Why do all my Bedrock quotas show 0 and every call returns ThrottlingException?

AWS says the quotas applied to an account can be lower than the published defaults depending on account age, usage, payment history and other factors. Several new-account reports on AWS re:Post describe applied values of 0 marked “Not adjustable”. The practical fix reported there is a Support case asking AWS to apply the default quotas, with the Region, model, quota codes and failed request IDs.

Why do I get AccessDeniedException for a model that is listed in my Region?

Common causes in the AWS docs: the AWS Marketplace subscription that starts on first use is still being set up (up to 15 minutes), the Anthropic first-time-use form has not been submitted, IAM or SCP conditions block the first invocation, or the model needs extra account-level access that AWS Sales grants.

Can I get a quota increase for a Legacy model?

No. AWS says quota increases are not granted for models in Legacy or Deprecated status and recommends migrating instead.

“Quota is 0” and “not available for this account”

Q: What does “Invocation of model ID … with on-demand throughput isn’t supported” mean?

That model is only offered in your Region through a cross-Region inference profile. Call it with the profile ID (for example the us., eu., apac. or global. prefixed ID) instead of the bare model ID, and allow the inference-profile ARN in IAM.

The five Bedrock problems I would have seen most as a support manager, in plain English. Each fix points at the AWS page it comes from. AWS re:Post answers are community or AI-assisted answers, not official AWS policy, so they are marked as such.

1. Every quota is 0 and the first call fails with ThrottlingException

What you see: ThrottlingException: Too many tokens per day on your very first prompt, even in the playground. In Service Quotas the AWS default is in the millions, but your applied value is 0 and marked “Not adjustable”.

Why: AWS documents that the quotas applied to an account “might be lower than the default values” based on “regional factors, account age, usage history, payment history, fraudulent usage” and past quota requests, and that increases are not granted automatically (Quotas for Amazon Bedrock). Several new-account reports on re:Post describe exactly this all-zero state (re:Post 1, re:Post 2, re:Post 3).

What to do:

Open Service Quotas, choose Amazon Bedrock, and check the applied value for your model in your Region. Check both endpoints: bedrock-runtime and bedrock-mantle quotas are tracked separately (AWS, mantle quotas).
Open a Support case. The re:Post answers suggest saying plainly that the applied value is 0 (you are asking for the defaults, not an increase), and including the Region, model ID, quota codes, a screenshot, and the x-amzn-requestid of a failed call.
From my time running Enterprise Support: a clear, single case with request IDs moves faster than several duplicate cases. Basic Support has no technical-support SLA, so if a launch depends on it, plan for the wait.

2. “Invocation of model ID … with on-demand throughput isn’t supported”

Why: In your Region the model is only offered through a cross-Region inference profile. AWS’s knowledge-center article quotes the exact error and the fix: pass the inference profile ID or ARN as modelId, for example us.anthropic.claude-3-5-sonnet-20241022-v2:0 instead of anthropic.claude-3-5-sonnet-20241022-v2:0 (AWS re:Post Knowledge Center).

Also check IAM: if your policy allows only the foundation-model ARN, calls through a profile return AccessDeniedException. Add arn:aws:bedrock:*::inference-profile/… to the Resource list (same article). On every model page here, the grid shows which Regions are “Inference profile only” and lists the exact profile IDs from the AWS model card.

3. AccessDeniedException on a model that is listed in your Region

Marketplace subscription still finishing. The first call to a third-party model starts an AWS Marketplace subscription in the background. Setup can take up to 15 minutes, and after you fix permissions it can take 2 more minutes (Request access to models). The MPAgreementBeingCreated error means “try again after 15 minutes” (error codes).
Anthropic first-time-use form. Anthropic models on bedrock-runtime need use-case details submitted once per account or once at the organization’s management account. The agreement can sit in PENDING; call GetFoundationModelAvailability and read agreementAvailability (AVAILABLE, PENDING, NOT_AVAILABLE, ERROR) rather than guessing (AWS). The error code for a missing form is FTUFormNotFilled.
The first call is blocked by your own policy. If bedrock:InvokeModel is restricted with conditions (for example an aws:SourceIp condition), the first, subscription-starting call must satisfy them or the subscription never starts (AWS).
Payment or location problems. “Invalid payment” and “restricted geo-location” are the common causes of a failed Marketplace agreement (error codes).
Some models need extra account-level access. AWS says that completing the activation steps doesn’t grant it, accounts in the same Organization can differ, and the route is AWS Sales (AWS).

4. Legacy models: “it worked last month”

Once a model enters Legacy, new customers can’t adopt it and existing customers may lose access after 15 days of inactivity. You also can’t create new Provisioned Throughput for it (Model lifecycle). AWS also says quota increases are not granted for Legacy models (runtime quotas). If a dev account that sat idle suddenly gets denied, check the retirement calendar first.

5. 429 vs 503 vs 529

429 ThrottlingException: you hit your account quota. Back off with jitter; for sustained load consider Reserved Tier or Provisioned Throughput (AWS).
503 ServiceUnavailable: service-side capacity, “not related to your account-level quotas”. Retry, try another Region, or use cross-Region inference (same page).
529 overloaded_error: the model itself is short on capacity. Retry with backoff and respect Retry-After (same page).

Asking for more quota

On bedrock-runtime, requesting an increase on “Cross-Region InvokeModel tokens per minute” for a model lets Support offer the on-demand TPM and the cross-model tokens-per-day quota at the same time. AWS notes that priority goes to customers who are already using their current allocation, so a request from an idle account may be denied (AWS).