Amazon Elastic Compute Cloud is the virtual-machine service that launched AWS in 2006 and remains the foundational compute primitive for most AWS workloads. EC2 offers hundreds of instance types optimized for different mixes of CPU, memory, storage, and networking, including GPU instances (P-series, G-series) for AI workloads.
EC2's role in the broader AWS product set has shifted over twenty years. The initial proposition — on-demand virtual machines — has been partially displaced by higher-level services (Lambda, Fargate, managed databases) for specific workload classes. But for workloads that need fine-grained control over the compute environment, EC2 remains the default.
For AI workloads specifically, EC2 offers GPU instances and Trainium-based instances (custom AWS silicon). These compete with the equivalent offerings from GCP (A3 VMs with H100, TPU VMs) and Azure (ND-series with H100, H200). Pricing and capacity availability vary significantly by region and instance type, and negotiating capacity commitments has become a strategic procurement function for AI-heavy enterprises.