Table of Contents
The CEO of AWS noted in his keynote session said: ‘We are in the early stages of a major computing shift driven by accelerated AI’
In sum – what to know:
AI as a long-term infrastructure shift – AWS CEO Matt Garman said the industry is entering a new phase defined by “a major computing shift driven by accelerated AI,” requiring sustained global capacity expansion.
Power availability shapes AI scalability – While avoiding specific forecasts, Garman tied AWS’s regional growth plans to working with utilities and policymakers to ensure enough electrical capacity for high-density AI clusters.
Commitment to silicon and dedicated AI environments – Garman cited 15 years of work with Nvidia and AWS’s own chip development as essential to giving customers performance and cost advantages for large-scale model training.
Amazon Web Services (AWS) used its re:Invent 2025 keynote to signal a long-term shift in how cloud and AI infrastructure will be built and financed, as CEO Matt Garman emphasized the growing interdependence between massive AI workloads, power availability, custom silicon and global data-center expansion.
Garman framed the moment as a structural transformation of the industry. “We are in the early stages of a major computing shift driven by accelerated AI,” he said, noting that customers are asking AWS for significantly more capacity across regions. He added that AWS continues to expand infrastructure globally, and highlighted that “some of the biggest customers are using Amazon Aurora at this scale,” as an example of what workloads now demand.
A central theme of Garman’s remarks was the critical role of electrical capacity in determining how far AI can scale. The executive repeatedly tied infrastructure expansion to available power, saying AWS is focused on “supporting customers with predictable performance at scale” while working closely with utilities and policymakers to ensure regions can support high-density AI clusters.
Garman also highlighted AWS’s continued investment in silicon development. “We’ve been working closely with Nvidia for more than 15 years,” he said, underscoring that AWS intends to continue offering a range of accelerators as model complexity grows. He also referenced AWS’s long-term work building custom chips, describing it as a way to optimize performance and cost for customers training and deploying AI models.
He linked these infrastructure and silicon investments directly to customer requirements for dedicated AI capacity. “Customers want more accessibility,” he said, adding that AWS is increasingly building environments that support large-scale model training with predictable availability and faster access to high-performance compute.
Garman added that AWS is focused on helping customers “run their largest and most demanding applications” across regions while continuing to expand foundational services underneath them.
