AWS EC2 Pricing & Compute Negotiation Guide.

How AWS actually prices EC2.

EC2 has five canonical pricing models, each with a different commitment shape and a different blended discount. Most enterprise customers blend three to four of them in production. The mix — not any single model — drives effective compute cost.

The published discount bands are starting points. Inside an Enterprise Discount Programme, the EDP discount stacks on top of the Savings Plan or RI discount on the on-demand-equivalent value. That stack is where the real negotiation happens. We have repeatedly secured EDP compute tiers that produce a 55-65% blended discount on what would otherwise be on-demand compute.

Instance family economics

Not all instance families are priced equally. The general-purpose M-family, the compute-optimized C-family, and the memory-optimized R-family all carry different per-vCPU costs. Within each family, the generation matters more than the size. An m6i.xlarge is materially cheaper per unit of work than an m5.xlarge, despite a smaller absolute price difference. Graviton-based instances (M6g, C7g, R7g) carry a 20% list discount over Intel equivalents and typically deliver 15-30% better price-performance once benchmarked.

Right-sizing alone — without any negotiation — usually frees 18-25% of EC2 spend in environments that have never been audited. Most AWS bills carry a long tail of oversized instances and forgotten dev/test capacity. We tag these in week one of every compute engagement.

Spot pricing reality

Spot is the largest source of unrealized savings in most environments. Interruption rates have dropped meaningfully — most production-quality instance pools see interruption rates under 5% per month. Workloads that tolerate restart (CI/CD, batch processing, async jobs, stateless services behind a queue) should be running on Spot. Workloads that cannot tolerate restart are still candidates for Spot via diversified pools and fallback strategies.