General-purpose Activate credits cover virtually every AWS-native service: EC2, ECS, EKS, Lambda, S3, RDS, Aurora, DynamoDB, CloudFront, Route 53, Bedrock, SageMaker, and roughly 200 more. They do not cover Marketplace SaaS purchases (Datadog, Snowflake, MongoDB Atlas, Confluent Cloud) or AWS Professional Services. Earmarked credits — Bedrock POC and Build for Startups — are tighter. This page is the comprehensive coverage reference: every service category, every gap, the Free Tier interaction, the Reserved Instance handling, and the tax/VAT mechanics.
The general-purpose Activate credit pools — Founders ($5K self-serve), partner-filed Founders ($5K–$25K), and Activate Portfolio ($50K–$100K) — are not service-restricted. They consume against virtually every AWS-native service on the monthly invoice. The coverage breakdown below is grouped by service category.
The mechanic is straightforward. When AWS finalizes your monthly invoice, the billing system itemizes consumption by service and region, then applies the available promotional credit balance against the eligible consumption lines before the final invoice is produced. General-purpose credits do not need to be directed; they auto-apply against every eligible line.
The list below covers the service categories that consume general-purpose Activate credits on a typical Series-A AWS invoice. It is not exhaustive — AWS publishes ~250 services and most consume against credits — but it covers the categories where the bulk of Activate credit consumption occurs.
Compute is the largest credit-consuming category for the median Series-A workload — typically 30–45% of the monthly invoice. Every standard compute service is covered: EC2 on-demand instances across all instance families (M-class general purpose, C-class compute-optimized, R-class memory-optimized, G-class GPU, P-class accelerated, Inferentia, Graviton), ECS task hours on both EC2 and Fargate launch types, EKS control plane hours plus the underlying node consumption, Lambda invocation costs and duration costs, App Runner service hours, Batch job hours, and Lightsail instance hours.
Spot instances also consume against credits at the spot price (not the on-demand price). Reserved Instance monthly fees consume against credits; Reserved Instance upfront payments generally do not (see Section VII on Reserved Instances and Savings Plans). Savings Plans monthly fees consume against credits with the same upfront caveat.
GPU-heavy compute — common for AI workloads pre-Bedrock — burns credits faster than CPU compute. A single p4d.24xlarge instance runs at ~$32/hour on-demand, meaning 24 hours of training consumes ~$770 of credits in a day. For startups projecting GPU-heavy training, the credit pool can exhaust in weeks rather than months. The Bedrock POC track is the structural alternative — Bedrock inference is itself a managed offering, and Bedrock POC credits ($10K–$50K) are earmarked specifically for that consumption.
Storage is the second-largest category for most workloads — typically 8–15% of the monthly invoice. Every AWS-native storage service is covered: S3 across all storage classes (Standard, Standard-IA, Intelligent-Tiering, Glacier Instant Retrieval, Glacier Flexible Retrieval, Glacier Deep Archive, S3 Express One Zone), EBS volume hours across all volume types (gp3, gp2, io1, io2, st1, sc1) plus snapshot costs, EFS standard and IA storage, FSx for Windows, Lustre, NetApp ONTAP, and OpenZFS, Glacier vault storage, and Storage Gateway charges.
S3 request charges (PUT, COPY, POST, LIST, GET) also consume against credits. Data transfer out from S3 to the internet is a significant credit burner for media or AI workloads — the first 10TB/month runs at ~$0.09/GB, meaning a CDN-misconfigured workload egressing 5TB/month consumes ~$450 of credits in egress alone.
S3 replication charges (cross-region, same-region) consume against credits. Object Lock and Object Versioning storage overages consume against credits at the standard per-GB rate for the underlying storage class.
Database consumption is typically 15–25% of the monthly Series-A invoice. Every managed database service is covered: RDS instance hours across MySQL, PostgreSQL, MariaDB, Oracle BYOL, and SQL Server engines, Aurora instance hours plus Aurora Storage (per GB-month), Aurora Serverless v2 ACU-hours, DynamoDB read/write capacity units plus DynamoDB Streams charges plus PITR storage, DocumentDB instance hours plus storage, Neptune instance hours plus storage, Timestream ingestion plus storage plus query, ElastiCache for Redis and Memcached node hours, MemoryDB for Redis node hours, Keyspaces request units, and QLDB read/write IOs.
Backup storage consumed by AWS Backup against any of the above database services consumes credits. Aurora Global Database cross-region replication consumes credits. Performance Insights at the standard 7-day retention is free; the long-term retention tier consumes credits.
RDS Reserved Instance monthly fees consume against credits. RDS Reserved Instance upfront and partial-upfront payments generally do not — they are paid out-of-pocket. Aurora Serverless v2 has no upfront payment mechanic; all consumption is metered and credit-eligible.
Networking is typically 6–12% of the monthly invoice, dominated by data transfer out and CloudFront. Every AWS-native networking service is covered: CloudFront request and data transfer charges across all regional rates, Route 53 hosted zone fees plus query charges plus DNSSEC, API Gateway REST API and HTTP API and WebSocket API request charges plus data transfer, VPC NAT Gateway hourly and data processing charges, VPC endpoint hourly charges, Direct Connect port hours and data transfer out, Transit Gateway attachment hours and data processing, AWS PrivateLink endpoint hours, Global Accelerator accelerator hours and data transfer.
Data transfer out to the internet (the largest networking line for most workloads) consumes credits at the standard regional rates — $0.09/GB for the first 10TB out of US-East, scaling down by tier. Inter-region data transfer also consumes credits. Intra-region data transfer between availability zones for cross-AZ workloads consumes credits at $0.01/GB each direction.
Domain registration through Route 53 — the annual $12/year fees and similar — is in a partial gray zone. AWS treats domain registration as a pass-through cost (paid to the registrar) rather than a service consumption, and these charges typically do not consume credits even though they appear on the AWS invoice. Hosted zone fees and query charges do consume credits normally.
Messaging and event streaming is typically 2–5% of the monthly invoice but consumes credits like any other category. Every messaging service is covered: SQS request charges (standard and FIFO), SNS publish and delivery charges, EventBridge event ingestion and rule evaluation, Kinesis Data Streams shard hours plus PUT payload units plus extended retention, Kinesis Firehose ingestion and conversion, Kinesis Video Streams ingestion and storage, MSK broker hours plus storage plus data transfer, Step Functions state transitions for standard and express workflows, Amazon MQ broker hours.
Express Step Functions workflows charge per request and per duration; both consume credits. Kinesis Data Analytics (Managed Service for Apache Flink) consumes credits per KPU-hour. SNS SMS delivery charges (per country rates) consume credits at the standard rates.
Analytics consumption is typically 4–10% of the invoice for B2B SaaS workloads with reporting features, and higher for analytics-focused products. Every analytics service is covered: Redshift cluster hours plus Concurrency Scaling plus Redshift Spectrum scan, Redshift Serverless RPU-hours, OpenSearch cluster hours plus storage plus UltraWarm, OpenSearch Serverless OCU-hours plus storage, Athena per-TB scan charges, Glue ETL job DPU-hours plus crawler runs plus Data Catalog requests, Lake Formation governed table costs, EMR cluster hours plus EMR on EKS, EMR Serverless worker hours, QuickSight reader sessions and author monthly fees, MWAA environment hours and worker hours, Athena for Apache Spark DPU-hours.
OpenSearch Serverless consumes against general-purpose credits and is also one of the supporting services covered by Bedrock POC credits when used for vector storage in a retrieval-augmented generation (RAG) workload.
AI/ML consumption varies enormously by workload — 1–3% of the invoice for SaaS workloads with light AI features, 30–60% for AI-first products. Every AI service is covered by general-purpose credits: Bedrock model invocation charges across Claude, Llama, Mistral, Titan, Cohere, and other available foundation models, SageMaker training instance hours plus inference endpoint hours plus notebook instance hours plus Studio user hours, Comprehend custom model training and inference, Rekognition image and video analysis charges, Textract per-page charges, Lex bot conversation charges, Polly characters synthesized, Translate characters processed, Transcribe audio minutes, and Personalize training plus inference.
Bedrock consumption against general-purpose Activate credits is allowed — meaning a startup with $100K Portfolio credits and a Bedrock-heavy workload can consume the full $100K against Bedrock if the workload runs that way. Bedrock POC credits are additive and earmarked (see Section II); they let a Bedrock-heavy workload extend the credit runway by stacking the two pools.
SageMaker training on GPU instances burns credits at the per-instance-hour rate for the underlying instance — meaning a ml.p4d.24xlarge training job consumes credits at ~$37.69/hour, and a 48-hour training run consumes ~$1,810. Inference endpoints are typically a smaller and more predictable line item.
Security, identity, and observability are typically 3–6% of the monthly invoice combined. Every native service in these categories is covered: Cognito monthly active users (above the always-free 50K MAU tier), KMS key requests plus customer master key monthly fees, Secrets Manager secret-month fees plus API calls, IAM Identity Center is free (no credit consumption needed), GuardDuty per-account per-region fees plus data analyzed, Security Hub finding ingestion and security checks, WAF web ACL fees plus rule evaluations plus requests, Shield Standard is free (Shield Advanced has its own structure — see note below), CloudWatch metric storage plus log ingestion plus log archival plus dashboard fees plus alarm fees, CloudWatch Logs Insights query scan charges, X-Ray trace recording and retrieval, AWS Config rule evaluations and config items recorded, CloudTrail data event ingestion (management events are free), Inspector vulnerability assessment per host or per Lambda function, Detective behavior analytics per GB ingested, Macie classification and discovery per GB.
Shield Advanced has a fixed $3K/month subscription fee plus data transfer protection charges. The subscription fee is technically a covered AWS service charge and consumes credits, but the subscription mechanic means a single $3K/month line item appears on the invoice for the subscription duration — material at the Series-A burn rate.
CloudWatch Logs ingestion is the most commonly underestimated observability line. A misconfigured application logging 100GB/month into CloudWatch Logs at $0.50/GB ingestion consumes $50 of credits per month for ingestion alone, plus storage. Aggressive log retention extending into the multi-year range can stack storage charges; the Logs Insights query scan is metered separately.
Every other native AWS service consumes against general-purpose credits at the standard service rate. This includes the IoT services (IoT Core message charges, IoT Greengrass device hours, IoT SiteWise data processing, IoT Analytics), the developer tooling (CodePipeline pipeline executions, CodeBuild build minutes, CodeDeploy deployments, ECR storage and data transfer, CodeArtifact storage and requests, Cloud9 instance hours), the frontend services (Amplify hosting and build minutes, AppSync API request and connection minutes, Device Farm device minutes), the integration services (AppFlow flow runs, MWAA, Step Functions), and the migration services (DataSync, DMS, Application Migration Service).
The general principle: if AWS bills the service on the standard AWS monthly invoice and it is not a Marketplace purchase, general-purpose Activate credits consume against it.
Some credit pools are earmarked — they only consume against a specific service or workload, even though they appear in the same AWS Billing dashboard view as general-purpose credits. Knowing which credits are earmarked and which are general-purpose matters for credit-stack planning.
Earmarked credits do not consume against general invoice consumption. If a Bedrock POC credit line of $25K is sitting in the account but the monthly invoice has no Bedrock charges, the Bedrock POC line does not burn down that month — even if there is $4K of EC2 + RDS + CloudFront on the invoice that could have consumed credits. The general-purpose Activate Portfolio line consumes against that $4K; the Bedrock POC line sits idle waiting for Bedrock charges to consume against.
The earmarking is enforced inside the AWS billing system at the service-mapping level. Each credit type has a defined set of eligible service codes; the billing system matches credit lines to consumption lines by service code before generating the final invoice.
Bedrock POC credits ($10K–$50K typical, Bedrock-earmarked) consume against Bedrock model invocations and a narrow set of tightly-coupled supporting services. The supporting services list is what makes Bedrock POC credits practically usable rather than just funding raw inference.
The covered services in a Bedrock POC pool are: Amazon Bedrock model invocations across all available foundation models (Claude, Llama, Mistral, Titan, Cohere, and others as added), Bedrock Knowledge Bases (the managed RAG service that bundles vector retrieval and prompt construction), Bedrock Agents (the orchestration service that calls tools and external APIs), OpenSearch Serverless when used as the vector store in a Bedrock RAG workflow, S3 when used for prompt logs, evaluation datasets, and Knowledge Base source documents, Lambda when used for Bedrock orchestration (the function that calls Bedrock and handles the response), and CloudWatch when used for Bedrock evaluation telemetry and inference monitoring.
Services not covered by Bedrock POC credits, even if they appear on the same invoice: EC2 unrelated to the Bedrock workload, RDS for the application database, CloudFront for the application frontend, Route 53 for the application DNS, Cognito for user authentication, and any other service that is part of the broader application rather than the Bedrock POC itself.
This is why the Bedrock POC credit is typically applied alongside a general-purpose Activate Portfolio pool, not instead of it. The Activate Portfolio covers the application infrastructure; the Bedrock POC extends the runway specifically on the Bedrock side.
Build for Startups credits (up to $25K, additive on top of Portfolio) are earmarked to the specific workload described in the application. The application form asks the partner to describe a discrete project — a video transcoding pipeline, a compliance push, a new vertical launch, a SOC 2 telemetry buildout — and the credits consume against the AWS services that workload uses.
In practice, the earmarking is loose. The billing system does not enforce Build for Startups credits against an isolated workload at the service-code level; it consumes them in expiration-date order alongside general-purpose credits. The earmarking is enforced at application-review time — meaning the reviewer approves the Build for Startups award contingent on a real additive workload existing, and double-counting the same workload between Portfolio and Build for Startups is the most common rejection cause for the secondary track.
Once approved and posted to the account, Build for Startups credits behave like general-purpose credits in terms of which services they consume against. The earmarking is a pre-issuance check, not a post-issuance enforcement.
Migration Acceleration Program (MAP) credits are earmarked to migration-related consumption. The covered services include the compute the migrated workload runs on (EC2, ECS, EKS, Lambda for application tiers), the storage and database services the migrated workload uses (S3, RDS, Aurora, DynamoDB), and the migration tooling itself (Application Migration Service, DMS, DataSync, Server Migration Service).
The earmarking is enforced by the linked-account structure of the MAP engagement. The migration workload typically lives in a dedicated AWS linked account (or a tagged set of resources within an existing account), and the MAP credits apply against consumption in that account/tag scope. Consumption outside the MAP scope — a separate development account, a marketing analytics workload, an internal tools account — does not consume MAP credits even if it is on the same AWS Organizations billing structure.
For Series-A migrations the MAP credit pool ranges from $25K at Assess phase to $50K–$200K at Migrate phase, depending on the scope of the migration. The credit pool is functionally limited to the migration workload, but the migration workload is typically the largest consumer of AWS services in the account during the engagement, so the practical effect is wide coverage.
There are five distinct categories of charges that appear on or alongside an AWS bill and are not eligible for credit consumption. Knowing the gap list is as important as knowing the coverage list — the gaps are where unexpected out-of-pocket spend appears.
The gaps fall into five categories: Marketplace SaaS purchases, AWS Professional Services, Reserved Instance and Savings Plans upfront payments, AWS Support plan fees, and a small set of regional pass-through charges. The Marketplace SaaS gap is the largest and most material for the median Series-A startup.
AWS Marketplace is a catalog of third-party software products that can be billed through the AWS invoice. Common Marketplace SaaS purchases include Datadog (billed via Marketplace for many enterprise customers), Snowflake compute and storage, MongoDB Atlas, Confluent Cloud, GitLab Ultimate licensing, Tableau, JFrog Artifactory and Xray, HashiCorp Vault and Terraform Cloud, PagerDuty, and many SaaS observability and security tools.
AWS credits — Activate, Founders, Build for Startups, Bedrock POC, MAP, Portfolio — do not consume against Marketplace SaaS charges. The Marketplace line items appear on the monthly AWS invoice and are paid in cash even when the credit balance would otherwise cover the AWS-native portion of the bill.
The economic effect is material. For a median Series-A B2B SaaS startup with a $4,000/month AWS bill where $600 is Datadog via Marketplace and $400 is MongoDB Atlas via Marketplace, the AWS-native portion ($3,000) consumes against credits and the Marketplace portion ($1,000) is paid in cash. The credit pool effectively funds 75% of the bill rather than 100%.
The same vendor billed directly (not through Marketplace) is paid in cash regardless — the Datadog bill paid via Datadog's own invoice is not on the AWS invoice at all and is unaffected by the credit structure. The Marketplace path makes the cash payment visible on the AWS invoice; the direct path keeps it separate.
For startups that have not yet committed to Marketplace billing, the credit-coverage consideration is one input into the decision: billing through Marketplace can simplify vendor management and provide some procurement consolidation, but it does not unlock credit coverage for those vendors. The math depends on the vendor — some Marketplace SaaS contracts include AWS-spend commitments that offset the credit non-coverage, and those need to be modeled individually.
AWS Professional Services (the in-house AWS consulting org) bills directly for architecture engagements, migration consulting, and specialized advisory work. ProServe engagements are typically priced at $250–$400/hour for senior consultants, with engagements running from $25K (small architecture review) to $500K+ (large migration leads).
ProServe charges appear on the AWS invoice as a separate line and do not consume against any credit type — Activate, Portfolio, Build for Startups, Bedrock POC, MAP. The credit pool may show ample remaining balance but the ProServe line is paid in cash.
For Series-A startups, ProServe engagement is unusual; the partner-led alternative covers the same work and is typically funded by AWS partner engagement programs (separate from your credits). For larger workloads or post-EDP customers, the ProServe gap can matter.
Reserved Instances and Savings Plans are commitment-based pricing mechanisms that trade upfront payment for discounted hourly rates. The payment structures are: all-upfront (full payment day one, no monthly charge), partial-upfront (some upfront plus reduced monthly), and no-upfront (full monthly, slightly less discount).
The upfront payment portion of Reserved Instances and Savings Plans generally does not consume against credits. The all-upfront payment is debited from the linked AWS payment method (credit card or invoice payment) at purchase. The partial-upfront payment behaves the same way for the upfront portion.
The monthly recurring charge portion does consume against credits. A partial-upfront RI with a $2K/month recurring charge will consume $2K of credits per month for the recurring portion, even though the original upfront payment was paid in cash.
Savings Plans behave the same way — monthly commitment fees consume against credits at the discounted rate. The strategic implication: Series-A startups with $100K of credits should generally use the no-upfront Savings Plans variant during the credit-funded period so the entire Savings Plan commitment consumes credits. Switching to partial-upfront or all-upfront makes sense after credits exhaust or as part of an EDP-era commitment structure.
AWS Support plans (Developer, Business, Enterprise On-Ramp, Enterprise) are priced as a percentage of monthly AWS spend with floors and tiers. Developer Support is $29/month or 3% of monthly spend, whichever is greater. Business Support is $100/month or 10% of monthly spend (with declining tiers above $10K, $80K, $250K). Enterprise On-Ramp is $5,500/month or 10% of spend. Enterprise is $15,000/month or 10% of spend (with declining tiers).
AWS Support plan fees generally do not consume against credits — they appear on the invoice as a separate line and are paid in cash. The Series-A pattern is to run Developer or Business Support during the credit-funded period and re-evaluate the support tier when transitioning to EDP.
The Business Support 10% calculation applies to pre-credit AWS spend in some contracts and post-credit spend in others — the exact calculation depends on the support contract terms. CloudRoute partners can confirm the calculation basis for a given account during the credit application process.
Certain regional and country-specific charges flow through the AWS invoice as pass-through to local providers and do not consume against credits. The most common categories are: domain registration fees (Route 53 registers domains through accredited registrars and the registration fee passes through), some telecommunications charges for SNS SMS in certain countries, certain regulatory or compliance certification fees in specific industries, and AWS Activate Console-related certification voucher fees.
These categories are small enough that they are not a material gap for the median Series-A startup — typically less than 1% of the monthly invoice combined. They are worth knowing about for invoice reconciliation rather than for credit-pool planning.
AWS Free Tier is a separate consumption discount that applies to the monthly invoice before promotional credits. Understanding the ordering matters for accurate credit-runway forecasting.
AWS Free Tier consists of three components: the always-free tier (services that have a perpetual free allocation, like Lambda's 1M free requests/month, DynamoDB's 25GB free storage, and SNS's 1M free publishes/month), the 12-month free tier (allocations that apply only during the first 12 months of an account, like 750 hours/month of t2.micro or t3.micro EC2, 5GB of S3 standard, and 750 hours/month of single-AZ RDS), and the $200 in promotional credits AWS introduced for new accounts in 2025 (covering the first $200 of consumption across covered services within the first 6 months of account creation).
When the AWS billing system finalizes the monthly invoice, the order of operations is: (1) calculate gross consumption by service, (2) subtract Free Tier allocations (always-free and 12-month, where applicable), (3) apply the new-account $200 promotional credit (during the first 6 months of the account), (4) apply general-purpose Activate credits (Founders, Portfolio, partner-filed) against the remaining consumption, (5) apply earmarked credits (Bedrock POC, Build for Startups) against their respective eligible consumption, (6) generate the final invoice.
The practical effect for a new AWS account: the first $200 of consumption is fully absorbed by the new-account credit, the 12-month Free Tier allocations cover the always-free band of small workloads (a single t3.micro running 24/7 is fully covered by the 750-hours free tier for the first 12 months), and Activate credits begin consuming only when consumption exceeds those allocations.
For a Series-A startup migrating an existing workload to AWS, the Free Tier interaction is usually small — the workload consumption is well above the 12-month Free Tier allocations from day one, so the Free Tier component is a rounding-error fraction of the bill. For pre-seed founders just starting on AWS, the Free Tier can cover the entire workload for the first few months, with Activate credits not consuming at all until consumption ramps.
The new-account $200 credit and the Free Tier do not extend the validity window of separately-issued Activate credits. They reduce the consumption that the Activate credits would otherwise apply against, which slightly extends the burn-rate calculation but does not change the expiration date of the Activate credit lines themselves.
The customer cannot direct which credits apply to which charges. AWS auto-applies credits in expiration-date order, with earmarked credits consuming only against their eligible services. The mechanics are deterministic but invisible — they happen inside the billing system between invoice generation and final billing.
When the monthly invoice is finalized, the AWS billing system performs the following matching: it iterates over the consumption lines on the invoice (each line is a service code, a region, a usage type, and a dollar amount), then iterates over the available credit lines in expiration-date-ascending order, and applies each credit line to the eligible consumption lines until either the credit line is exhausted or the consumption lines are exhausted.
Within a single credit line, the application is service-eligibility-aware. A Bedrock POC line of $25K with a Bedrock invoice line of $3K consumes $3K of the line; the remaining $22K of the Bedrock POC balance is not eligible for the $1K EC2 consumption line on the same invoice and does not consume against it.
General-purpose credit lines (Founders, Portfolio, partner-filed Founders) are eligible against virtually every service code on the invoice except the gap categories described in Section III. They consume against the eligible consumption lines in invoice order.
The customer cannot manually direct which credit consumes against which charge. There is no console toggle, no API call, and no support ticket that changes the credit-application order. The expiration-date-ascending order is the design and is enforced consistently across all accounts.
The one nuance: for credit lines with identical expiration dates (a Portfolio award and a Build for Startups award issued on the same day, both expiring exactly 12 months later), the application order within the matched date is FIFO by issuance timestamp. This rarely matters in practice — credit lines with identical expiration dates are usually portions of the same award.
Reserved Instances and Savings Plans add commitment-based discounts that interact with the credit pool in a specific way. The hourly recurring portion consumes credits at the discounted rate; the upfront portion generally does not.
Reserved Instances (RIs) commit to a specific instance type, region, and tenancy for 1 or 3 years in exchange for a discount of 20–72% off on-demand pricing depending on the term and payment structure. Savings Plans (SPs) are a more flexible variant that commits to a dollar-per-hour spend level rather than a specific instance type, with similar discount magnitudes.
The all-upfront payment structure pays the entire 1-year or 3-year commitment day one. The all-upfront line item appears as a one-time charge on the next monthly AWS invoice. This one-time charge is paid via the linked payment method and does not consume against credits in the standard mechanic.
The partial-upfront payment structure pays a smaller upfront amount plus a reduced monthly recurring charge. The upfront portion behaves the same as the all-upfront case (paid in cash). The monthly recurring portion consumes against credits at the discounted rate.
The no-upfront payment structure pays only the monthly recurring charge with no upfront. The entire commitment payment is monthly, and the entire monthly payment consumes against credits.
The strategic implication for credit-funded Series-A workloads: the no-upfront variant of Savings Plans is generally the right choice during the credit-funded period because every dollar of the commitment consumes against credits. The all-upfront variant requires paying the full commitment in cash even though the underlying consumption would have been credit-eligible — meaning the cash outlay is real even though the credit balance is intact.
After credits exhaust, the optimal payment structure shifts. The all-upfront variant offers the largest discount and may be the right choice in steady-state, but the credit-funded period favors no-upfront. CloudRoute partners modeling the credit-runway during the application typically include a Savings Plans transition plan that switches from no-upfront during the credit-funded period to a longer-term commitment structure when credits exhaust.
AWS credits apply to the pre-tax invoice amount in most jurisdictions. The post-credit invoice is then taxed at the applicable regional rate. Several jurisdictions have credit-related tax nuances worth knowing.
In the United States, AWS bills the pre-tax usage and applies state-level sales tax or use tax on top depending on the state of the linked AWS billing address. Credits consume against the pre-tax usage amount. A $4,000 monthly invoice in a state with 8% sales tax becomes a $4,000 + $320 = $4,320 final bill before credits; if $3,500 of credits apply against the $4,000 pre-tax amount, the final invoice is $500 of usage + $40 of tax on that $500 = $540.
In the European Union, VAT applies on top of the AWS service charge at the customer's country VAT rate (typically 19–27%). Credits consume against the pre-VAT amount in most EU jurisdictions. The post-credit invoice is VAT-rated at the customer's country rate. A €4,000 monthly invoice with 21% VAT and €3,500 of applied credits becomes €500 of usage + €105 of VAT = €605 final.
In India, AWS uses the AISPL (Amazon Internet Services Private Limited) entity for billing Indian customers. The AISPL invoice structure has specific GST handling that interacts with credits differently from the AWS global invoice. Credits consume against the pre-GST AISPL invoice amount, but the GST input credit treatment depends on the customer's GST registration status and business activity classification. Indian Series-A startups receiving Activate credits should validate the GST handling with their accountant during the credit application process.
In the United Kingdom post-Brexit, the AWS billing flow uses the AWS UK entity. UK VAT (20%) applies on top of the post-credit usage amount. The UK invoicing behaves like the EU invoicing structurally — credits offset pre-VAT, VAT applies to remaining usage.
In the UAE and Saudi Arabia, VAT (5% in UAE, 15% in KSA as of 2026) applies on top of post-credit usage. The GCC VAT structure treats AWS as a B2B digital service and applies the reverse-charge mechanism for registered businesses, which has specific reporting implications but does not change the credit-pre-VAT mechanic.
The general principle across jurisdictions: credits reduce the pre-tax (pre-VAT, pre-GST, pre-sales-tax) invoice amount. The applicable tax then applies to whatever portion of the invoice the credits did not cover. A startup with a $4K monthly invoice and $4K of credit consumption pays $0 of usage + $0 of tax. A startup with a $4K monthly invoice and only $3K of credit consumption (because $1K was Marketplace SaaS) pays $1K of usage + tax on $1K.
A representative monthly AWS bill for a Series-A B2B SaaS startup with $100K of Portfolio credits in the account, broken down line by line to show what consumes credits and what does not.
The startup runs a multi-tenant B2B SaaS application on AWS. 12 engineers. ~3,000 paying customers. Monthly active users ~45,000. The workload consists of an application tier on ECS Fargate, an Aurora PostgreSQL database, S3 for asset storage, CloudFront for delivery, Cognito for authentication, Datadog (billed via Marketplace) for observability, and MongoDB Atlas (billed via Marketplace) for a specific event-store workload. Total monthly AWS invoice: $4,120.
The pre-credit invoice breakdown: ECS Fargate compute = $980, Aurora PostgreSQL instance + storage + backup = $720, S3 storage + requests + data transfer = $310, CloudFront requests + data transfer out = $420, Route 53 hosted zone + queries = $35, API Gateway requests + data transfer = $180, Cognito MAU above the always-free tier = $90, KMS keys + requests = $25, Secrets Manager secret-months = $30, CloudWatch logs + metrics + dashboards = $260, GuardDuty + Security Hub = $120, AWS Backup storage = $80, Step Functions + EventBridge + SQS = $70, Lambda invocations + duration = $120, Datadog via Marketplace = $580, MongoDB Atlas via Marketplace = $260, AWS Business Support plan = $400 (the 10% tier of the AWS-native portion = $322 plus the $100 minimum = $400).
The total invoice is $4,700 gross. Of that, $4,000 is AWS-native consumption (eligible for general-purpose credits), $840 is Marketplace SaaS (not credit-eligible), and the $400 Support plan fee is treated as a separate line not eligible for credit consumption. Wait — let us recalculate cleanly. The $4,120 figure was rough; let us redo with the precise lines.
Recalculated: the AWS-native consumption (ECS, Aurora, S3, CloudFront, Route 53, API Gateway, Cognito, KMS, Secrets Manager, CloudWatch, GuardDuty, Backup, Step Functions, Lambda) totals approximately $3,440. The Marketplace SaaS portion (Datadog + MongoDB Atlas) totals $840. The Business Support fee is $400. Gross invoice: $4,680.
Of this gross invoice, $3,440 consumes against the Portfolio credit balance, leaving the AWS-native portion at $0 out-of-pocket. The Marketplace SaaS portion ($840) is paid in cash. The Support plan fee ($400) is paid in cash. Cash out-of-pocket per month: $1,240 plus applicable state sales tax on the $1,240 paid portion. The headline number — "the credit pool covers the AWS bill" — is true for the AWS-native portion (73% of the gross invoice) and misleading for the full bill (27% of the gross invoice is paid in cash).
Over the 20-month credit-burn window typical at this consumption rate, the cumulative numbers are: $68,800 of Portfolio credit consumption against AWS-native services, $16,800 of cash payment for Marketplace SaaS, $8,000 of cash payment for Support plan fees. The Portfolio credit balance is exhausted at month 20 (with $3,440/month consumption against a $100K pool — actually exhausted at month 29 — adjusting: $3,440/month against $100K = 29 months, but the validity window is 24 months, so the credit consumption hits 24 months of $3,440 = $82,560 against the $100K pool, leaving $17,440 forfeited at expiration unless burn ramps).
The example illustrates the importance of modeling Marketplace SaaS exposure during the credit application. A startup whose monthly burn projection includes 25% Marketplace SaaS is structurally going to leave material credit balance unused at expiration unless the AWS-native portion scales materially within the validity window. The CloudRoute-routed credit application process explicitly models this exposure during the partner pre-call.
For the worked example, 18% of the gross AWS invoice is Marketplace SaaS — $840 of $4,680/month. Over 24 months that compounds to $20,160 of cash payment that the credit pool does not cover. The strategic implication is to consider direct billing with these vendors (outside Marketplace) when the credit-pool offset would otherwise be material — though direct billing has its own trade-offs around vendor consolidation and procurement workflow.
The coverage map for the five most common credit types, by service category. Use this as the canonical reference when planning credit-stack consumption.
| Service category | General-purpose Activate | Bedrock POC | Build for Startups | MAP credits |
|---|---|---|---|---|
| Compute (EC2, ECS, EKS, Lambda) | Yes | Only Lambda for Bedrock orchestration | Yes (within scoped workload) | Yes (within migration scope) |
| Storage (S3, EBS, EFS) | Yes | Only S3 for prompt logs / Knowledge Base | Yes | Yes (migration storage) |
| Databases (RDS, Aurora, DynamoDB) | Yes | No | Yes | Yes (migrated database) |
| Networking (CloudFront, Route 53, API Gateway) | Yes | No | Yes | Yes |
| AI/ML — Bedrock | Yes | Yes (primary) | Yes (if workload scope) | No |
| AI/ML — SageMaker | Yes | No (not Bedrock) | Yes (if workload scope) | No |
| OpenSearch Serverless (vector store) | Yes | Yes (RAG support) | Yes | No |
| Analytics (Redshift, Glue, Athena) | Yes | No | Yes | Yes (analytics in scope) |
| Security / IAM / CloudWatch | Yes | Only Bedrock evaluation telemetry | Yes | Yes |
| Marketplace SaaS (Datadog, Snowflake) | No | No | No | No |
| AWS Professional Services | No | No | No | No |
| Reserved Instance upfront | Generally no | No | No | No |
| RI / SP monthly recurring | Yes | No | Yes | Yes |
| AWS Support plan fees | No | No | No | No |
Situation: Series-A startup approved for $100K Portfolio credits in mid-2024. The team modeled the credit runway as 22 months against a projected $4,500/month AWS burn. The projection treated the gross invoice as credit-eligible; the actual breakdown was $3,440/month AWS-native + $840/month Marketplace SaaS (Datadog + MongoDB Atlas) + $400/month Business Support plan fee.
What CloudRoute did: The credit consumption ran at $3,440/month against the AWS-native portion of the invoice. The Marketplace SaaS and Support plan fees ran at $1,240/month cash out-of-pocket. The credit pool extended further than projected against the AWS-native portion alone (29 months at $3,440/month vs the 24-month validity window), creating a forfeit-at-expiration risk on the $17K of remaining balance.
Outcome: The CloudRoute partner reworked the credit consumption plan at month 9: the team migrated MongoDB Atlas to a direct Atlas contract (out of Marketplace) to free up Marketplace spend for other purposes, and consolidated Datadog to a direct Datadog contract negotiated alongside the Atlas move. The Marketplace exposure dropped to $0; the AWS-native portion grew slightly with self-hosted observability. The Portfolio credit consumed at $4,100/month and exhausted at month 24, matching the validity window. Cost to customer for the partner re-engagement: $0.
engagement window: 24 months · marketplace exposure mitigated: $840/month · forfeit at expiration: $0 · cost to customer: $0
CloudRoute routes you to an AWS partner who models the credit coverage against your actual service mix — including Marketplace SaaS exposure — before filing the application. Customer pays $0. No procurement loop.