Concept card

BDC Capacity Units

As of 2026-09-27

Edited by Cédric Mary, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is BDC Capacity Units?

The 128-CU entry floor rarely covers a real Tier-1 workload: six workload types share one capacity-unit pool, so size a BDC tenant to its peak, not its average.

What it is

Capacity Units are the billing and sizing currency of SAP Business Data Cloud — a bundled measure of compute, memory, storage allotment, and data-flow input/output operations per second, sold in tenant-sized blocks rather than priced per resource. One capacity unit buys roughly one vCPU-equivalent of HANA Cloud compute, about four gigabytes of HANA Cloud memory, a modest storage allotment before overflow pricing kicks in, and a slice of data-flow throughput. The smallest Business Data Cloud tenant starts at 128 capacity units, noticeably larger than a standalone Datasphere tenant's 64-unit floor, reflecting the wider workload mix — replication, live analytics, planning, Joule, Knowledge Graph — that a Business Data Cloud tenant is expected to carry simultaneously. For a consultant sizing an engagement, capacity-unit fluency is the line between a proposal with a predictable monthly bill and the outcome every finance stakeholder dreads: a runaway invoice discovered after go-live.

Why it matters

In Analytics Legends' sizing estimates (2026, not an SAP figure), a Tier-1 pattern sizes to 130-240 CU steady-state and 280-360 CU at peak — more than double the 128-CU tenant floor, so quoting the floor as sufficient sets up a mid-project capacity crisis.
Memory (~4GB per CU) is the dominant cost driver for in-memory analytics, not compute — a sizing conversation anchored on vCPU count is pricing the wrong variable.
Six distinct workload types (SAC live queries, replication flows, Analytic Model evaluation, Joule grounding, Knowledge Graph traversal, Databricks cluster spin-up) compete for one shared CU pool, so a spike in any one starves the rest.

Key points

1 CU = 1 vCPU + 4 GB HANA + storage + IOPS.
BDC tenant min: 128 CU (€100-180k/yr FY26 entry).
Tier-1 sizing (Analytics Legends estimate, 2026): 130-240 steady, 280-360 peak; provision at peak + 20% buffer.
Six workload types: replication · AM queries · SAC Live · Joule · Knowledge Graph · Databricks side.
Joule grounding load scales with active users — size it from measured usage in the tenant, not from a published rule of thumb.
Sizing-at-average = anti-pattern (peak doubles average).
Chargeback: per-Space rolling-30d consumption from Tenant Monitor.
Elastic CU bursting in preview Q2 2026, H2 2026 GA target.

Terms used on this page

Capacity Unit (CU): BDC's billing primitive: ~1 vCPU + 4 GB HANA Cloud + storage + IOPS combined.
Tenant CU pool: Total contracted CU shared across all workloads in a single BDC tenant.
Sizing at peak: Defensive provisioning to measured peak load + 20% buffer. Default; sizing at average is anti-pattern.
Joule grounding load: Compute consumed by Joule agents grounding answers in tenant data; it multiplies with active users. SAP publishes no fixed idle-consumption figure.
Per-Space chargeback: Allocation pattern using Tenant Monitor rolling-30d Space-level consumption to apportion costs across LoBs.
Tenant Monitor: Built-in BDC dashboard showing CU consumption, query workload, replication health per Space.
Elastic CU bursting: Preview Q2 2026 / H2 2026 GA target — temporary exceeding contracted CU at 1.3-1.5× burst rate, billed quarterly.
Surge billing: Charge applied when tenant exceeds storage allotment or IOPS allotment. Replication parallelism is the most-common cause.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →

Guides that answer with this page

These guides cite this page as one of the sources their answer rests on.