Sovereign architecture Multi-cloud portable Intelligence above infrastructure

Own the enterprise intelligence layer. Consume hyperscaler infrastructure.

The strategic architecture is not Azure-first or AWS-first. It is a sovereign enterprise AI architecture where bp Sphere owns finance ontology, policies, agents, evidence, memory, learning, replay, controls, and decision logic while hyperscalers provide replaceable infrastructure services.

Open platform architecture Open fabric documentation Open runtime architecture

North star

Enterprise-owned decision intelligence

Cloud posture

Azure + AWS as infrastructure providers

Portability

Models, clouds, and data stores remain swappable

Strategic asset

Policies, controls, memory, learning, and replay

Executive positioning

The core message: leverage Azure and AWS where they create value, but keep the finance decision intelligence, agent runtime, policy engine, knowledge fabric, and business logic portable and independent of any hyperscaler.

Azure and AWS should provide infrastructure. bp Sphere should provide intelligence. The enterprise should own the knowledge, policies, controls, decision logic, learning, memory, evidence, and agent capabilities that differentiate how finance operates.

bp Sphere vs hyperscalers vs strategic ownership

This table is the executive boundary: what the platform contributes, what cloud providers already offer, and where ownership should sit.

Capability layer	bp Sphere brings	Azure / AWS already provide	Recommended strategy
Enterprise Finance Ontology	Business entities, relationships, finance semantics	None	Enterprise owns completely
Finance Data Model	Cross-process finance model across P2P, O2C, R2R, Treasury, FP&A	Data storage only	Enterprise owns
Enterprise Event Fabric	Financial event normalization and routing	Event Grid, EventBridge, managed Kafka	Enterprise-owned logical layer using cloud infrastructure
Policy Intelligence Runtime	Finance policies, controls, approval rules	Basic workflow and rules services	Enterprise owns
Decision Runtime	Context -> Evidence -> Recommendation -> Approval -> Action	None	Core strategic asset
Agent Runtime	Agent orchestration, authority boundaries, approvals	Generic cloud agent services	Enterprise owns orchestration layer; may consume cloud services
Skill Fabric	Finance-specific skills and reasoning	Generic AI frameworks	Enterprise owns
Evidence Intelligence Fabric	Evidence collection, lineage, audit packs	Storage only	Enterprise owns
Decision Replay	Full decision reconstruction and auditability	None	Enterprise owns
Learning Runtime	Outcome tracking and continuous improvement	Basic ML services	Enterprise owns
Control Intelligence Hub	Controls monitoring, testing, assurance	Security controls only	Enterprise owns
Continuous Close Engine	Close orchestration and risk prediction	None	Enterprise owns
CFO Command Center	Executive insights and decision monitoring	BI dashboards	Enterprise owns
Process Intelligence	Process mining and optimization	Analytics tooling	Sphere + process-mining tools + enterprise ownership
Enterprise Knowledge Fabric	SOPs, policies, work instructions, controls	Search infrastructure	Enterprise-owned abstraction layer
Agent Registry	Agent catalog, metadata, authority management	None	Enterprise owns
Prompt Management	Prompt governance and versioning	Partial support	Enterprise owns
AI Governance	AI policy, compliance, approvals	AI governance services	Shared; enterprise defines policies
Human-in-the-Loop Controls	Approval workflows and authority boundaries	Workflow tooling	Enterprise owns
Runtime Observability	Agent monitoring and business diagnostics	Cloud monitoring services	Enterprise owns business observability; cloud provides infrastructure observability
FinOps for AI	Token tracking, model economics, optimization	Cost reporting	Enterprise owns optimization logic
Security & Identity Integration	Business authorization model	Entra ID, IAM, Cognito	Shared
SAP Cost Optimization Layer	Read minimization, caching, event sourcing	None	Sphere capability
Hybrid SAP Coexistence	ECC + CFIN + S/4 abstraction	None	Sphere capability
Enterprise Memory Layer	Organizational memory and decision history	Vector DB infrastructure	Enterprise owns
Multi-Agent Collaboration	Finance-specific collaboration patterns	Generic agent frameworks	Enterprise owns
Transformation Discovery Agent	Opportunity identification and redesign	None	Sphere capability
Autonomous Operations Layer	Governed execution with approvals	None	Long-term enterprise asset

Capabilities the enterprise should strongly own

These become intellectual property regardless of cloud provider because they encode enterprise semantics, risk posture, controls, knowledge, and operating discipline.

Strategic capability	Why ownership matters
Enterprise Ontology	Defines enterprise business semantics and should remain portable.
Policy Runtime	Encodes controls, compliance, approval thresholds, and human-boundary rules.
Agent Registry	Governs all enterprise agents, ownership, authority, lifecycle, and auditability.
Decision Runtime	Turns context, evidence, policy, recommendation, approval, action, and replay into one reusable operating discipline.
Control Intelligence Hub	Becomes the assurance layer across finance, controls, and audit.
Evidence Intelligence Fabric	Creates regulator, auditor, and executive trust through lineage and evidence packs.
Enterprise Memory	Captures institutional knowledge, outcomes, exceptions, and prior decisions.
Learning Runtime	Creates compounding advantage from outcomes, overrides, backtests, and controlled improvement.
Decision Replay	Provides accountability and reconstructability for sensitive decisions.
Event Fabric	Acts as the enterprise nervous system for finance and operations events.
Finance Skill Library	Encodes finance expertise in reusable, governed capabilities.
AI Governance Framework	Defines enterprise risk posture, policy enforcement, model controls, and approval boundaries.

Capabilities that can be consumed from Azure or AWS

These are infrastructure services. They are important, but not strategic differentiators in the finance intelligence layer.

Capability	Azure	AWS
Compute	AKS, VM Scale Sets	EKS, EC2
GPU	Azure GPU	EC2 GPU
Storage	Blob Storage	S3
Event Infrastructure	Event Grid	EventBridge
Identity	Entra ID	IAM
Secrets	Key Vault	Secrets Manager
Monitoring	Azure Monitor	CloudWatch
LLM Hosting	Azure OpenAI	Bedrock
Vector Storage	Azure AI Search	OpenSearch
Databases	SQL, Cosmos DB	Aurora, DynamoDB
Networking	VNET	VPC

Recommended ownership north star

Own	Buy / consume	Reason
Ontology, policies, agents	Compute, storage, GPU	The business semantics and controls stay portable.
Decision runtime, evidence, replay	Networking, identity, monitoring	The intelligence layer does not become cloud-specific.
Learning, memory, event fabric	Managed databases, model hosting, search infrastructure	Infrastructure can evolve without losing institutional intelligence.

The long-term objective is not to build a finance platform on one hyperscaler. The objective is to build an enterprise-owned Decision Intelligence Platform where clouds provide elastic infrastructure while the enterprise owns policies, controls, knowledge, agents, learning, and decision-making assets.

Sovereign enterprise AI architecture

Business layer

Enterprise business functions

Finance

Procurement

Supply Chain

Trading

Operations

↓

Intelligence layer

bp Sphere

Policies

Agents

Evidence

Replay

Audit

Learning

↓

Enterprise data fabric

Data, events, and systems of record

UDP

Databricks

Foundry

SAP

Events

↓

Infrastructure layer

Hyperscaler and private infrastructure

Azure

AWS

Private / on-prem

AI model strategy

Do not build around one model. Build around a model abstraction layer so reasoning, extraction, matching, validation, and drafting can use the right model class at the right cost and risk level.

Model tier	Best use	Operating principle
Tier 1 reasoning models	Discovery, research, process redesign, complex explanation	Swappable large models behind a model gateway.
Tier 2 mid-size models	Operations and analyst assistance	Lower-cost models for routine research and drafting.
Tier 3 local models	Classification, extraction, routing, validation, matching	High-volume transaction support with low cost and data-control benefits.

AI foundation layer

Model gateway, prompt gateway, policy gateway, guardrails, observability, and cost management remain independent of hyperscalers.

Agent platform

Agent registry, skill registry, authority framework, decision replay, and learning runtime remain enterprise-owned.

Decision platform

Finance, procurement, supply chain, trading, retail, and operations reuse the same context-evidence-policy-action-replay discipline.

Executive summary. bp Sphere creates a cloud-portable, multi-cloud enterprise AI architecture. The enterprise can change models, cloud providers, or infrastructure without losing the intelligence assets created through ontology, policy, evidence, controls, agents, memory, learning, and replay.

Sovereign Enterprise AI Architecture