likely leverage turnkey agent services even as they create their own tailored agents. Jassy mentioned existing successful examples, such as Claude Code, Codex, and Amazon’s own Kiro, which has reportedly become 50% more cost-effective than competitors and experienced a threefold increase in usage from one quarter to the next.
As part of the broader strategy, Amazon is scaling back its current Nova foundation models, including high-end offerings like Premier and Omni, and models focused on video and image generation, such as Reel and Canvas. However, Jassy reassured stakeholders that Kiro, which specializes in coding, will remain an influential part of Amazon’s AI portfolio.
Looking ahead, Jassy forecasted that within a few years, the market might witness at least half a dozen AI models that will stand comparably against each other, all of which will be housed within Amazon Bedrock, including their custom-built model.
In the wake of these updates, Amazon’s financial performance has displayed resilience. Despite missing certain revenue projections, Amazon reported total net sales of $200.6 billion for the second quarter of 2026, exceeding estimates and demonstrating a substantial year-over-year increase. The company’s stock saw a notable rise in after-hours trading, climbing to $258 per share, the highest since June 1, 2026.
While Amazon has projected revenue of $197 billion to $202 billion for the third quarter, analysts had initially anticipated sales around $204 billion, indicating a challenging outlook ahead. Nevertheless, the company’s robust performance in the AI sector and its focus on niche model development may create new opportunities for growth that stakeholders will keenly watch in the coming months.