Key Takeaways

  • SAP AI Core is a managed AI inference runtime deployed on SAP's Business Technology Platform (BTP), billed via consumption-based cloud credits with no perpetual licensing option.
  • SAP AI Launchpad provides enterprise-grade model management, data governance, and orchestration for deployments at scale, with strict Professional user licensing requirements.
  • The consumption model creates hidden cost exposure: credit prices vary by SAP region, burst workloads scale unpredictably, and there's no ceiling protection.
  • BTP integration is mandatory, coupling AI Core adoption to your broader cloud strategy and limiting exit optionality after initial deployment.
  • Professional user licensing in AI Launchpad is enforced per actual user, not per instance, making large MLOps teams significantly more expensive than enterprises anticipate.
  • Enterprise deployment often requires S/4HANA integration, SAP Data Warehouse Cloud adoption, and AI Business Services bundles, locking buyers into further SAP ecosystem spend.
  • Negotiation leverage exists: volume commitments, consumption guarantees, and fixed credit caps are all on the table if raised before LOI signature.

What is SAP AI Core?

SAP AI Core is a managed generative AI and machine learning inference runtime deployed on SAP Business Technology Platform (BTP), the company's unified cloud operating system. It sits in the intersection of three enterprise pressures: the need to operationalize AI models in production, SAP's desire to deepen customer lock-in through platform services, and the emerging regulatory requirement to trace AI deployments and manage data lineage.

Unlike traditional perpetual SAP licenses, AI Core operates on pure consumption: you provision compute capacity, deploy pre-trained or custom models (whether Joule foundation models or third-party LLMs via SAP's orchestration layer), and pay for each inference request plus underlying infrastructure. There is no "SAP AI Core license." The product is the consumption model itself.

Technically, AI Core provides containerized model serving via Kubernetes, integrated secrets management, GPU allocation (variable by region and demand), and multi-tenancy isolation within your BTP account. For enterprise architects, this sounds reasonable. For procurement teams, this translates to "variable monthly spend with limited predictability."

SAP AI Launchpad: The Enterprise Control Plane

SAP AI Launchpad is the companion product and, in practice, the mandatory enterprise wrapper around AI Core. It provides model lifecycle management, data governance, deployment orchestration, audit logging, and role-based access control (RBAC) across teams. If AI Core is the runtime, Launchpad is the command center.

Launchpad is licensed per Professional user—a user classification that includes data scientists, ML engineers, and anyone with model management responsibilities. SAP's definition of "Professional" is deliberately broad: it includes users who perform any of these tasks:

  • Create, train, or fine-tune models
  • Deploy models to production via Launchpad orchestration
  • Access model versioning, lineage, or governance features
  • Manage deployment pipelines or monitoring dashboards
  • Execute audit or compliance reports on model activity

Enterprises with MLOps teams of 8–15 people often discover, during licensing reviews, that all of them are classified as Professional users. At ~€600/month per Professional user (enterprise scale), a moderately sized ML team bills €5,000–€9,000 monthly in Launchpad licensing alone—before consuming a single AI Core inference credit.

Licensing Models and Pricing Structures

BTP Credits: The Foundation Currency

Both AI Core and AI Launchpad consume BTP credits, SAP's unified cloud consumption currency. A single credit is approximately €0.00012 (subject to volume discount), and enterprises purchase credit blocks in advance. A typical enterprise might purchase 100,000 to 1,000,000 credits annually, depending on compute footprint.

The challenge: credit consumption varies significantly by resource type and region. GPU inference in Europe (Frankfurt) costs 2.5x more per inference than on CPU in North America. Batch processing is cheaper than real-time requests. Prompting Joule foundation models via SAP Business AI has different credit profiles than bringing your own model.

SAP publishes credit pricing in a matrix, but does not guarantee fixed rates across the contract term. Pricing adjustments are contractually permissible, creating multi-year budget uncertainty for large deployments. For a structured approach to forecasting SAP AI costs across all billing dimensions, see our guide to SAP AI budget and forecasting for 2026.

Professional User Licensing

AI Launchpad Professional licenses are sold as fixed monthly subscriptions, with volume discounts at specific seat thresholds. They do not support "consumption-based" pricing—you either license a user or you don't. This creates a planning problem: as your MLOps org scales, licensing costs jump in discrete increments.

Limited and Free User Engagement (FUE) classifications exist but have severe restrictions. A Limited user can view deployed models and run inference via API but cannot manage deployments or access governance tools. FUE users can observe model metadata but not execute any operational function. Neither qualifies for production MLOps work.

Bundle Traps: AI Launchpad + Data Warehouse Cloud + Business AI

SAP strongly incentivizes bundling AI Launchpad with SAP Data Warehouse Cloud (for metadata management and data lineage) and SAP Business AI (for pre-built generative AI applications like "Invoice Recognition" and "Intelligent Document Processing"). Bundled pricing is more attractive than à la carte, but it locks enterprises into buying SAP's data platform and pre-built AI services ecosystem—whether or not they directly use those services.

In practice, many large deployments include all three, with total annual costs reaching €50,000–€500,000 depending on user count, model complexity, and inference volume.

S/4HANA & BTP Integration: The Lock-In Vector

SAP's strategic intent is to position AI Core as the inference layer for S/4HANA–based AI applications. This means:

  • Master data synchronization: AI models must pull training data from S/4HANA financial, supply chain, or HR modules via direct API integration. This requires active S/4HANA connectivity and ongoing data governance.
  • Fiori embedded AI: SAP is shipping pre-built AI features in Fiori UI applications (e.g., "Demand Sensing," "Accounts Receivable Aging"). These invoke AI Core models behind the scenes, creating inference dependencies on S/4HANA licensing decisions.
  • RISE with SAP coupling: Enterprises on RISE with SAP contracts often discover that AI Core consumption is treated as part of the overall RISE spend cap—not separately priced. This incentivizes underestimating AI workload volumes during contract negotiation.

The result: adopting AI Core effectively requires maintaining an active S/4HANA subscription, SAP HANA Cloud for training data, and sufficient BTP credits. It is not a "best-of-breed AI platform" option; it is a commitment to the SAP stack.

Consumption-Based Pricing & Hidden Cost Exposure

AI Core consumption is billed monthly in arrears based on actual resource usage. SAP provides a billing API and consumption portal, but granular cost attribution (which team, which model, which inference type) requires custom instrumentation. Out-of-the-box reporting does not show cost per model or cost per user journey.

Common hidden costs include:

  • Burst overage: A peak inference load during month-end financial closing can consume 30% of monthly credits in a single day, but there is no monthly cap or prepaid ceiling. You are liable for all overage at per-credit rates.
  • Model experimentation: Data scientists fine-tuning models in dev/test environments incur identical per-inference costs to production, even at 0.01% scale. Many enterprises see 40–60% of credit consumption in lower environments.
  • Cross-region failover: If your primary region experiences latency, automatic failover to backup regions executes at 2–3x higher credit cost. You receive the bill; failover decisions are not gated by cost.
  • Logging and observability: Detailed model inference logging (for regulatory compliance or debugging) adds 15–25% overhead to credit consumption. SAP does not automatically manage this; it's an option you enable.

Enterprise Deployment Architecture

A typical large enterprise AI Core deployment involves:

  1. Foundational layer: S/4HANA as master data source, SAP HANA Cloud as training data warehouse, BTP as integration backbone.
  2. Model layer: Multiple AI Core deployments per region (Europe, Americas, APAC), supporting both generative models (e.g., invoice analysis via Joule) and custom ML models (demand forecasting, churn prediction).
  3. Governance layer: AI Launchpad for central model registry, data lineage, and audit. SAP Data Warehouse Cloud for metadata governance.
  4. Application layer: Fiori embedded AI, custom Joule-powered applications, third-party BI tools querying AI-augmented data sets.

This architecture demands deep BTP expertise, ongoing data quality management, and continuous monitoring of model performance. It is not a "turn-key" deployment; it is a 12–18 month journey, often requiring external consulting (SAP Services or partners).

Compliance, Governance & Risk

AI Core deployments carry specific compliance risks that enterprises must mitigate contractually:

Data Residency and Sovereignty

AI Core instances run within SAP's BTP data centers, which are regional but not always customer-selectable. For regulated industries (finance, healthcare, energy), data sovereignty requirements may mandate specific regions or even on-premise deployment. SAP's contractual language on data residency is permissive—it reserves the right to move workloads between "nearby" regions for load balancing. This is incompatible with strict residency requirements and must be negotiated.

Model Audit and Explainability

Regulatory bodies increasingly require explainability for AI decisions affecting lending, hiring, or resource allocation. SAP AI Launchpad provides model versioning and lineage tracking, but it does not automatically generate regulatory-grade explanations or impact assessments. Enterprises must build custom audit workflows on top of Launchpad APIs.

IP and Model Ownership

When you deploy a custom model in AI Core, SAP retains the right to anonymize model logs for product improvement and security research. Enterprise IP concerns (proprietary models, competitive advantage models) require explicit contractual carve-outs before deployment. Standard terms are ambiguous.

SAP's Generative AI: Joule, Business AI, and Licensing Implications

SAP Joule is the company's foundation model platform, available via both direct API access and via pre-built "Business AI" applications. Joule deployments in AI Core are metered separately from custom ML models and incur per-token or per-request pricing in addition to infrastructure credits.

Business AI applications include pre-built use cases like:

  • Invoice recognition and processing
  • Contract analysis and risk flagging
  • Demand sensing and supply chain optimization
  • Narrative generation for financial reports

These are licensed per module (e.g., €8,000–€15,000/year for "Invoice AI") plus per-document processing fees. The licensing structure is deliberately opaque: it's not clear whether a Business AI license includes unlimited inference or if there's a transaction cap. SAP forces enterprises to estimate volume in the sales process, with penalties for over/under-estimation.

Negotiation Strategies & Contract Red Flags

Enterprises have more negotiating leverage on AI Core contracts than they typically assume. Key tactics:

1. Volume Commitments with Consumption Floors

Rather than purchasing credits on demand, negotiate a 12-month commitment with a minimum guaranteed consumption (e.g., 500,000 credits/month). If you consume less, you still pay for the minimum; if you consume more, overage is charged at a discount (e.g., 10% less than standalone credit rates). This caps your downside and simplifies budgeting.

2. Fixed Credit Caps

Push back on "consumption models" by imposing a monthly or annual ceiling. Once you hit the cap, new inference requests are queued or rate-limited, not billed at overage rates. This requires governance discipline (which models run, how often), but it eliminates surprise bills.

3. Professional User Overages

Challenge the broad definition of "Professional user." Negotiate tiered classifications: developers who touch code are Professional; business users who run pre-built models via Fiori are Limited; observers are FUE. Ensure the contract allows reassignment of users to lower tiers as roles shift.

4. Pricing Term Lock

Standard SAP terms allow "pricing adjustments" annually. Negotiate a fixed credit price for the full contract term (typically 2–3 years). If SAP insists on price escalation clauses, cap them at 3% annual increase (rather than the standard 5–8%).

5. Data Residency Carve-Outs

Explicitly state which data centers your AI Core instances must use and prohibit SAP from migrating workloads without 90-day written notice and customer approval. This is especially critical for GDPR, HIPAA, and industry-specific data residency mandates.

6. Exit Rights & Model Portability

Require SAP to provide 30-day notice before discontinuing AI Core or raising prices materially. Negotiate a right to export trained models, training data lineage, and inference logs in standard formats (ONNX for models, Parquet for data). This limits lock-in after 2–3 years.

Enterprise Case Study: Multinational Financial Services Firm

A large European bank deployed AI Core for invoice processing, demand forecasting, and regulatory reporting. Initial contract: 2-year term, 400,000 BTP credits/month (€48,000/month), 12 Professional users in AI Launchpad (€7,200/month), plus SAP Data Warehouse Cloud and Business AI "Invoice" module.

Year 1 reality: Actual consumption climbed to 680,000 credits/month due to higher-than-estimated model experimentation and cross-region failover. Invoice AI processing doubled in Q3, pushing transaction fees 60% above budget. Professional user count grew to 18 (data engineers joining the team), adding €7,200/month in unexpected licensing.

Cost impact: €400,000 overage in Year 1 alone. During contract renegotiation, SAP threatened to discontinue below-market discounts unless the bank expanded to include additional Business AI modules (adding €150,000/year).

Resolution: The bank engaged independent SAP licensing advisors to challenge the upgrade pressure, renegotiated consumption floors with strict monthly caps, converted 8 users to Limited classification, and secured fixed credit pricing for Year 2. New cost structure: €540,000/month baseline (lower per-credit rate from volume commitment) with €50,000 monthly ceiling on Professional users. Three-year deal, no further automatic price increases.

Frequently Asked Questions

Can we run custom models in AI Core without using Joule or SAP Business AI? +
Yes. You can deploy any containerized model (PyTorch, TensorFlow, ONNX) in AI Core, even if you're not using Joule or Business AI. However, SAP's sales strategy heavily incentivizes bundling, and discounts often require commitment to the full stack. Negotiate separately for custom-only deployments if that's your use case.
What happens if we hit our BTP credit limit mid-month? +
By default, you are liable for all overages at per-credit rates. New inference requests will continue to be serviced and billed. To prevent surprise bills, negotiate a monthly cap clause in your contract, which will either queue requests or rate-limit them once the cap is hit. This is a critical negotiation point.
Is AI Launchpad required if we only deploy one model? +
Technically, you can deploy a model to AI Core via raw APIs without Launchpad. However, SAP strongly incentivizes Launchpad for governance, versioning, and audit compliance. For enterprise deployments, Launchpad is practically mandatory; for sandbox or POC deployments, you may skip it.
Can we negotiate a fixed price for Professional users instead of per-user licensing? +
This is difficult but possible. SAP prefers per-user licensing for revenue predictability. However, if you can commit to a fixed user count for 2+ years, you may negotiate a flat monthly fee. Ensure the contract allows one reassignment per quarter to accommodate team changes without penalty.
What's the difference between AI Core and Azure OpenAI / AWS Bedrock? +
AI Core is SAP's managed inference platform tightly integrated with BTP and S/4HANA. Azure OpenAI and AWS Bedrock are cloud-agnostic inference services supporting multiple models. AI Core has lower latency for S/4HANA-integrated workflows but higher lock-in and operational overhead. For enterprises already on Azure or AWS, those platforms are often cheaper and simpler.
Are there any hidden costs in Professional user licensing? +
Yes. SAP's definition of "Professional" is broad and includes anyone who touches model management. Contractors, temporary staff, and consultants count as Professional users. Negotiate flex licensing for peak periods (e.g., ability to add/remove users quarterly rather than monthly). Also clarify whether named users or concurrent users are licensed; SAP defaults to named users, which is more expensive.
Can we use open-source alternatives to AI Core? +
Absolutely. Kubernetes-based alternatives like KServe, Seldon, or Ray support the same workloads and can be deployed on any cloud or on-premises. The trade-off: you own the operational overhead (monitoring, scaling, security patches). For enterprises comfortable with DevOps, open-source is often cheaper than SAP's managed offering.
Is there a way to avoid SAP Data Warehouse Cloud if we use AI Core? +
Data Warehouse Cloud is not mandatory, but SAP bundles it aggressively and often makes pricing discounts conditional on adoption. If you want to avoid it, push back early in negotiations. You can manage data lineage and metadata via custom solutions or third-party tools. Negotiate explicitly for discount parity with a Data Warehouse Cloud-free configuration.

Pre-Signature Negotiation Checklist

Before signing any SAP AI Core or Launchpad contract, ensure your team has:

Critical Negotiations

  • Define "Professional user" explicitly and reserve the right to reassign users quarterly.
  • Establish a monthly BTP credit consumption ceiling (hard cap, not soft limit).
  • Negotiate fixed credit pricing for the full contract term.
  • Require written notice (90 days) before any discontinuation of AI Core or price increases >3% annually.
  • Prohibit data residency migration without customer approval.
  • Secure export rights for trained models and training data in standard formats.
  • Limit SAP's right to use model logs for product improvement (carve-out for proprietary models).
  • Clarify whether overage charges apply to burst periods or if rate-limiting is enforced.

Next Steps: Getting Specialized Advice

SAP AI Core licensing is complex, and the terms are negotiable. Most enterprises leave 20-40% of potential savings on the table by not engaging in detailed contract review before signature. This guide provides the foundational knowledge; the next step is engaging independent advisors to conduct a licensing audit and negotiate on your behalf.

Our SAP Contract Negotiation service includes a comprehensive review of any proposed AI Core or Launchpad agreement, identification of cost reduction opportunities, and negotiation support with SAP. We've helped enterprises reduce AI Core spend by 25-45% through strategic renegotiation.

For those considering a RISE with SAP engagement or S/4HANA migration that includes AI Core, our RISE with SAP Advisory service covers the full licensing and cost analysis of bundled offerings, ensuring AI costs don't become a hidden expense in the broader agreement.

To explore SAP license optimization specific to your deployment, or to access our complete RISE with SAP guide, reach out for a free consultation.

Related Articles in This Cluster

Newsletter

Get Weekly SAP Licensing Insights

Exclusive analysis of SAP contract trends, pricing changes, and negotiation tactics. Delivered to your inbox every Friday.