specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Power BI
providerId: power-bi
created: '2026-05-04'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-05'
reconciled: true
tags:
  - Analytics
  - Business Intelligence
  - Microsoft
  - Rate Limiting
description: >-
  Power BI REST APIs apply per-user throttling to prevent overuse, returning
  HTTP 429 with a Retry-After header when a calling identity exceeds the
  documented per-time-window threshold. Numeric per-second/minute ceilings are
  not exhaustively published; specific operation groups (admin, embed-token,
  capacity, dataset refresh) carry their own published per-hour and per-day
  caps in the relevant operation reference. For Power BI Embedded and Fabric
  F-SKU capacity, throughput is governed by the v-core capacity rather than
  by API quota: utilization is evaluated every 30 seconds and capacities
  exceeding 100% enter "interactive request delay" mode.
sources:
  - https://learn.microsoft.com/en-us/rest/api/power-bi/
  - https://learn.microsoft.com/en-us/power-bi/developer/embedded/embedded-faq
responseCodes:
  throttled: 429
  capacityOverload: 429
headers:
  retryAfter: Retry-After
  requestId: RequestId
limits:
  - name: Per-user request throttling
    scope: user
    metric: varies
    limit: per-operation thresholds; see operation reference
    notes: Power BI throttles when a single user exceeds the per-operation time-window threshold; HTTP 429 + Retry-After is returned.
  - name: Embed-token generation (Pro / PPU)
    scope: user
    metric: varies
    limit: limited number per master account or service principal (development testing only)
    notes: Production embedding requires assigning a Power BI Embedded (A SKU) or Fabric (F SKU) capacity; capacity removes the embed-token cap.
  - name: Embedded / Premium capacity utilization
    scope: capacity
    metric: v-core_seconds
    limit: '30 seconds of CPU per v-core per 30-second cycle'
    notes: Each evaluation cycle aggregates interactive (current cycle) plus background (1/2880 of past 24h) operations.
  - name: Capacity overload behavior
    scope: capacity
    metric: percent_overload
    limit: 'no delay below 10% overload; up to 20s interactive delay at 100% overload'
    notes: Capacity remains in interactive-request-delay mode until the previous evaluation drops below 100% utilization.
policies:
  - name: Honor Retry-After
    description: On 429, clients must wait the number of seconds indicated in Retry-After before retrying.
  - name: Capacity-bound throttling
    description: For Embedded and Fabric workloads, scale up the v-core capacity (or use autoscale where supported) rather than retrying through 429 storms.
  - name: Cross-region operations
    description: Operations that download files from a different region than the call origin may take longer; treat regional differences as latency, not throttling.
  - name: Operation-level limits
    description: Specific operations (admin scans, dataset refresh, embed-token generation) have their own caps documented in the per-operation reference; consult the operation page rather than assuming a uniform global limit.
  - name: Development vs production embedding
    description: Trial embed tokens with Pro/PPU are for development testing only; production embedding must run on purchased capacity.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com
notes: >-
  Microsoft documents the throttling mechanism (429 + Retry-After) and the
  capacity utilization model in detail but does not publish a single global
  per-second numeric limit for the Power BI REST API. Per-operation limits
  appear in each operation's reference page.