AWS EC2 compute evolves to meet agentic AI and physical AI demand

Coverage from SiliconANGLE's livestreaming video studiohelp_outline UPDATED 11:51 EDT / JULY 25 2026 AI AWS EC2 compute evolves to meet agentic AI and physical AI demand by Thomas Godwin Twenty years on, AWS EC2 compute is meeting demand shaped by agentic AI, physical AI, and customers pushing general-purpose cloud infrastructure into new territory its earliest architects never anticipated. That growth is permeating throughout every layer of the platform, according to Art Baudo (pictured), principal product marketing manager and head of EC2 product marketing at Amazon Web Services Inc. Customers are now using AMD-based instances for AI inference workloads alongside traditional high-performance computing tasks, and the breadth of what is moving into the cloud has surprised even longtime cloud advocates. “As that performance has gone up, so has the price performance that we’ve delivered to customers,” Baudo said. “With AI taking off, we’re seeing people use it in multiple different spaces across the industry and across the instance types as well.” Baudo spoke with theCUBE’s Dave Vellante and John Furrier at AMD Advancing AI event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how the AWS and AMD partnership has matured since the first EPYC instances launched in 2018, and how the Nitro system underpins security and performance throughout the full EC2 portfolio. They also covered what the acceleration of AI workloads means for EC2 compute. (* Disclosure below.) EC2 compute expansion reflects AMD partnership depth and agentic AI momentum AWS introduced AMD EPYC processors into EC2 in 2018 and has shipped every subsequent generation, including Turin-based instances launched in 2026. The shift from multithreaded to single-threaded instances between the sixth and seventh generation delivered substantial performance gains that customers in AI and EDA workloads have been able to take advantage of. More recently, AWS has introduced high-frequency instances that combine 5GHz clock speeds with expanded memory configurations to serve workloads that require peak compute and rapid data access. “We’ve tried to continuously make sure we optimize on price performance for customers,” Baudo said. “Some of our most recent instances continue to see a reduction in price performance. That is a continuous focus from the AWS side, and that is no different on the AMD instances.” The AWS Nitro System resides beneath all of this as the architectural foundation that makes EC2 compute’s breadth and security possible. By offloading hypervisor functions to dedicated hardware and software, Nitro allows AWS to deploy instance types much faster and at greater scale while enforcing zero operator access. Baudo said the system extends naturally into AI workloads throughout the full portfolio, from AMD and Graviton instances to Trainium and Inferentia, providing customers with a consistent security baseline, regardless of which compute tier they choose. “The fundamentals of the cloud, the ability to deliver security, the ability to deliver performance on demand, are even greater now,” Baudo said. “Especially in workloads where you have a really heavy burst and a need for lots of CPU performance, and then suddenly you may not need it for a period of time, the cloud is phenomenal at doing that.” On the question of cost, Baudo said AWS is applying the same customer-centric optimization discipline to AI workloads that it applied to general cloud cost management in 2022. The pattern parallels what AWS has done historically, making the economics work so