Amazon has reorganized several advanced engineering teams into a single group that now manages artificial intelligence models, chip design and quantum computing research.
This updated structure includes the team responsible for the Nova family of core models, which powers a growing number of services used by both internal teams and external cloud customers. It also includes the Annapurna Labs group, which develops custom processors such as Graviton and Trainium that support many of Amazon’s compute and machine learning workloads.
Previously, these teams were split into separate parts of Amazon’s corporate structure. AI model development was managed largely within the AGI group, operating alongside the Alexa voice assistant team. Chip engineering was centered on AWS infrastructure, while quantum computing was managed as a separate research effort.
These units now report under a common leadership framework, with shared goals and aligned deadlines. The new setup connects machine learning model development directly to the underlying infrastructure.
Why it’s important: The teams now working together build and maintain the systems used across Amazon’s technology stack. With this new structure, they follow a unique development process that connects research and deployment more directly than before.
- Development of foundation models in the context of infrastructure engineering: Nova templates power applications for customer service, developer tools, and enterprise tasks. They rely on large-scale IT systems and frequent updates. By moving model development to the same group that manages the underlying hardware, Amazon reduced the steps needed to test and release changes. Engineers responsible for model behavior now work directly with those running the environments in which models operate, removing a handoff that previously slowed builds and tuning.
- A shared workflow for developing custom processors and models: Amazon’s internal chips are designed for specific purposes. Graviton focuses on general-purpose computing, while Trainium is designed for model training. The teams that design these processors now work closely with the teams that build the models that use them. Testing and architectural planning occur within a single workflow. This eliminates the need for separate alignment meetings, since technical decisions and deployment plans are handled within the same reporting structure.
- Quantum computing research is integrated with applied engineering: Amazon quantum The team works on the development of physical hardware and algorithms, alongside long-term research. The group is now part of the same engineering stream that supports production models and computational infrastructure. Quantum engineers work alongside teams that manage deployed systems. Although their research objectives remain distinct, they now use shared planning processes and tools.
- Management changes accompany reorganization: Rohit Prasad, who led the AGI group and played a pivotal role in AmazonEarly conversational AI efforts will leave the company. His job included shaping the direction of Alexa and building the team behind Nova. Pieter Abbeel, who joined Amazon through the acquisition of Covariant, now leads research into foundation models. His experience spans robotics and learning systems, and he continues to guide applied work across the company.
- AI engineering is no longer distributed across business units: Previously, Amazon’s AI efforts were split between separate divisions. Alexa and AWS followed different timelines and ran separate infrastructure, even when working on related features. This separation has been removed. Model training, hardware design, deployment coordination, and long-term system improvements are now within a single organization. This reduces duplicate work and allows teams to share tools and systems without additional approval levels.
Go further -> Andy Jassy Makes Leadership Announcement at Amazon – Amaarea
Amazon appoints new AI chief amid battle to take on tech rivals – Bloomberg
Trusted information for technology leaders
Our readers are CIOs, CTOs and senior IT executives who rely on The IOC national magazine for smart, curated approaches to the trends shaping business, from GenAI to cybersecurity and beyond.
Subscribe to our newsletter 4 times a week to stay up to date with the information that matters.