Bedrock Provisioned Throughput: what it buys, when you need it, and when it pays off.

IWhat Provisioned Throughput is

On-demand bills per token. Provisioned Throughput bills per hour for reserved capacity.

With on-demand inference you pay for each input and output token, and your throughput is limited by shared account quotas. With Provisioned Throughput you buy a number of model units for one specific model. Each MU gives a set number of input tokens and output tokens per minute, across all requests. You pay a fixed price per MU per hour, whatever you actually use.

The hourly price depends on three things: the model (a customized model costs the same as the base model it came from), the number of MUs, and the term. After you buy it, you get a provisioned model ARN and call that instead of the base model ID.

IIOn-demand vs Provisioned Throughput

The choice is usage pricing and shared limits vs capacity pricing and reserved throughput.

IIIWhen you must use Provisioned Throughput

One case is a rule. One is a judgment call.

• Customized models: AWS's PT page says: "If you customized a model, you must purchase Provisioned Throughput to be able to use it." That covers fine-tuned and distilled models. Budget the hosting before you start a fine-tune; see Bedrock fine-tuning and model distillation.
• Paths that cannot be throttled: Even on a base model, a checkout flow or a real-time agent under an SLA may need reserved throughput. First check whether a quota increase or cross-Region inference solves it; see Bedrock quotas.

IVPricing: model units and the three terms

Longer commitments cost less per hour and lock you in.

• No commitment: Highest hourly price. Delete any time. Good for testing a customized model or covering a short spike.
• 1 month: Lower hourly price. You cannot delete the PT until the month ends.
• 6 months: Lowest hourly price. You cannot delete it until the six months end. Only for a stable model and proven volume.

An AWS example shows the shape (pricing page, checked 10 October 2026): for Cohere Command, $49.50 per MU per hour with no commitment, $39.60 on a 1-month term and $23.77 on a 6-month term. Another example: one MU of a Llama 2 model on a 1-month term at $21.18 per hour comes to $15,757.92 for a 31-day month. Rates differ by model, so check the pricing page for yours.

VThe break-even against on-demand

PT wins on a base model only above a certain monthly token volume. Work it out from your own logs.

• 1. Price the PT per month: Hourly price per MU, times the number of MUs, times the hours in the month (about 730). It is the same whether you use it or not.
• 2. Price your traffic on-demand: Take a month of input and output tokens from your logs and multiply by the model's on-demand rates.
• 3. Compare: If step 2 is higher than step 1, and stays higher, PT saves money. If not, stay on on-demand.
• 4. Check capacity: Make sure the MUs you price can carry your peak tokens per minute. Your account manager can give the MU figures.

Two common mistakes: sizing from average traffic (so peaks still throttle) and forgetting that idle nights and weekends are billed. Utilization is the whole game. If your traffic is spiky or still growing, PT on a base model is usually the wrong call.

VIHow to buy and manage Provisioned Throughput

A short checklist.

• Pick the model: A base model, or your customized model. Throughput is reserved for that one model.
• Pick the term and the number of MUs: Size from measured peak tokens per minute. Ask your account manager about MU figures and limit increases.
• Buy it and switch the model ID: Call the provisioned model ARN instead of the base model ID in your inference calls.
• Watch utilization: Track invocations and token throughput in Amazon CloudWatch. Low use means you over-bought.
• Delete it when you are done: Billing continues until you delete the PT. A forgotten no-commitment PT is a silent hourly charge. For committed terms, check whether auto-renew is on.

VIIWho can file AWS funding for a Bedrock build

There are two kinds of AWS money for a Bedrock build. A partner files one. You apply for the other yourself.

Partner-filed funding. An AWS consulting partner can file POC (proof of concept) funding to test a Bedrock workload, whether or not you run on AWS today. If you already run on AWS and are adding new AI work, the partner can file incremental workload (IW) funding instead. For startups this is often $50K to $150K. Depends on the workload; AWS approves. The work has to be real, and the partner has to be the one doing it.

AWS Activate, which you apply for yourself. Activate Founders gives $1,000 to start, up to $5,000, with no investor needed. Activate Portfolio gives up to $200,000 if your VC, accelerator or angel group is an Activate Provider and gives you its Org ID. A consulting partner cannot apply for Activate on your behalf. AWS made Activate credits usable for third-party models on Bedrock in April 2024. Note that AWS promotional credits do not pay upfront fees for Savings Plans or Reserved Instances; check the credit terms for any commitment you plan.

For most startups without a VC, the partner route is the larger of the two. CloudRoute matches you with one vetted AWS partner for free. Partners pay CloudRoute a monthly membership plus 10% of AWS-approved funding.