01
PRIMARYStandard Bedrock access via Cursor
No existing Bedrock commit. Marketplace procurement.
Ready to ship.
View section →CONFIDENTIAL · CURSOR × AWS
This brief is restricted. Enter the shared password to continue.
Cursor × AWS · Partnership Architecture
01
PRIMARY PATH · READY TO SHIPCustomer has no existing Bedrock commit and consumes Bedrock through Cursor’s tenancy.
Ready to ship. Prior Bedrock performance issues have been resolved. Preferred procurement path is AWS Marketplace. AWS sellers receive full quota attainment.
BUYER
No existing Bedrock commit. Procures via AWS Marketplace.
CONTROL PLANE · VPC
Runs Bedrock in Cursor’s VPC. Consumption decrements against Cursor’s PPA.
RETAINS 100% TOP-LINEAWS MODEL PROVIDER
Native model picker integration. Full product parity with Anthropic, OpenAI.
WHY THIS PATH MATTERS
Use Case 1 is the straightforward path for customers without AWS commitments. Bring Bedrock back into the model picker natively — prior performance issues are resolved. Simple enough to ship as-is; AWS sellers get full quota attainment via Marketplace.
VS. USE CASE 2
No decrement routing complexity.
VS. USE CASE 3
Standard Marketplace tenancy.
Customer → Cursor
Pays Cursor via AWS Marketplace. Cursor retains 100% top-line.
Cursor → AWS
Cursor pays AWS for all Bedrock inference served from Cursor’s VPC.
Cursor PPA decrement
Bedrock consumption decrements Cursor’s own Bedrock commitment.
Infra decrement
Standard services (EC2, S3) also decrement Cursor’s PPA.
— USE CASE 1 · STANDARD — ↓ NEXT USE CASE —
02
KEY UNLOCK NEEDEDCustomer has an existing Bedrock PPA and wants Cursor usage to count toward it.
AWS to confirm per-customer attribution support in PRM so Bedrock consumption routed through Cursor decrements the customer’s Bedrock PPA — not bulk account totals. This is the single unlock required to move Use Case 2 from design to production.
ENTERPRISE BUYER
Holds existing Bedrock PPA with AWS. Contracts with Cursor via AWS Marketplace.
CONTROL PLANE · VPC
Runs Bedrock inside Cursor’s VPC. Tags consumption per customer and reports via PRM.
RETAINS 100% TOP-LINEAWS MODEL PROVIDER
Serves inference from Cursor’s VPC. 100% of Bedrock revenue flows to AWS.
WHY THIS ARCHITECTURE MATTERS
Use Case 2 is the only path that lets Cursor retain full top-line economics while giving the customer decrement credit against their Bedrock PPA — unlocking enterprise buyers whose AWS commitments would otherwise block adoption.
VS. USE CASE 1
Same Cursor VPC architecture; customer now gets credit on their existing Bedrock commitment.
VS. USE CASE 3 (BYOK)
No credential-propagation rebuild. Primarily commercial/ops design.
Customer → Cursor
Full list via Marketplace. Cursor retains 100% top-line.
Cursor → AWS
Pays AWS for Bedrock inference from Cursor’s VPC.
AWS → Cursor
Rev-share rebate, preserving net economics.
Bedrock decrement
PRM per customer → decrements Customer’s PPA.
Infra decrement
EC2, S3 decrement Cursor’s PPA.
— USE CASE 2 · BYO-Commit — ↓ NEXT USE CASE —
03
EXCEPTION PATH · FAST FOLLOWRegulated customers (banks, JPMC, Citi) who require data residency in their own VPC.
Credential propagation bug blocks Tab, codebase indexing. GitHub issue filed 2026. Fixable — not invested because it undermines revenue without a commercial model.
Revenue retention construct for Cursor still needs to be documented to move this use case forward.
REGULATED BUYER
Data residency required. Banks, JPMC, Citi tier.
AWS BEDROCK
Runs in customer’s VPC. Decrements customer’s account.
CONTROL PLANE ONLY
Cross-account IAM role access to customer’s Bedrock. Credential propagation bug blocks core features.
REVENUE MODEL UNDEFINEDAWS COMMERCIAL SURFACE
Bedrock commit decrements stay in the customer’s account — no cross-account rebate path.
WHY THIS PATH EXISTS
Use Case 3 exists for regulated customers who cannot accept data egress from their VPC under any commercial construct. It is technically feasible but strategically subordinate — investment follows only once the primary architecture (UC1/UC2) is settled.
VS. USE CASE 2
Cursor revenue retention not yet modeled.
VS. USE CASE 1
Fundamentally different trust boundary.
Customer → Cursor (?)
Subscription path depends on commercial construct. Currently blocking.
Cursor → Customer Bedrock
Cross-account IAM exists today but is not production-grade.
⚠ Feature propagation
Tab, codebase indexing broken. Customers told to disable these features.
Customer-account decrement
All Bedrock consumption decrements customer’s own account directly.
GUIDING PRINCIPLE
Address economic-motivation customers (UC1 & UC2) first. BYOK for data-control and regulated customers is a fast follow — it carries more engineering effort and investment will be prioritized once the primary architecture is settled.