SAP AI Core & Launchpad: Pricing and Budget Planning

SAP AI Core and SAP AI Launchpad represent a seismic shift in how enterprises approach application intelligence. But for procurement teams and CFOs, the pricing model remains opaque. Unlike traditional SAP licensing—where named users and instances drive cost—AI pricing is consumption-based, hidden in BTP (Business Technology Platform) credits, and prone to cost overruns that dwarf initial estimates.

This guide decodes SAP AI Core pricing, reveals the real costs behind Launchpad deployment, and provides a budget planning framework to prevent bill shock in 2026 and beyond.

Understanding SAP AI Core Pricing: The Consumption Model

SAP AI Core operates on a consumption-based pricing model, fundamentally different from traditional SAP licensing. Instead of paying for named users or instance counts, you pay for what you use in real time. This model introduces both opportunity and risk.

Pricing is denominated in SAP BTP (Business Technology Platform) credits. One credit unit equals €1. Consumption is measured in four dimensions:

• Compute hours: CPU and GPU time allocated to model inference and training.
• Storage: Data persistence in SAP HANA Cloud, object storage, or hybrid repositories.
• API calls: Each inference request against a deployed model.
• Data egress: Transferring data out of SAP Cloud infrastructure to third-party systems or on-premises.

A mid-market enterprise deploying three AI models for demand forecasting, inventory optimization, and customer churn prediction should expect:

This €98,400 annual cost excludes SAP AI Launchpad (see below), licensing for underlying S/4HANA or BTP subscriptions, and cost inflation beyond the base year.

The AI Unit Pricing Model: GPU vs. CPU Trade-offs

Within SAP AI Core, compute is priced asymmetrically. GPU-accelerated workloads (essential for deep learning models) cost significantly more than CPU-only inference.

• CPU compute (inference): €0.15–€0.25 per hour.
• GPU compute (NVIDIA A100 inference): €2.50–€3.50 per hour.
• GPU compute (model training): €3.50–€5.00 per hour (due to full GPU allocation).

The pricing gap explains why enterprises often underdeploy GPU-backed models. A typical LLM-fine-tuning workload (20 GPU hours/week) costs €260/week or ~€13,500 annually. Scale this to five concurrent research initiatives, and you're at €67,500 per year—easily overlooked during budget planning but material at renewal.

SAP AI Launchpad: The Hidden "Free" Management Layer

SAP markets SAP AI Launchpad as a "free" management and governance layer for SAP AI Core. In marketing materials, it appears bundled. In reality, it has its own consumption model.

SAP AI Launchpad enables:

• Model lifecycle management (versioning, deployment, rollback).
• Data governance and lineage tracking.
• Explainability and bias detection (for regulated industries).
• Monitoring and cost tracking dashboards.

While the Launchpad console itself is included in a BTP subscription, data operations within Launchpad incur charges:

• Metadata storage and retrieval: €0.10–€0.30 per 1,000 API calls.
• Model versioning (git-style storage): €50–€200/month depending on model size and frequency of updates.
• Governance audit logs: €0.05 per 1,000 log entries.

For an enterprise managing 10–15 AI models in production, Launchpad operational costs are typically €300–€800 per month or €3,600–€9,600 annually. This is often buried in BTP bills and not flagged as a separate line item.

Hidden Costs: Where Budget Overruns Originate

Data Egress: The Silent Cost Multiplier

SAP AI Core must integrate with legacy on-premises ERP, data warehouses, and third-party CRMs. Moving data in and out of SAP Cloud incurs egress charges.

• Inbound (data to SAP Cloud): Free or minimal cost.
• Outbound (data from SAP Cloud): €0.10–€0.30 per GB, depending on destination.

An enterprise syncing 500 GB of daily forecasts, recommendations, and model outputs back to on-premises systems incurs €50–€150/day or €18,250–€54,750 annually in egress costs alone.

Model Training vs. Inference Pricing Differential

SAP distinguishes pricing based on whether a model is training (learning weights) or inferring (making predictions). Training incurs premium pricing:

• Inference: €0.15–€0.25/CPU hour, €2.50/GPU hour.
• Training: €0.30–€0.40/CPU hour, €5.00/GPU hour.

Retraining a demand forecasting model monthly (40 GPU hours per retraining cycle) costs €200/month or €2,400 annually. Extend this to five models, and you're at €12,000 annually—a cost many teams omit from budget forecasts.

HANA Cloud Dependency Tax

SAP AI Core requires SAP HANA Cloud as a mandatory data layer. You cannot "bring your own database" without significant engineering effort. HANA Cloud pricing is separate from AI Core:

• HANA Cloud (single-node): €3,000–€8,000 per month.
• HANA Cloud (multi-node HA): €15,000–€40,000 per month.

This is a fixed cost orthogonal to AI consumption. Even if your AI workloads scale down in a given month, HANA Cloud charges persist. Plan for €36,000–€480,000 annually in HANA Cloud costs.

BTP Credit Depletion Rates: The Real Math

SAP BTP provides a unified credit pool for AI Core, integration, analytics, and other services. AI workloads deplete credits 2–5x faster than standard BTP services because of GPU compute and data egress intensity.

Example depletion scenario for a mid-market enterprise:

At €23,000/month, this enterprise requires a BTP commitment of €276,000 annually. If AI workloads grow by 50% (a common scenario), credits spike to €12,300/month, and total BTP consumption exceeds €334,000—a 21% cost increase without new licenses or users.

Competitive Pricing: SAP AI Core vs. Cloud Alternatives

How does SAP AI Core stack against AWS, Azure, and Google Cloud?

SAP AI Core GPU inference costs are 85–220% more expensive than AWS, Azure, or Google. The differential widens with training workloads.

Why pay more?

• Ecosystem lock-in: If you're already invested in S/4HANA, RISE, or SAP Analytics Cloud, SAP AI Core integrates natively with minimal engineering overhead.
• Governed integrations: Pre-built connectors for SAP Ariba, SuccessFactors, and other SAP modules reduce middleware costs.
• Compliance: For regulated industries (banking, pharma), SAP's data residency and audit trails are aligned with existing SAP deployments.

The business case for SAP AI Core is strongest when: (1) your compute footprint is modest (<5,000 GPU hours/year), (2) your team lacks cloud engineering expertise, or (3) regulatory requirements mandate SAP tenancy.

The "Free Tier" Trap: Expiring AI Credits

SAP bundles initial AI credits with RISE with SAP, S/4HANA Cloud, or BTP subscriptions. These credits are marketed as "free" but come with a critical catch: they expire after 12 months.

Initial credit allocations typically range from €5,000–€50,000 depending on the contract. Upon expiration, if you don't renew your commitment, consumption converts to on-demand pricing with 20–30% premium rates.

Example:

• Year 1: €30,000 AI credits (bundled, "free").
• Year 2: Credits expire. On-demand GPU pricing rises from €3.50/hour to €4.55/hour (30% premium).
• Year 3: BTP credits expire, on-demand rates increase again.

Enterprises that fail to forecast and commit to ongoing AI spending face sudden cost shocks in Year 2. Budget conservatively: assume expiring credits and plan for baseline ongoing consumption + 25% growth buffer.

Budget Planning Framework for Enterprises

Phase 1: Consumption Estimation (Months 0–3)

• Baseline compute: Inventory all planned AI models. Estimate training frequency (weekly, monthly, quarterly), inference volume (requests/day), and GPU vs. CPU split.
• Storage: Estimate datasets (training, inference caches, versioned models). Plan for 20% annual growth.
• Egress: Map data flows back to on-premises systems. Quantify daily/monthly volumes.
• HANA Cloud: Confirm mandatory HANA Cloud sizing (single vs. multi-node). This is fixed and non-negotiable.

Phase 2: Cost Modeling (Months 3–6)

• Build a consumption cost model in a spreadsheet with scenarios: base case, 30% growth, 50% growth.
• Include all hidden costs: data egress, model training retraining, HANA Cloud, Launchpad governance, storage.
• Annualize and apply inflation factors: GPU costs typically escalate 3–5% annually; SAP rarely discounts compute upfront.
• Add contingency: 20–30% buffer for growth and price changes.

Phase 3: Commitment Negotiation (Months 6–12)

• Request multi-year AI credit commitments with fixed pricing. SAP offers 2–3% discounts for 2-year prepayment.
• Negotiate cap-and-commit terms: agree to a baseline consumption level with price protection; overage costs are negotiable.
• Secure service level agreements (SLAs) for inference latency and model availability, especially if AI powers customer-facing products.
• Request annual cost reviews with adjustment mechanisms if consumption deviates >20% from forecast.

Phase 4: Ongoing Governance (Year 1+)

• Establish monthly cost tracking dashboards. SAP provides BTP consumption analytics, but it's opaque by default. Mandate detailed AI-specific reporting.
• Review model utilization quarterly. Decommission low-value models to free up compute budget.
• Plan infrastructure right-sizing annually. Most enterprises overprovision GPU by 30–50% upfront; as models mature, consolidate and reduce allocation.
• Negotiate annual price reviews. SAP typically publishes escalation clauses (3–5% per year). Push back on anything >CPI+1%.

Right-Sizing: Avoiding Overprovision

A common pitfall: enterprises over-allocate GPU compute upfront, fearing "not enough capacity." In reality, most AI models mature and stabilize within 6–12 months. Inference workloads become predictable; retraining frequency drops.

Right-sizing best practices:

• Start lean: Pilot AI models with minimal GPU allocation (e.g., 2–4 GPU hours/week per model). Monitor performance and user feedback.
• Auto-scaling: Use SAP AI Core's auto-scaling policies to match compute to actual demand. This prevents idle GPU hours.
• Batch vs. real-time: Shift inference to batch processing where possible (e.g., overnight or off-peak). Batch inference is 40–60% cheaper than real-time APIs.
• Model consolidation: Combine multiple small models into a single ensemble model. Reduces inference API calls and deployment overhead.

Annual Escalation and Multi-Year Contracts

SAP AI rider contracts typically include escalation clauses that increase baseline costs 3–5% annually. This is standard but negotiable.

Example impact over 3 years:

Over 3 years, a 4% escalation adds €12,160 to total spend. Negotiate for 2–3% caps, or tie escalation to published SAP cost indices rather than a fixed percentage.

FAQ: SAP AI Core Pricing and Budgeting