Rohit Prasad leaves Amazon AI
The tech world is experiencing a rapid transformation from personnel news to structural changes. Amazon's recent restructuring of its AI leadership, coupled with the departure of Rohit Prasad, is an example of this. This reorganization integrates AI models, chips, and quantum computing under new leadership.
Amazon's AI Strategy
Amazon is reorganizing its AI leadership. Rohit Prasad will leave the company at the end of 2025. This decision is part of a larger reorganization of the AI organization. Reuters reported on this development on December 17, 2025.
Amazon CEO Andy Jassy confirmed in a published leadership update, that Prasad is leaving "at the end of this year." The core of the announcement is a revamp: Peter DeSantis will take over a new organization, tasked with jointly leading Amazon's AI models, custom silicon, and quantum computing. Jassy emphasized that DeSantis will report directly to him. Reuters also describes this new unit as a bundle of AI models, chips, and quantum, , signaling a greater AI focus.

Source: news18.com
Rohit Prasad, the long-time head of AI at Amazon and architect of Alexa, is leaving the company.
Jassy places Prasad's start at Amazon in 2013, connected to the early Alexa years. Alexa is described as a service that touches "hundreds of millions" of customers daily. Reuters also positions Prasad as a key figure in the Alexa environment and places his departure in the context of Amazon's AI strategy.
Over the past two years, Prasad was responsible for building "Amazon Nova" and the organization that Amazon internally calls "AGI." Amazon speaks of twelve "foundation models", that would be used by "tens of thousands of companies."
Technological Focus
In the memo, Amazon portrays DeSantis as a long-time AWS manager who was involved, among other things, in central infrastructure components like EC2. Jassy also mentions the acquisition of Annapurna Labs in 2015 as a milestone closely related to Amazon's custom silicon. Reuters directly links this chip line to Amazon's current processor and AI accelerator families, , known in the market under names like Graviton and Trainium.
Amazon itself describes the logic behind this as optimization "across models, chips, and cloud software and infrastructure." This affects the daily work of teams developing GenAI products: : model quality, latency, costs, and hardware availability are closely linked.

Source: msn.com
Rohit Prasad was instrumental in the development of Alexa and other AI-driven products at Amazon.
Product and Application
Amazon positions "Amazon Nova" as a model family available through Amazon Bedrock. AWS documents the use of Nova models in Bedrock via the corresponding APIs and parameters in the official documentation. AWS announced "Amazon Nova 2" for Bedrock on December 2, 2025, , including "Nova 2 Lite" and "Nova 2 Pro (Preview)."
In parallel, Amazon is visibly pushing the topic of "agentic AI," i.e., systems that not only respond but also execute tasks in tools and interfaces. In this context, Amazon describes "Nova Act" as an approach, that automates browser actions, thereby mapping real UI workflows. Amazon's own re:Invent overview bundles Nova, chips, and other AI components as a related line in its product communication. The AWS blog also summarizes the re:Invent 2025 announcements regarding Nova in a collective post.
Research and Development
Jassy writes that Pieter Abbeel will lead the "frontier model research team" within the AGI organization. . In the same memo, Amazon names Abbeel as a co-founder of Covariant and links his experience with generative AI and reinforcement learning to the advancement of its own models. Reuters also picks up on this personnel matter and positions Abbeel as head of frontier model research.
It is not new that Amazon Robotics and foundation model work are being considered more closely together: Amazon announced in 2024 that it would hire Covariant founders and take a non-exclusive license for Covariant's "robotic foundation models." TechCrunch also reported on the hiring of the founders and the licensing deal at the time.
Outlook and Implications
Amazon communicates Rohit Prasad's departure as part of a larger realignment that brings AI models, custom chips, and quantum under a unified leadership. The decision to have Peter DeSantis report directly to the CEO, signals that this topic is no longer just an AWS component but a company-wide focus with infrastructure and product implications. Crucially for customers and development teams, it remains to be seen whether this bundling will lead to better, cheaper, and more reliable AI products more quickly—and whether Amazon will consistently close the gap from models via Bedrock to agentic applications. The Nova models in Bedrock and the documentation of their parameters as well as the focus on agentic AI are central elements here.