Bedrock and GenAI spend on AWS is driven by tokens and model choice, not by infrastructure. Bedrock charges per input and output token on demand, with optional Provisioned Throughput that reserves model capacity for a committed term. The biggest savings come before any pricing decision: choosing the smallest model that meets the task, trimming prompt and output length, and caching repeated calls. Treat GenAI as its own governance line, because it grows faster than any other part of the bill and the usual rightsizing playbook does not apply.
This sits alongside the rest of AWS data and network spend, but the mechanics are distinct. Here is how to keep it under control without slowing the teams shipping AI features.
What actually drives the GenAI bill
Three variables set the cost of every Bedrock call: the model, the input token count, and the output token count. Output tokens usually cost more than input, and a larger model multiplies the rate on both. That means the most expensive pattern is a verbose prompt sent to an oversized model that returns a long answer, repeated at scale.
The levers follow directly. Right size the model to the task, because a smaller model often handles classification, extraction, and routing at a fraction of the cost of a frontier model reserved for genuinely hard generation. Trim prompts and cap output length. And cache deterministic or repeated calls so you are not paying tokens twice for the same answer. These three moves change the bill more than any infrastructure tuning.
On demand tokens or Provisioned Throughput?
Bedrock gives you two pricing modes, and the choice mirrors the commitment decision everywhere else on AWS.
- On demand per token. You pay only for what you call. Ideal for spiky, experimental, or low volume workloads where reserving capacity would mean paying for idle time.
- Provisioned Throughput. You reserve model capacity for a committed term at a flat hourly rate, which buys guaranteed throughput and a lower effective rate once volume is high and steady. It is the GenAI equivalent of a commitment: it pays off only against a defensible forecast of sustained usage.
The trap is buying Provisioned Throughput for a workload that is still finding its shape. Reserve capacity only once traffic is steady enough that the committed hours beat accumulated token charges. Until then, on demand keeps you flexible.
Govern it like a budget, not an experiment
GenAI spend escapes control when no one owns it. Put the same discipline on it that you put on the rest of the estate: tag AI workloads so the Cost and Usage Report attributes every token to a team, set budgets and alerts per application, and review GenAI as its own line in the monthly cadence. Watch for the quiet drivers, such as retrieval pipelines that re embed unchanged documents or agents that loop more than expected.
A European SaaS company shipped an AI assistant that defaulted every call to a frontier model and capped nothing. Routing the simple intents to a smaller model, trimming the system prompt, and caching repeated lookups cut the per request token cost sharply, and only the genuinely hard generation stayed on the larger model. With volume then steady, a slice of traffic moved to Provisioned Throughput at a lower effective rate. Figures are verified against billing data and anonymised.
Where GenAI meets the rest of the bill
AI workloads rarely stand alone. The retrieval layer pulls from S3 and moves data across regions, the logging is heavy, and inference on your own models brings GPU capacity into the picture. Watch CloudWatch Logs, the quiet budget eater, when AI services log verbosely, and govern training and tuning compute alongside inference, which is covered in SageMaker cost control for ML teams. The whole estate picture lives in the AWS cost optimization guide.
Frequently asked questions
How is Bedrock priced on AWS?
When does Provisioned Throughput pay off?
What drives GenAI cost most on AWS?
Govern your GenAI spend with us
We help enterprises put governance on Bedrock and GenAI before it becomes the largest unmanaged line on the bill. Our guarantee: we reduce your cloud spend or we reimburse our service fee. Pricing is either a Fixed Fee scoped up front or Gainshare, a share of verified savings with no retainer and no risk.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.