# Atlas Cost Commitments, Cross-Org Billing and FinOps Tooling > Source-anchored research for Atlas Cost Commitments, Cross-Org Billing and FinOps Tooling. Run status: COMPLETED. Parent: [MongoDB Atlas Cost Optimization](https://llms-explorer.com/tree/mongodb-atlas-cost-optimization/) · 7 facets · 56 facts · page: https://llms-explorer.com/tree/atlas-cost-commitments-cross-org-billing-and-finops-tooling/ ## Definitions - ZeroEntropy (a competing embedding vendor) argues MRL is not lossless: on its internal 22-dataset harness a learned post-hoc projection beat the best MRL configuration by about 0.02-0.03 NDCG@10 from 2560 down to 80 dimensions, and it says the original MRL paper's equal-accuracy claim was shown on ImageNet, not text retrieval. The evaluation uses LLM-judged relevance and is unreplicated. (low confidence) (verified as of 2026-04-21) — [source](https://zeroentropy.dev/articles/matryoshka-is-dead/) ## Structure and components - MongoDB docs strongly recommend proactively monitoring the list of non-compliant resources via the nonCompliantResources endpoint. Policies enforce end-state goals, not intermediate states, so a failing downstream component can temporarily leave a policy non-compliant. (high confidence) — [source](https://mongodb.com/docs/atlas/atlas-resource-policies) ## How it works - Atlas has native billing alerts in USD: the last daily amount billed (excluding credits applied) per org or project, and monthly totals per organization or per project that reset when the pending invoice closes. These work as budget thresholds. (high confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/reference/alert-conditions/) - Vantage ingests Atlas costs through the Invoices API, not FOCUS, using an org API key with Organization Billing Viewer and Organization Read Only. Atlas credits do not appear in Vantage because the provider API does not expose credit information; per-cluster Active Resources need Project Read Only on each project. (low confidence) (verified as of 2026-10-19) — [source](https://docs.vantage.sh/connecting_mongodb-atlas) - Atlas automatic quantization cuts index RAM to about 1/3.75 of float32 for scalar (int8) and 1/24 for binary (not 1/32, because the HNSW graph is not compressed); binary requires numDimensions to be a multiple of 8. MongoDB recommends quantization for large collections (over 100,000 vectors). Because binary needs a multiple of 8, truncated dimension choices must respect that. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/vector-search/indexes/vector-search-type.md) - Autoscaling drift: if a cluster's existing autoscaling limits exceed a new policy's bounds, the cluster keeps scaling automatically beyond them, but manual scaling past the policy limits is refused. Docs say to use nonCompliantResources to find and fix such clusters, which matters for cost because compute autoscaling ceilings can stay above the cap. (medium confidence) — [source](https://mongodb.com/docs/atlas/atlas-resource-policies) ## Measurements and reference values - Google Cloud documents a 100% drawdown rate: spend on Google Cloud Marketplace services counts toward a customer's Minimum Commitment unless excluded. Documented exclusions are Google Maps offerings, Google Ads, services not running on Google Cloud, resold services where the vendor has not opted into the resale program, services owned by the buyer/reseller/affiliate, and professional services. Google may change the list at its sole discretion. (high confidence) (verified as of 2026-10-13) — [source](https://docs.cloud.google.com/marketplace/docs/commit-drawdown) - Since 2025-06-09, Google Cloud Marketplace Channel Private Offers (purchases through a Google-authorized channel partner) draw down at 100% of the private offer's final price, but the amount applied to the customer's minimum commitment is capped at 25%. (high confidence) (verified as of 2025-05-15) — [source](https://cloud.google.com/blog/topics/partners/upgrades-to-google-cloud-marketplace-for-partners) - Truncation and quantization are multiplicative, complementary levers: Voyage docs state 8-bit gives 4x and 1-bit gives 32x smaller vectors, so a 2048-dimension float32 vector (8 KB) truncated to 512 dimensions and binary-quantized is 64 bytes raw. Atlas's measured RAM ratio for binary is 1/24, so expect less than the raw arithmetic. (medium confidence) — [source](https://www.mongodb.com/docs/voyageai/tutorials/dimensions-and-quantization.md) - A May 2026 study (arXiv 2605.16608) reports that truncated embeddings of ordinary non-MRL models are competitive with, and often beat, MRL-trained models up to roughly 70-80% size reduction; MRL helps mainly at heavier truncation (about 80%+), mostly for smaller models. (low confidence) (verified as of 2026-05-01) — [source](https://www.emergentmind.com/papers/2605.16608) ## Problems, failure modes and limitations - Microsoft MACC eligibility is per offer: each Marketplace offer must be enrolled (Partner Center shows Enrolled/Not Enrolled), free and BYOL offers are not eligible, and only licenses used exclusively in Azure contribute. Atlas clusters deployed on non-Azure clouds are therefore a risk for MACC credit; the Atlas page does not address this. (medium confidence) (verified as of 2026-10-01) — [source](https://learn.microsoft.com/en-us/partner-center/marketplace-offers/azure-consumption-commitment-enrollment) - Atlas GCP Marketplace billing (listing 'MongoDB Atlas (Pay as You Go)') has no upfront commitment. An admin picks a Google Cloud Billing Account at subscription, then links an Atlas organization; outstanding or failed invoices must be resolved first. While linked, the payment method shows 'GCP Marketplace Subscription' and cannot be edited; removing it requires unlinking the billing account. Cancelling is done in GCP under Your Orders. (medium confidence) — [source](https://mongodb.com/docs/atlas/billing/gcp-self-serve-marketplace/) - Paying and linked organizations cannot have failed payments, an active self-serve support plan, an active self-serve marketplace subscription, or overlapping monthly commitment subscriptions. (high confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - A paying organization can link at most 250 organizations (UI or Admin API); more requires contacting support. A paying organization cannot itself be a linked organization. The subscription enabling the feature is bought through MongoDB Sales. (medium confidence) (verified as of 2026-10-19) — [source](https://www.mongodb.com/docs/atlas/billing) - Atlas Vector Search fixes numDimensions in the index definition. For Voyage autoEmbed fields it accepts 256, 512, 1024 or 2048 (default 1024), and the autoEmbed fields quantization and numDimensions cannot be edited after creation; changing them means building a new index and deleting the old one. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/vector-search/indexes/vector-search-type.md) - In the mongodbatlas provider, tags on mongodbatlas_advanced_cluster are written with a map-like literal (tags = { Key = "Value" }) in the registry example but are 'effectively implemented as a set of objects' per a maintainer, so lifecycle ignore_changes cannot target a single tag key such as tags["costcenter"]. The issue was closed March 14, 2024 with no schema change visible. (medium confidence) (verified as of 2026-10-19) — [source](https://github.com/mongodb/terraform-provider-mongodbatlas/issues/2006) - A plan-time tag policy for Atlas would check mongodbatlas_advanced_cluster resources in the plan for required tag keys (e.g. environment, application, cost-center) and fail on missing keys. No published Atlas-specific Sentinel/OPA example was found; HashiCorp's own tag examples target AWS and Azure resources. (low confidence) — [source](https://developer.hashicorp.com/terraform/cloud-docs/policy-enforcement) ## Comparisons and alternatives - FOCUS data is invoice-scoped: it is retrieved per invoice (current or past billing period), so an ingestion pipeline polls invoices by month rather than streaming real-time spend. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/billing/focus-api/) - Atlas also has native percentage-increase spend alerts: org daily vs previous day, org weekly vs previous 7 days, org monthly vs previous 30 days, and project daily and weekly. This is a simple built-in anomaly-style detector. (medium confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/reference/alert-conditions/) ## Facts and statements - Atlas on Azure Marketplace (pay-as-you-go listing) is advertised as drawing down existing Azure committed spend (MACC and Azure commit-to-consume), with one monthly Azure invoice including Atlas usage and upgraded support billed through Azure. MongoDB's page gives no rates or eligibility rules. (medium confidence) (verified as of 2025-03-18) — [source](https://www.mongodb.com/company/blog/news/introducing-pay-as-you-go-atlas-azure-marketplace) - Google excludes Marketplace services not running on Google Cloud from drawdown, and Google defers to the vendor on which services run on Google Cloud. For Atlas, clusters on AWS or Azure billed through the GCP Marketplace listing may therefore not count; this is an inference, and no Atlas or Google source states it for Atlas. (low confidence) (verified as of 2026-10-13) — [source](https://docs.cloud.google.com/marketplace/docs/commit-drawdown) - Atlas docs list Marketplace as a distinct subscription type: invoices for usage come from the marketplace the customer subscribed through, not from MongoDB. The self-serve GCP page does not cover commitment drawdown, private offers, or changing the linked billing account. (low confidence) (verified as of 2026-10-19) — [source](https://mongodb.com/docs/atlas/billing/gcp-self-serve-marketplace/) - Cross-organization billing shares one billing subscription across many organizations; the paying organization's invoice lists charges incurred by all linked organizations. (high confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - Linked organizations are bound by the paying organization's subscription agreement: support level, subscription charges, discounts and payment terms. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - Each linked organization gets its own monthly invoice with its usage plus a percentage of the total subscription cost proportional to its usage; payment still flows only through the paying organization. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - Daily subscription uplift is billed to each organization on its own usage. At period end an unmet monthly minimum is split across linked and paying organizations by usage share (docs example: $199 unmet split $132.70/$66.30 for 2/3 and 1/3 of charges). (medium confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - Linking prerequisites: paying org has an Atlas subscription; Organization Billing Admin or Organization Owner on both orgs; linked org has the same minimums, uplifts and SLA as the paying org or no support plan (Atlas Basic). (high confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - Documented use case: keep separate organizations per team for authorization isolation while paying through one organization; a 250+ project company can create new organizations and link them to avoid project limits and keep one invoice source. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/billing) - Charts billing ingestion: for cross-org billing, enter the paying organization's ID and toggle ingestion of linked invoices. The API key pair needs Organization Billing Admin and Billing Viewer roles. (medium confidence) — [source](https://www.mongodb.com/docs/charts/dashboards/manage-dashboards/billing-dashboards/) - Enabling linked invoices on an existing Charts ingestion limits the dashboard to the last three months of each linked org's data; a new ingestion gets twelve months. Invoice updates are picked up only for invoices within three months; ingestion runs daily. (medium confidence) (verified as of 2026-10-19) — [source](https://www.mongodb.com/docs/charts/dashboards/manage-dashboards/billing-dashboards/) - Querying a linked invoice through the Admin API requires Organization Billing Admin or Organization Owner, while the same endpoint's role table lists only Organization Billing Viewer, so the page is internally inconsistent about the role needed. (low confidence) — [source](https://www.mongodb.com/docs/api/doc/atlas-admin-api-v2/2025-02-19/operation/operation-getorginvoice) - Third-party FinOps tooling such as DoiT connects each Atlas organization separately (one service account per org) and maps organization.id to a billing-account dimension; the page does not describe paying/linked org handling. (low confidence) (verified as of 2026-10-19) — [source](https://www.doit.com/platform/mongodb-intelligence/) - Atlas provides billing cost and usage data in FOCUS 1.3 format. Users get it from the Invoices page (choose 'CSV FOCUS Cost & Usage Report') or from the FOCUS Data API; MongoDB announced GA on 2026-08-19. (high confidence) (verified as of 2026-08-19) — [source](https://www.mongodb.com/docs/atlas/billing/focus-report/) - FOCUS removes the manual conversion of Atlas Credit Units into billing currency: BilledCost and EffectiveCost are returned in billing currency, and PricingCurrency/x_PricingCurrencyRate expose the Atlas Credit rate. (high confidence) — [source](https://www.mongodb.com/docs/atlas/billing/focus-report/) - The ContractApplied field identifies which Atlas credit covered usage (Prepaid, Promotional or Elastic); InvoiceIssuer distinguishes direct MongoDB billing from marketplace billing; Tags carries resource tags. These give commitment and allocation analysis without custom SKU mapping. (high confidence) — [source](https://www.mongodb.com/docs/atlas/billing/focus-report/) - The FOCUS Data API adds three endpoints to the Atlas Administration API. The first, List Invoices for an Organization (GET /api/atlas/v2/orgs/{orgId}/associatedInvoices), takes month and year filters and can include linked organizations for cross-org billing. All endpoints need Organization Billing Admin, Organization Owner or Organization Billing Viewer. (high confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/billing/focus-api/) - Atlas sends billing alerts to Organization Owners and Billing Admins by default; a billing email address can be added or used exclusively. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/billing/) - DoiT Cloud Intelligence added a MongoDB Intelligence module (announced 2025-11-07) that pulls Atlas cost and usage data automatically from a connected organization. Its changelog describes anomaly investigation through the Ava assistant, not a dedicated Atlas anomaly feature, and does not mention budgets. (low confidence) (verified as of 2025-11-07) — [source](https://www.doit.com/changelog/mongodb-intelligence-finops-visibility-for-atlas-built-into-doit-cloud-intellige) - FOCUS v1.4 lists anomaly detection, forecasting and exchange-rate modeling only as capabilities future versions 'will consider'. FOCUS standardises the data shape, so anomaly detection and budgets stay tool-specific. (medium confidence) (verified as of 2026-10-19) — [source](https://focus.finops.org/docs/specification/v1-4/sections/introduction/) - FOCUS SaaS support is described by the spec as 'foundational' (columns for pricing currency, effective cost and contracted pricing in non-monetary units such as credits), so cross-vendor comparisons of Atlas data may still need vendor-specific handling. (medium confidence) — [source](https://focus.finops.org/docs/specification/v1-4/sections/introduction/) - Voyage AI documents that vector storage and compute cost scales linearly with embedding dimensionality and with bits per value; its Matryoshka embeddings let a 2048-dimension vector be truncated to its first 256, 512 or 1024 dimensions, each slightly lower in retrieval quality, and the docs show truncation followed by re-normalization. (high confidence) — [source](https://www.mongodb.com/docs/voyageai/tutorials/dimensions-and-quantization.md) - OpenAI states the text-embedding-3 models can be shortened with the dimensions parameter (dropping trailing values) and that text-embedding-3-large cut to 256 dimensions still beats the full 1536-dimension text-embedding-ada-002 on MTEB; this is an OpenAI-reported figure, not independently verified here. (medium confidence) — [source](https://openai.com/index/new-embedding-models-and-api-updates/) - Truncate only models documented as Matryoshka-trained (Voyage 4 series, OpenAI text-embedding-3) and re-normalize after slicing; validate recall on your own queries at the target dimension before re-indexing, since quality retention depends on model, corpus and query mix. (medium confidence) — [source](https://www.mongodb.com/docs/voyageai/tutorials/dimensions-and-quantization.md) - The endpoint is GET /api/atlas/v2/orgs/{orgId}/nonCompliantResources with Accept header application/vnd.atlas.2024-08-05+json. It returns the projects, clusters or networks in an organization that violate current resource policies. The Atlas CLI exposes it 1:1 as `atlas api resourcePolicies getNonCompliantResources --orgId --version 2024-08-05`. (high confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/cli/current/command/atlas-api-resourcePolicies-getNonCompliantResources/) - Atlas does not auto-remediate or block existing out-of-compliance resources. A resource that predates a policy stays unchanged and is only reported. Example: a pre-existing 0.0.0.0/0 access-list entry was not removed after a blocking policy was applied, but appeared as non-compliant. (high confidence) — [source](https://foojay.io/today/enforcing-governance-in-mongodb-atlas-with-resource-policies/) - A new policy applies to new clusters immediately. A non-compliant existing cluster can only be changed in ways that bring it into compliance (for example, if AWS is blocked, an AWS cluster can only move to Azure or Google Cloud). (medium confidence) — [source](https://mongodb.com/docs/atlas/atlas-resource-policies) - Shard-count false positives: Atlas treats a replica set as having one shard. A minShardCount policy above 1 without a cluster-type check flags every replica set in scope as non-compliant, flooding the endpoint. Use minShardCount only for minimums and maxShardCount only for maximums, and add the !context.cluster.isConvertingToSharded exception for replica-set-to-sharded conversions. (high confidence) — [source](https://mongodb.com/docs/atlas/atlas-resource-policies) - Policies are allow-by-default and organization-wide, with per-project exclusions possible. Each policy needs a unique name (duplicate returns 400). Create, update and delete generate activity-feed events. Docs recommend several small policies over one large one, which keeps non-compliance reports attributable to a specific policy. (medium confidence) — [source](https://mongodb.com/docs/atlas/atlas-resource-policies) - Policies can be managed with the Atlas Terraform provider (resource mongodbatlas_resource_policy; provider 1.33.0 or later needed for minimum-TLS and cipher-suite rules). The MongoDB docs describe no Terraform data source for nonCompliantResources, so drift monitoring is done by polling the API or CLI, for example from a scheduled job. (low confidence) (verified as of 2026-10-19) — [source](https://mongodb.com/docs/atlas/atlas-resource-policies) - Atlas resource tags are key-value pairs; max 50 tags per cluster, keys and values up to 255 characters, keys unique and case sensitive ('environment' and 'Environment' are distinct). Managing them needs Project Owner or Organization Owner role. (high confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/tags) - Tags reach billing: tag keys appear as column headings in the invoice CSV export and as a field in the Admin API invoice payload; the pending-invoices endpoint returns results.lineItems.tags. Tags typically show on line items within 48 hours of being added. (high confidence) (verified as of 2026-10-18) — [source](https://mongodb.com/blog/post/deployment-tags-billing-invoices-now-generally-available) - Invoice tagging scope is limited: line items not billed at the cluster level are not eligible for tagging, and cluster line items without a given tag key show blank cells for that column. Untagged spend therefore remains in any report. (medium confidence) (verified as of 2026-10-18) — [source](https://mongodb.com/blog/post/deployment-tags-billing-invoices-now-generally-available) - MongoDB's guidance: define a strategy enforcing mandatory tags on all deployments (start with environment and application), use a standard case-sensitive format, use namespaces/prefixes per business unit, manage tags via Terraform or Admin API, and run a governance routine that checks for untagged or improperly tagged deployments. Cost keys suggested: cost-center or billing-unit. (high confidence) — [source](https://mongodb.com/blog/post/deployment-tags-billing-invoices-now-generally-available) - Do not put PII/PHI in tags: Billing and other MongoDB services can read them, and cluster tags are sent to Datadog and Prometheus integrations. Tag changes are recorded in the Project Activity Feed. (high confidence) — [source](https://www.mongodb.com/docs/atlas/tags) - AWS provider default_tags (since v3.38.0, April 30 2021) is a provider-block feature where resource-level tags win and Auto Scaling Groups are excluded. The Atlas registry page for advanced_cluster documents no provider-level default tags, so Atlas tags must be set per resource or merged in a module via locals and merge(). (low confidence) (verified as of 2026-10-19) — [source](https://www.hashicorp.com/en/blog/default-tags-in-the-terraform-aws-provider) - HCP Terraform evaluates Sentinel or OPA policy sets against the plan on each workspace run; failed policies can stop the run depending on enforcement level, and permitted users can override. Each policy set uses one framework, but sets from different frameworks can be attached to one workspace. A newer HCL-based 'Terraform policy' (beta) also covers Stacks. (medium confidence) (verified as of 2026-10-19) — [source](https://developer.hashicorp.com/terraform/cloud-docs/policy-enforcement)