Normalized Compute Unit

Definition.
A Normalized Compute Unit (NCU) is a standardized unit of measure that represents compute consumption in a consistent and comparable form. It translates underlying computational work into a normalized unit that can be used for estimation, comparison and control.

In other words, NCU makes compute measurable in a way that is in a way that remains consistent across different hardware and software environments.

Why this exists.
While compute can be measured, it is not inherently comparable across:

• models
• architectures
• execution environments
• workloads

This creates three structural problems:

• Inconsistency: identical workloads may produce different compute measurements depending on model or system
• Non-comparability: compute usage cannot be directly compared across providers or workflows
• Inoperability: raw compute measurements are not usable for pricing, budgeting, or control

NCU resolves this by standardizing compute into a consistent unit.

What “normalized” means.
Normalization refers to the process of converting raw compute measurements into a common unit that is independent of underlying system differences.

Rather than relying on model-specific or infrastructure-specific metrics, NCU expresses compute in a form that is:

• consistent across environments
• comparable across requests
• stable over time

The public description is intentionally abstract; the specific normalization methods are not disclosed.

NCU’s defining characteristic: comparability.
NCU enables direct comparison between workloads that would otherwise be incomparable.

This allows:

• one request to be evaluated against another
• one system to be evaluated against another
• one provider to be evaluated against another

Without normalization, compute remains fragmented and difficult to reason about.

How it works.
Raw compute measurements are translated into NCUs through a normalization process that accounts for differences in execution context.

The result is a standardized unit that can be:

• estimated prior to execution
• applied consistently during execution
• reconciled after execution

This enables compute to function as a stable unit of measure across the system.

What measurement means in NCU.
An NCU represents the standardized value of compute required to perform a task.

It is not:

• tied to a specific model
• dependent on infrastructure
• limited to a single execution environment

It is a normalized representation of compute that enables consistent interpretation.

Relationship to FBM.
FBM measures the underlying computational work performed.

NCU translates that measurement into a standardized unit.

Together:

• FBM provides accuracy
• NCU provides consistency

Without FBM, NCU lacks grounding.
Without NCU, FBM lacks usability.

Relationship to G-PEP.
G-PEP governs execution based on defined constraints.

NCU provides the unit used to define those constraints.

This enables:

• pre-execution estimation
• enforceable limits
• consistent policy application

Without a normalized unit, governance cannot be applied reliably.

Difference from token-based units.
Token-based systems use units tied to text output.

NCU is independent of text and reflects standardized compute.

Tokens answer: how much text was produced
NCU answers: how much standardized compute was consumed

This distinction enables meaningful comparison and control across diverse workloads.

Cost discipline.
NCU enables cost discipline by providing a stable unit for budgeting, pricing and control.

When combined with FBM and G-PEP, it allows:

• consistent estimation
• comparable measurement
• enforceable constraints

Without a normalized unit, cost cannot be consistently managed across systems.

Current state of AI measurement.

AI systems today lack a universally comparable unit of compute.

Measurement is fragmented across:

• tokens
• infrastructure metrics
• provider-specific abstractions

This fragmentation prevents consistent control and comparison.

NCU addresses this gap by providing a normalized unit.

Origin and engagement.
The Normalized Compute Unit is part of a broader architectural system for aligning AI cost, measurement and control.

Organizations evaluating compute-based approaches typically consider NCU alongside:

• compute-based measurement systems
• pre-execution control mechanisms
• governance frameworks

Public discussion is intentionally incomplete; failure modes only become clear at the architectural level. Confidential architectural review available upon request.