Forvis Mazars × UAE Presidential Office
A working session on infrastructure, architecture, and AI. Prepared for Aymen Al Saadi and Ali Al Khumairi ahead of their Thursday, June 18 session during Vivatech Paris 2026.
×
What we'll cover in this session
Infrastructure, architecture, and AI across four key topics, plus strategic intelligence as supporting evidence.
An open conversation, not a vendor pitch
Reference architectures, honest sovereignty trade-offs, and production patterns only. Open conversation with your architecture and procurement teams.
What we will share
- Reference architectures and operating models we use with clients today.
- Honest lessons learned on sovereignty, including trade-offs we discovered.
- Concrete patterns for governed AI delivery: frameworks, tagging, access control.
- How we adapt infrastructure from cloud baseline to fully sovereign on-prem.
What we will not do
- Claim sovereignty where connectivity or vendor paths undermine it.
- Recommend rebuilding everything in-house when market accelerators fit.
- Present innovation theatre: production patterns and proof points only.
- Hide vendor constraints that block your chosen infrastructure path.
Sovereignty is a spectrum, not a label
Physical location, residency, air-gap, and vendor models each land at different points on the spectrum.
Infrastructure assessment
- Where is compute? Bare metal, colocation, national cloud, hybrid?
- Who operates it? Internal team, managed services, 24/7 SLA?
- What connectivity exists? Cross-border paths can undermine residency claims.
- What attack surface? Risk assessment and isolation per data perimeter.
What we'll cover together
Reference architectures with operating models, transparent sovereign vs. assumed sovereign assessment, and proof from organizations facing similar compliance bars.
GAIA: what running your own AI infrastructure actually means
Forvis Mazars operates its own GPU servers. Here is what that looks like in practice.
Server operations
Managed GPU servers under the Zettafox domain. Physical compute owned and operated by Forvis Mazars, not rented per token from a hyperscaler.
Model refresh cycle
Open-weight models evolve fast. Running your own infra means deciding when to pull new model versions, test them against your use cases, and cut over without asking a vendor for access.
API exposure
Models are exposed via a unified API endpoint. Applications, agents, and workflows call the same endpoint regardless of which model is active underneath.
Use-case routing
Requests are routed to the right model based on data classification and use case. Restricted workloads stay on sovereign compute. Lower-sensitivity tasks can use lighter models or fallback tiers. LiteLLM handles the routing layer.
Our sovereignty journey: transparency builds trust
Forvis Mazars evolved from owned servers to colocation, then reassessed connectivity and vendor paths.
Forvis Mazars evolved its infrastructure posture as sovereignty requirements became clearer. We moved from owned physical servers to European colocation, then discovered that connectivity paths and vendor dependencies can still create exposure. We are now actively working with European hosting partners to close those gaps.
Owned servers
Physical servers in our own facilities.
European colocation
Scale and resilience through in-region hosting.
Deeper assessment
Connectivity, vendor paths, GPU compliance re-evaluated.
Infrastructure matched to purpose
Azure for internal IT; dedicated servers for R&D and client needs; mission intent drives the choice.
Private architecture is an operating model, not just hardware
Ownership, resilience, and observability matter as much as the compute stack.
Who runs it
Internal ops, managed services partner, or hybrid. Clear ownership for prod workloads.
Resilience & SLA
24/7 support, documented runbooks, failover. Does the operator know your applications?
Observability
Your own alerting even when the provider is responsible. Sensors, dashboards, escalation.
Operating principle
Infrastructure documentation, architecture records, and independent monitoring matter as much as the hardware. If the managed services partner fails silently, you still need alerts in your control plane.
Private AI: the hard problems are operational, not theoretical
Compute, isolation, and threat modeling decide whether sovereign claims hold in practice.
Compute
- Bare metal vs. virtualized GPU clusters.
- Right people in the right jurisdiction to operate.
- Compliance minimums on GPU and inference stack.
Isolation
- Air-gapped vs. hybrid sovereign patterns.
- Network segmentation between data perimeters.
- Agent sandboxes for untrusted code execution.
Threat model
- Attack surface assessment per tier.
- Connectivity audit: paths outside desired jurisdiction.
- Vendor dependency mapping on sovereign claims.
Private architecture operating model checklist
Clear ownership, documented runbooks, and independent alerting in your control plane.
Ownership & ops
- Who owns prod infrastructure: internal, MSP, or hybrid?
- How do you select and audit the managed services partner?
- Does the operator know your applications and dependencies?
- Documented architecture, runbooks, and change control.
Resilience & visibility
- 24/7 support SLA for all production applications.
- Independent monitoring and alerting in your control plane.
- Escalation when the provider fails to act on alerts.
- Regular resilience testing and failover validation.
Rapid AI delivery requires framework convictions
Formal ontologies are already the deployed standard in the most demanding environments. The delivery question is how to build on what already works.
Why a reusable framework
- One controlled delivery chain, not exotic one-off applications.
- Security patches and fixes propagate across all deployments.
- Governance boundaries enforced by design, not after the fact.
- Duplicate and scale without losing control or auditability.
BFO/CCO: already deployed at scale
- Basic Formal Ontology (ISO/IEC 21838-2:2021) and Common Core Ontologies are the existing standard for national security and biomedical research programs.
- Not an emerging approach: adopted by DARPA, NIH, and defense research communities.
- Provides the shared semantic layer that makes AI systems auditable and interoperable across agencies.
- The right foundation for high-assurance government AI is what already holds in the most demanding environments.
ABAC: segment first, then attribute access
Access control solves who can reach what. Semantic governance solves whether the AI knows what things mean and can be held accountable for its reasoning.
Effective access control combines isolating perimeters and mapping permissions. But governing AI at scale requires a further layer: auditing that the semantic structures inside AI systems are sound. The National Center for Ontological Research (NCOR) is emerging as the authority for this kind of vetting, analogous to how financial regulators audit firm-level controls.
Segmentation
- Network and logical isolation between prod, maintenance, and dev zones.
- Separate agent sandboxes for code execution before promotion.
- Data perimeters: each workload sees only its authorized datasets.
Attribution
- Map business processes to people, roles, and authorized systems.
- Ontology and graph models for complex enterprise authorization (e.g. JP Morgan pattern).
- Agents verify permissions before querying production data platforms.
Classification and tagging drive model and infrastructure routing
Tagged files route to the right model tier; untagged assets enter assessment before processing.
Tag by design
- Every file carries a sensitivity tag: public, internal, confidential, restricted.
- Microsoft and enterprise platforms now enforce tagging at ingestion.
- Untagged assets enter an assessment workflow before AI processing.
Routing logic
- Restricted: full local/sovereign treatment, no external compute fallback.
- Confidential: sovereign GPU or approved private endpoints only.
- Internal/public: may use approved cloud models with audit logging.
- Framework accommodates multiple security typologies by design.
NCOR*: the emerging governance authority
- The National Center for Ontological Research is the vetting authority for semantic structure in AI systems.
- Audits whether AI systems are built on sound, interoperable ontological foundations.
- Analogous to financial regulation: independent review of how AI systems reason, not just what they output.
*Working with TotalEnergies, we are pioneering a new class of services at the intersection of audit, consulting, and technology, vetting their ontology and semantic knowledge graph against BFO. Jeremy Ravenel, CEO of Naas.ai, is working directly with NCOR to operationalize the standard, bringing Forvis Mazars' audit and advisory expertise together with technology partners to deliver this in production.
Cyber in the agentic era: red team, blue team, and the ontology layer
Agentic AI expands the attack surface. Ontology-grounded security frameworks make adversarial behavior and defensive response machine-readable, auditable, and governable at scale.
Adversarial behavior
- Agents introduce new attack vectors: prompt injection, lateral movement through APIs, unauthorized data access.
- MITRE ATT&CK formalizes adversarial tactics as a structured ontology, applicable to AI agent behavior.
- Red team exercises must now include agentic threat models, not just perimeter attacks.
Defensive countermeasures
- D3FEND maps defensive techniques as a formal ontology aligned to ATT&CK.
- Agents can be instructed to monitor, respond to, and escalate using the same vocabulary as the security team.
- Shared ontological language closes the gap between what agents do and what security operators see.
The bridge layer
- ATT&CK and D3FEND are domain vocabularies, not formal ontologies. BFO and CCO provide the semantic layer needed for reasoning, provenance, and AI assurance.
- Mapping these two worlds is active research: work presented at STIDS 2026 is building that bridge, aligning cyber frameworks into the BFO/CCO ecosystem.
- When the alignment lands, the same semantic spine that governs data also governs security, making AI behavior auditable end to end.
Forvis Mazars adapts stack to client constraints
Azure for internal IT; dedicated servers for R&D and client engagements; sovereign options where mission requires.
Infrastructure choice is driven by mission intent, not a single default. Azure covers internal IT operations. Dedicated servers support R&D and client-specific needs. Sovereign and restricted requirements extend to European national clouds and fully on-prem architectures depending on the engagement context.
Azure baseline
Enterprise AI, productivity, and collaboration at scale.
European sovereign
National cloud providers (France, Germany) for residency-sensitive clients.
Client-dedicated
UK-only deployments where cross-border transfer is prohibited.
On-prem sovereign
GPU compute with minimum compliance for restricted workloads.
Match tier to class
Each workload class maps to the lowest-risk tier that still meets operational needs.
Meta-platform: many purpose-built platforms, one interoperability layer
Human and technology platforms that appear independent but share data and integration services.
How to read it
- Platform engineering is not a single platform: purpose-built surfaces share services.
- Mirrors Gotham, Foundry, and Apollo as a meta-platform family.
- GAIA is the sovereign infrastructure layer hosting multiple platforms with controlled interop.
Services, modules, components: a disciplined stack
Shared platform services abstract infrastructure; modules and components carry domain logic.
Platform services
- Relational, document, vector, and graph databases.
- Object storage, email, API gateway, secrets store.
- Modules consume services, not raw infrastructure.
Modules and components
Standard hierarchy enforced across all BOB/ABI deployments: service → module → component → asset.
Semantic spine stays yours; vendors accelerate where they fit
Keep terminology, governance, and classification in-house; use vendors when the path aligns.
Semantic spine (yours)
- Your terminology, procedures, process catalog, and governance rules.
- Data classification policy and routing logic.
- How your organization defines roles, access, and audit requirements.
Accelerators (when they fit)
- Managed data platforms (Databricks, Snowflake) when cloud path aligns.
- Security tools matched to infrastructure (DLP, antivirus, identity).
- Upfront fit analysis: vendor highway vs. your national road constraints.
Highway vs. national road
Vendors bring speed and depth, but each carries architecture constraints. A product that only runs on Azure is the wrong choice for a sovereign on-prem path. Analysis before commitment, not after.
BOB current state: 3 interconnected modules in research preview
Three modules sharing a semantic spine on GAIA sovereign infrastructure. Each at a different technology readiness level.
Market Intelligence
Generate structured intelligence on target organizations. Company positioning, market dynamics, competitive landscape, and key strategic challenges. Usable prototype in operational environment.
Relationship Intelligence
Reveal strategic connections across the firm. Who knows who, surface prior interactions, turn relationship knowledge into firm-wide intelligence. Technology validated in beta.
Offering & Engagement Intelligence
Transform proposals into reusable assets. Interactive proposals and client engagement tracking. Every engagement becomes reusable firm intelligence. Prototype deployed in operational environment.
An open-source alternative to Palantir
BOB is built on ABI, an open-source agentic intelligence platform. Jeremy Ravenel, CEO of Naas.ai and Senior Advisor to the Forvis Mazars board, bridges both organisations: Naas.ai is the technology partner, Forvis Mazars is a contributor and design partner applying the platform internally as proof of practice. The project positions an open, ontology-grounded, sovereign-ready alternative to proprietary platforms like Palantir, running on GAIA compute and grounded in BFO/CCO for interoperability and auditability.
BOB target state: from intent to outcome
The semantic spine connects what the firm knows to what it does. Data flows in, governed intelligence flows out.
Client knowledge graph
Engagement data, service offering, market intelligence, and industry trends unified through a single ontology-grounded knowledge graph.
Sovereign AI in production
Proactive business opportunities, cross-selling signals, and benchmarking. All auditable and reusable across engagements.
The seven BFO categories: a chart of accounts for AI
BFO operationalizes ISO/IEC 21838-2:2021 the way auditors operationalize accounting standards, turning a standard into a working guidance framework. The seven categories are not theory: they are the foundation of Trusted Intelligence: a target state where AI Assurance becomes as significant to organisations as Financial Audit Assurance, giving every intelligent system a shared, auditable structure for what it knows, who acts, when, where, and why.
Continuants (analog to assets)
- Material Entity (WHO): organizations, teams, systems
- Site (WHERE): infrastructure locations and zones
- Quality (HOW IT IS): states, levels, classifications
- Information (HOW TO KNOW): knowledge assets, documents
- Role / Realizable (WHY): roles, mandates, dispositions
Occurrents (analog to liabilities)
- Process (WHAT): macro, micro, nano (activities, workflows, engagements)
- Temporal Region (WHEN): timeframes, cycles, deadlines
Dedicated instances, GPU scale, and vendor negotiation leverage
GPU investments create leverage; the risk is service parity, not access.
Major data platform vendors will deploy dedicated government instances when the commercial case is strong. The UAE's GPU investments create that leverage. The critical question is not whether vendors will come, but whether dedicated instances deliver the same service quality, feature parity, and support as their global deployments.
What to watch
- Dedicated UAE instances may have different SLAs than worldwide US deployments.
- Enterprise contracts sometimes lag individual-tier feature availability.
What to negotiate
- Negotiate outcomes and parity, not just procurement volume.
- Map vendor critical path against sovereignty and compliance requirements early.
Our sovereignty evolution: a lesson we share openly
Sovereignty requirements evolve; architecture must be reassessed when they change.
Forvis Mazars did not start with a perfect sovereign architecture. We operated owned physical servers, moved to European colocation for scale, and later identified connectivity and vendor-path risks that our initial assessment missed. We are now closing those gaps with European hosting partners. This transparency is intentional: it reflects how sovereignty requirements actually evolve in practice.
Verifiable properties
- Sovereignty is a percentage and a set of verifiable properties, not a marketing term.
- Colocation in-region does not automatically mean data never crosses borders.
Re-assess when requirements change
- GPU location, vendor APIs, and backup paths all affect the real posture.
- Architecture is not set-and-forget.
Dedicated instances: leverage vs. service-level parity
GPU scale creates negotiation leverage; parity risk is feature lag and weaker SLAs.
UAE GPU investments create strong vendor negotiation leverage. Major platforms will deploy dedicated government instances. The risk is not access, it is parity: dedicated deployments may lag global feature sets, enterprise contracts may restrict capabilities vs. individual tiers, and SLAs may differ from worldwide US ops.
Before signing
- Benchmark dedicated instance capabilities against global deployment.
- Map critical-path vendors early against sovereignty requirements.
Contract outcomes
- Negotiate feature parity and SLA outcomes, not just spend volume.
- Plan fallback or hybrid paths when vendor constraints block architecture.
Questions for us to explore together
Workloads, GPU posture, sovereignty audit, platform investments, and ABAC maturity.
Which workloads must remain fully air-gapped vs. hybrid sovereign?
What is your current GPU and colocation posture? Dedicated instances planned?
How do you define and audit sovereignty today: residency, connectivity, vendor paths?
Existing data platform investments (Databricks, Snowflake, national cloud) to integrate?
Process catalog and ABAC maturity: ontology-driven or role-based today?
Priority use case for a first 90-day governed AI proof point?
Strategic intelligence follows
Cross-sector evidence on why semantic infrastructure matters for sovereign AI at scale. Jump to slide 32 or use the section menu in the top bar.
Why semantic infrastructure matters for sovereign AI at scale
Cross-sector evidence on governed meaning, knowledge graphs, and durable semantic assets.
Semantic Infrastructure & Knowledge Graphs
Cross-sector evidence on why governed meaning matters for AI-ready government at scale.
UAE government digital programs already accumulate data platforms, AI pilots, analytics dashboards, and automation workflows. The strategic issue is coherence: leadership needs to see why governed meaning matters, where it creates measurable value, and how it fits across ministries, data products, sovereign AI, and citizen services.
Forvis Mazars draws on cross-sector semantic infrastructure work with healthcare, defense, finance, and energy organizations. The pattern is consistent: when decisions are costly, regulated, and shared across systems, formal semantics become infrastructure. This deck places the UAE government question in that broader market context.
- Cross-sector evidence from 25 organization profiles and 26 public source notes.
- Healthcare, defense, aerospace, finance, energy, and internet platform reference cases.
- Semantic layer, knowledge catalog, knowledge graph, ontology, and agent context patterns.
- 25 strategic intelligence organization profiles with linked evidence notes in the portal.
- Evidence notes, organization profiles, and benchmark patterns available in the UAE gov portal workstream folder.
This section is the strategic evidence layer. Use it to place the UAE government question in a broader market context: control of meaning is becoming a strategic capability for AI-ready organizations, not a specialist modeling topic.
The most advanced organizations are converging toward the same pattern: models change, platforms change, but semantic assets become durable.
Systems do not converge by themselves
Operational systems, documents, dashboards, copilots, and data platforms keep multiplying. The result is more access to data, not necessarily more shared understanding.
Meaning becomes reusable infrastructure
Healthcare, defense, finance, retail, cloud, and energy examples show the same move: controlled terms become ontologies, then knowledge graphs, then governed workflows.
Representation may beat model choice
The next advantage may not come from having a better AI model. It may come from having a better representation of operational reality.
Core message
Strategic semantic infrastructure lets an enterprise change AI models, platforms, and applications while preserving the governed meaning those systems depend on.
AI makes knowledge easier to generate, but harder to organize.
More output, less coherence
Organizations are accumulating documents, reports, dashboards, copilots, agents, vector indexes, and generated summaries without a shared model of the things those systems talk about.
Systems need shared entities
AI workflows need stable understanding of people, products, locations, processes, regulations, assets, events, evidence, and decisions.
What increases
- Content volume.
- Model experimentation.
- Platform choices.
- Agent workflows.
What does not automatically increase
- Shared definitions.
- Decision traceability.
- Governed mappings.
- Cross-platform meaning.
Result
More AI does not necessarily create more coherence. Semantic infrastructure is the control layer that keeps AI grounded in the same reality as operations.
Search interest for ontology AI is moving from niche to visible demand.
Executive readout
- Both terms stayed near zero through 2024, then accelerated sharply from mid-2025.
- By May 2026, forward deployed engineer reaches the Trends maximum of 100, while ontology AI reaches 66.
- The signal is not mature adoption. It is market attention moving toward ontology-backed AI delivery and embedded implementation roles.
Implication: semantic infrastructure is becoming part of the AI operating model, not a specialist data architecture topic.
The market uses similar words for different layers. The winners increasingly combine them.
Analytics
Defines business metrics, reporting logic, dimensions, and reusable calculation meaning.
Discovery
Documents data assets, ownership, lineage, quality, access, and governance responsibilities.
Relationships
Connects entities, events, documents, and facts so systems can traverse operational context.
Meaning
Defines entity types, relations, identifiers, definitions, constraints, rules, and intended interpretation.
Grounding
Gives AI agents approved concepts, context, provenance, and action boundaries.
Observation
The strategic infrastructure is not one tool. It is the governed connection between meaning, evidence, systems, and AI execution.
Semantic authority is not the same as platform capability.
What platforms can do
- Integrate data.
- Visualize workflows.
- Query graphs.
- Automate actions.
- Support AI search and agents.
What the enterprise must own
- Authoritative identifiers.
- Approved definitions.
- Relations and constraints.
- Mappings and provenance.
- Validation and inference expectations.
Internal strategic assessment
A platform object model, property graph, schema, dashboard, or AI summary may support operations. It does not automatically prove that enterprise meaning is preserved, governed, exportable, or reusable outside the platform.
Decision rule
Operational acceptance should not be confused with semantic acceptance.
Evaluate semantic claims by evidence, not terminology.
What model is authoritative?
There must be a designated model for enterprise meaning, with ownership, scope, and version control.
Can it leave the platform?
Definitions, identifiers, relations, constraints, mappings, and provenance must be inspectable, exportable, reconstructable, and reusable.
What semantic loss is accepted?
Derived products may omit detail when needed. They should not contradict, redefine, or silently drift from approved meaning.
Executive test
If the enterprise cannot inspect, govern, validate, export, and reuse its meaning outside a vendor system, it does not yet control its semantic infrastructure.
Semantic lock-in is the next vendor lock-in.
What leaders already recognize
- Infrastructure dependency.
- Data residency.
- API coupling.
- License and switching cost.
What AI makes more dangerous
- Meaning embedded in object models.
- Relationships hidden in platform logic.
- Rules buried in workflows.
- Agent context controlled by vendors.
Why it matters
If enterprise meaning exists only inside a platform, future migration becomes semantic rework. AI increases the risk because agents act on hidden context, not only visible data.
Strategic position
Vendors should compete to operationalize meaning, not to own or redefine it.
The same lesson appears across sectors: durable meaning outlasts systems.
SNOMED
Clinical interoperability requires shared definitions. The ontology becomes infrastructure, not an application.
OBO Foundry
Federated ontologies let research organizations collaborate without redesigning meaning.
Mission models
Interoperability starts with common operational understanding, not APIs alone.
Entity-centric systems outperform document-centric systems for search, assistants, and recommendations.
AWS / Linux
Standard infrastructure layers become reusable foundations. Semantic assets may follow the same path.
Strategic lesson
Organizations that control their semantic infrastructure are better positioned to change models, adopt platforms, integrate acquisitions, comply with regulation, and preserve institutional knowledge.
Semantic infrastructure connects systems, data, meaning, graph memory, and AI execution.
How to read it
- Data products make evidence reusable, but the ontology defines what that evidence means.
- The knowledge graph connects approved meaning to operational memory: entities, events, provenance, and context.
- Agents consume governed context, execute workflows, and leave an audit trail back to operations.
Strategic point: platform choice matters less than control of the semantic contract.
The organization evidence points to the same architecture pattern.
Palantir TechnologiesEnterprise, defense, operations
OBO FoundryLife sciences ontology ecosystem
Department of War / DoD and ICDefense ontology standards
U.S. Customs and Border Protection / CBPBorder operations ontology
National Institutes of Health / NCBO BioPortalBiomedical infrastructure
NATO research communityDefense interoperability
GoogleSearch and AI infrastructure
MicrosoftEnterprise data and AI
Amazon Web ServicesCloud graph infrastructure
IKEAConsumer goods, retail, and digital experience
EDM Council FIBOFinance
Goldman Sachs / FINOS LegendFinance
JPMorgan ChaseFinance
Gene Ontology ConsortiumLife sciences
Open PHACTSLife sciences / pharma
AstraZenecaLife sciences / pharma
RocheLife sciences / pharma
NovartisLife sciences / pharma
PfizerLife sciences / pharma
DARPADefense research
BoeingAerospace
Airbus SkywiseAerospace
The Open Group OSDUEnergy
EquinorEnergyPalantir Technologies
Probably the most visible commercial success story for ontology as operating architecture.
Signal
Palantir is the clearest commercial proof that ontology can become operating architecture, not a documentation layer.
Key points
- Foundry, Gotham, and AIP center work around object types, properties, links, actions, functions, and security controls.
- The ontology connects data, logic, decisions, and system write-back so users and agents operate on shared business objects.
- Its market message turns semantic structure into an operating system for decisions, not a back-office data model.
Evidence
Palantir describes the Ontology as the system at the heart of its architecture and defines core concepts for objects, links, actions, functions, and operational workflows.
Strategic takeaway
Decision systems need a governed model of the business before AI agents can act with confidence.
OBO Foundry
The gold standard for open, interoperable biomedical ontologies.
Signal
OBO shows that ontology value depends as much on governance and reuse discipline as on formal modeling.
Key points
- The Foundry coordinates interoperable biomedical ontologies under shared design principles and community review.
- Ontologies such as ECO encode evidence and conclusions so research claims can be connected and reused.
- The model proves that federated domains can share meaning without collapsing into one central application.
Evidence
OBO publishes open ontology principles and reusable biomedical ontologies, including the Evidence and Conclusion Ontology for representing evidence-backed scientific assertions.
Strategic takeaway
Mature semantic programs need operating rules: ownership, design principles, reuse, versioning, and review.
United States Department of War / DoD and Intelligence Community
A public signal that formal ontology is becoming mission interoperability infrastructure.
Signal
Defense adoption signals that formal ontology is becoming mission interoperability infrastructure.
Key points
- Public reporting states that BFO and CCO were selected as baseline standards for formal ontology work.
- BFO provides top-level categories, while CCO adds reusable mid-level classes for mission-specific domains.
- The target is data sharing, federated search, analytic efficiency, and interoperability across complex mission systems.
Evidence
Public sources describe Basic Formal Ontology and Common Core Ontologies as baseline standards for DoD and Intelligence Community ontology work and as a shared foundation for mission data integration.
Strategic takeaway
Standards have to precede scale when many systems must coordinate under high operational risk.
U.S. Customs and Border Protection / CBP
CBP is building ontology-backed knowledge graphs for live border operations, not just exchange standards.
Signal
CBP proves that ontology can become mission infrastructure when response windows are measured in minutes and many sensor feeds must fuse into one operational picture.
Key points
- Border Patrol leadership is leading an ontology development effort across CBP and DHS for enterprise-wide data integration and knowledge sharing.
- A proof of concept used RDF triples linking sensors, acts of sensing, and unmanned aerial vehicles for real-time mapped airspace during drone interdiction.
- A 30-day test collected 17,000 records meeting time and space criteria for qualified interdiction scenarios at the border.
Evidence
Public reporting from the Data Centric Architecture Forum describes CBP's ontology work and the sensor-to-drone RDF pattern used to give field agents precise, mapped airspace information under operational pressure.
Strategic takeaway
Semantic models pay off when many feeds must fuse into one decision picture and the cost of ambiguity is measured in missed response windows.
National Institutes of Health / NCBO BioPortal
NIH-backed infrastructure has made biomedical ontologies searchable, reusable, and queryable at scale.
Signal
BioPortal shows how ontology becomes useful when it is searchable, mapped, queryable, and exposed as infrastructure.
Key points
- BioPortal aggregates biomedical ontologies in a common repository for lookup, annotation, mapping, and reuse.
- It exposes APIs, mappings, and SPARQL access so ontology assets can feed applications and analysis workflows.
- The case proves that shared meaning needs service interfaces, not only files and stewardship documents.
Evidence
NCBO BioPortal describes itself as a comprehensive repository of biomedical ontologies, with services for search, annotation, mappings, APIs, and SPARQL access.
Strategic takeaway
Semantic assets gain strategic value when they are consumable by platforms, products, analysts, and AI workflows.
NATO research community
NATO research shows ontology as an interoperability tool for coalition command and control.
Signal
NATO research makes the interoperability problem explicit: coalition systems need semantic mediation, not just connectivity.
Key points
- IST-075 and IST-094 explored ontology-based semantic interoperability for heterogeneous command-and-control systems.
- The work covers mediation, model harmonization, service discovery, and translation across tactical abstractions.
- It proves that common meaning matters most when multiple actors must coordinate without one canonical system.
Evidence
NATO research papers document ontology-based approaches for semantic interoperability and tactical service discovery across military networks and command-and-control environments.
Strategic takeaway
Federated operations need a shared semantic layer so local systems can remain different while decisions remain coherent.
Google made knowledge graphs mainstream for entity-centric search.
Signal
Google made the strategic shift visible: understanding entities and relationships beats matching strings.
Key points
- The Knowledge Graph models real-world things and relationships to improve search, panels, answers, and disambiguation.
- It moves retrieval from document-centric matching toward entity-centric understanding.
- The same logic applies inside enterprises where operational entities are scattered across documents and systems.
Evidence
Google introduced the Knowledge Graph as an intelligent model of real-world entities and relationships, commonly summarized as a move toward things, not strings.
Strategic takeaway
AI search and assistants need durable entity understanding before they can produce reliable operational answers.
Microsoft
Microsoft is now making ontology a native enterprise data product concept inside Fabric.
Signal
Microsoft bringing ontology into Fabric shows that enterprise platforms are mainstreaming semantic assets.
Key points
- Fabric IQ positions ontology as an enterprise vocabulary and semantic layer across domains and OneLake sources.
- Documentation covers entity types, properties, relationships, data bindings, graph views, and agent-consumable concepts.
- The platform can generate ontology items from Power BI semantic models, making semantic modeling part of the analytics stack.
Evidence
Microsoft Fabric documentation describes ontology as a way to unify enterprise meaning and generate ontology concepts from existing semantic models.
Strategic takeaway
When platforms consume ontology natively, the strategic question becomes who governs the meaning those platforms inherit.
Amazon Web Services
AWS provides the infrastructure layer for RDF, SPARQL, and enterprise knowledge graphs through Neptune.
Signal
AWS validates graph infrastructure, while also showing that storage capability is not the same as semantic authority.
Key points
- Amazon Neptune supports RDF and SPARQL for knowledge graph workloads alongside property graph use cases.
- AWS guidance shows OWL ontologies loaded into Neptune and paired with reasoning engines such as RDFox.
- The infrastructure enables graph operations, but governance of concepts, constraints, and inference remains an enterprise responsibility.
Evidence
Amazon Neptune documentation covers SPARQL access, and AWS guidance explains how OWL ontologies can be used in model-driven graphs on Neptune.
Strategic takeaway
Cloud graph services can host semantic assets, but they do not decide which meanings are authoritative.
IKEA
A rare public consumer-goods example where ontology and knowledge graph practice directly support product discovery, recommendations, and customer experience.
Signal
IKEA shows that ontology can support commercial experience, not only regulated or scientific domains.
Key points
- Public material describes a three-layer knowledge graph: concepts, categories, and product data.
- The graph supports product discovery, recommendations, search, navigation, APIs, and explainable customer experiences.
- The case proves that semantic structure helps when product context and business rules must travel across channels.
Evidence
IKEA Knowledge Hub and public talks describe how concept, category, and data layers support recommendations and digital experience use cases.
Strategic takeaway
Semantic models become business infrastructure when recommendations and decisions must be explainable, reusable, and channel-independent.
EDM Council FIBO
The most important open financial ontology standard.
Signal
FIBO shows how formal meaning becomes a risk and reporting control in a regulated industry.
Key points
- FIBO defines financial instruments, contracts, entities, indices, derivatives, and regulatory concepts in formal ontology assets.
- The model uses OWL and Description Logic with industry review and standards alignment.
- It proves that shared definitions can reduce ambiguity in reporting, lineage, risk, and entity resolution.
Evidence
EDM Council and FIBO specification pages describe FIBO as a formal financial industry ontology for business concepts and relationships.
Strategic takeaway
Where ambiguity creates financial or regulatory exposure, business meaning needs the same rigor as data quality.
Goldman Sachs / FINOS Legend
A rare public example of a major bank open-sourcing its internal data modeling and governance platform.
Signal
Legend shows that regulated enterprises need business meaning to be modeled, governed, and executable by engineering teams.
Key points
- Goldman Sachs open-sourced Legend through FINOS as a modeling, governance, lineage, and collaborative data platform.
- Legend combines visual modeling, mappings, execution, SDLC, and review workflows for financial data products.
- The case connects semantic modeling directly to delivery discipline, not only architecture diagrams.
Evidence
FINOS public materials describe the Legend case study and Goldman Sachs decision to open-source its data modeling platform through FINOS.
Strategic takeaway
Semantic governance scales when models are versioned, reviewed, mapped, and executable in the delivery workflow.
JPMorgan Chase
A public example of enterprise knowledge graphs used for mission-critical financial applications.
Signal
JPMorgan Chase shows that entity resolution becomes core infrastructure when decisions depend on noisy text and internal identifiers.
Key points
- Publications describe knowledge graphs used for risk assessment, fraud detection, investment advice, and financial news linking.
- JEL links company mentions in text to entities in an enterprise company knowledge graph.
- The case proves that graphs are valuable when they connect unstructured evidence to governed enterprise entities.
Evidence
JPMorgan Chase publications on JEL describe financial-news entity linking against a company knowledge graph and broader enterprise knowledge graph applications.
Strategic takeaway
High-value decisions require entity resolution that connects documents, events, and enterprise-specific objects.
Gene Ontology Consortium
The early proof that scientific knowledge needs shared computable meaning.
Signal
Gene Ontology is the early proof that scientific knowledge needs shared computable meaning before large-scale analysis works.
Key points
- GO provides controlled terms for gene functions, biological processes, and cellular components across organisms.
- The knowledgebase combines evidence-supported annotations with GO-CAM models that connect annotations into pathways.
- The case established the pattern of identifiers, evidence, relations, and curation before analytics.
Evidence
Gene Ontology public resources and the 2023 knowledgebase paper describe a computable structure for gene function, evidence-backed annotations, and biological models.
Strategic takeaway
Reusable knowledge depends on stable identifiers, evidence codes, relationships, and sustained expert curation.
Open PHACTS
A landmark public-private semantic web initiative for drug discovery.
Signal
Open PHACTS showed that drug discovery needs answerable cross-domain questions, not isolated datasets.
Key points
- The initiative integrated compounds, targets, diseases, tissues, pathways, and public drug-discovery databases.
- It used semantic web technology, APIs, and query mechanisms to help researchers traverse related evidence.
- The case demonstrates why domain questions should drive graph design.
Evidence
IHI project materials and the Open PHACTS triple store paper describe a semantic platform for integrating pharmacological data and answering complex drug discovery questions.
Strategic takeaway
Semantic infrastructure should be judged by the cross-domain questions it makes answerable.
AstraZeneca
A strong public example of an internal pharma knowledge graph for drug development.
Signal
AstraZeneca shows how competitive knowledge graphs combine public science, internal evidence, and machine learning readiness.
Key points
- BIKG integrates public, licensed, proprietary, and literature-extracted biological data for drug development.
- The graph connects genes, proteins, diseases, compounds, NLP pipelines, and ontology alignment.
- Kazu documentation shows the adjacent need for entity recognition and normalization in scientific text.
Evidence
AstraZeneca public materials describe BIKG as an internal Biological Insights Knowledge Graph and Kazu as tooling for biomedical entity recognition and normalization.
Strategic takeaway
The strongest knowledge assets connect external evidence, internal data, text extraction, and governed graph structure.
Roche
A mature pharma example of FAIR data, semantic hubs, and ontology-backed interoperability.
Signal
Roche shows that ontology can become data-governance plumbing for large scientific operations.
Key points
- Public sources describe RDF, RDFS, OWL, community vocabularies, BFO-aligned models, and semantic harmonization.
- The FAIR approach connects metadata, terminologies, domain ontologies, application models, and productive applications.
- The case proves that interoperability requires design discipline before data is consumed by downstream platforms.
Evidence
Roche FAIR data by design material and the FAIR in vivo platform paper describe ontology-backed data harmonization for life sciences knowledge graphs.
Strategic takeaway
FAIR data becomes operational only when metadata, terms, models, applications, and governance are designed together.
Novartis
A visible drug-discovery knowledge graph case using biomedical entities and literature evidence.
Signal
Novartis is a clean public example of experts traversing evidence relationships instead of searching isolated repositories.
Key points
- Public case studies describe a graph connecting genes, diseases, compounds, text-mined literature, historical data, and image-derived data.
- Researchers use the graph to evaluate relationship strength and identify promising compounds or disease hypotheses.
- OntoBrowser adds evidence of ontology tooling around open-source science and browsing.
Evidence
Neo4j and Novartis public sources describe drug discovery knowledge graph use cases and the OntoBrowser project for working with biomedical ontologies.
Strategic takeaway
Experts need connected evidence paths that reveal why an answer is plausible, not only where a document is stored.
Pfizer
A visible pharma example of semantic integration, ontologies, and knowledge graph AI.
Signal
Pfizer reinforces the pattern: the higher the cost of scientific ambiguity, the more valuable governed semantic integration becomes.
Key points
- Public sources describe data standards, vocabularies, ontologies, linked data, and an intelligent data framework.
- Recent public material describes biomedical knowledge graphs with Data4Cure for continuously updated scientific insight.
- The work connects literature, identifiers, compounds, disease areas, public data, and internal data for drug discovery.
Evidence
Bio-IT World and Drug Discovery Online sources describe Pfizer semantic integration work and knowledge graph collaboration for AI and data-driven discovery.
Strategic takeaway
Semantic integration matters most when evidence is fragmented and the cost of a wrong interpretation is high.
DARPA
The historical bridge between early semantic web research and defense needs.
Signal
DARPA explains why defense communities saw semantic interoperability early: autonomous systems need explicit meaning.
Key points
- DAML aimed to make web information machine-readable through semantic annotations and ontologies.
- The program created transition paths to military command-and-control and intelligence activities.
- DAML and DAML+OIL influenced OWL and later ontology and knowledge graph standards.
Evidence
DAML.org and the DAML BAA describe early semantic web work focused on machine-readable information and defense-relevant transition paths.
Strategic takeaway
Agentic and autonomous systems increase the value of explicit semantics because machines act on representations.
Boeing
A public aerospace example where semantics support model-based systems engineering and lifecycle data.
Signal
Boeing makes the aerospace analogy concrete: lifecycle decisions require shared meaning across parts, requirements, and evidence.
Key points
- Public sources describe semantic capabilities for MBSE, aircraft data hierarchy, impact analysis, and lifecycle consistency.
- The work connects parts, requirements, engineering data, system models, and analysis workflows.
- The case maps closely to asset-intensive operations where versions, safety, and maintenance evidence matter.
Evidence
MarkLogic and Boeing public repositories describe semantic support for model-based systems engineering and aircraft data hierarchy work.
Strategic takeaway
Complex assets need one governed model of reality across design, operations, maintenance, safety, and change impact.
Airbus Skywise
The clearest public aviation platform example of ontology-backed operational data integration.
Signal
Skywise shows ontology becoming operational when it is embedded into fleet, maintenance, and equipment workflows.
Key points
- Airbus launched Skywise with Palantir to integrate aviation data across airlines and operational use cases.
- Developer documentation exposes ontology APIs for aircraft, parts, events, maintenance, and work packs.
- The platform supports predictive maintenance, disruption reduction, fleet operations, and analytics over aviation data.
Evidence
Airbus public launch material and Skywise developer documentation describe an aviation data platform with ontology APIs for operational entities and workflows.
Strategic takeaway
Ontology becomes strategic when it is wired into operational workflows where asset context changes decisions.
The Open Group OSDU
The energy sector's clearest open data-standardization move.
Signal
OSDU is the energy sector proof point that data standardization is now a shared platform concern.
Key points
- OSDU provides an open-source, standards-based, technology-agnostic data platform for energy data.
- Public ontology work converts OSDU schemas into OWL/RDF, moving schema standardization toward formal semantics.
- The case is necessary infrastructure, but it does not replace enterprise-level concept governance.
Evidence
OSDU Forum and OSDU Ontology public materials describe common energy data platform standards and ontology conversion work over OSDU schemas.
Strategic takeaway
Sector standards help align platforms, but enterprises still need to govern the meanings that drive decisions.
Equinor
A public operator example of ontology-based access and contextualized industrial data.
Signal
Equinor shows the energy operator bottleneck clearly: finding trusted data can be harder than analyzing it.
Key points
- Public research describes ontology-based data access for exploration data at Equinor.
- OmniaPlant principles publish contextualized industrial data through open APIs for plant and operational data.
- The work connects geologists, timeseries metadata, plant context, and industrial applications.
Evidence
SINTEF publication material and Equinor OmniaPlant describe ontology-based data access and contextualized industrial data APIs.
Strategic takeaway
Discovery improves when operational data is exposed through governed context, not only stored in more repositories.
Cognite
A market-facing industrial knowledge graph platform used in energy and process industries.
Signal
Cognite validates industrial knowledge graphs as a market category, especially for asset-intensive operations.
Key points
- CDF contextualizes asset, timeseries, document, 3D, maintenance, and process data.
- Cognite publishes core and process industry data models for assets, equipment, timeseries, maintenance orders, and notifications.
- The case proves industrial platforms want semantic structure, while formal authority still has to span all platforms.
Evidence
Cognite public materials describe an industrial knowledge graph and process industry data models for contextualized asset and operations data.
Strategic takeaway
Industrial platforms can operationalize knowledge graphs, but enterprise meaning should remain governed above any single platform.
The external evidence supports a governed semantic spine as strategic control infrastructure.
Own meaning once
The government semantic layer should define governed concepts that data platforms, analytics tools, sovereign AI systems, and agents consume. Otherwise, each platform rebuilds meaning locally.
Anchor in decisions
The message should start with operational questions: service delivery, policy compliance, citizen outcomes, security risk, budget impact, and explainable AI.
Make evidence traceable
Every concept should carry ownership, source evidence, lifecycle status, mapping rules, and validation expectations.
Stay platform-neutral
The semantic layer should feed platforms without becoming captive to one vendor's object model, graph model, or AI interface.
Executive sentence
Ontologies and semantic knowledge graphs are the control system for enterprise meaning: they let operational, data, and AI platforms act on the same reality.
The pattern across 25 organisations is unambiguous: costly shared decisions require semantic infrastructure.
Twenty-five organization profiles across healthcare, defense, aerospace, energy platforms, internet infrastructure, open standards, and manufacturing have produced a consistent pattern. When decisions are expensive and interpretations must be shared across systems, people, and time, semantic infrastructure is built. Roche, Novartis, and Pfizer built it because drug discovery errors are fatal. Boeing and Airbus built it because lifecycle asset decisions cannot tolerate ambiguity at scale. Google, Amazon, and Microsoft embedded it because search, commerce, and AI cannot function without shared meaning. DARPA funded it because autonomous systems act on representations, not on ambiguous data. The convergence is cross-sector and the direction is clear.
The implication for UAE government is specific. National digital programs operate at the same scale and decision complexity as the organizations in this deck. Policy execution, citizen services, sovereign AI, cross-ministry coordination, and regulated data sharing all carry the same characteristics: fragmented evidence, high cost of interpretation errors, and decisions that must be traceable across time. The question is not whether semantic infrastructure matters. It is how quickly to make governed meaning operational across platforms and AI workflows.
- Healthcare, defense, aerospace, and internet platforms have all converged on semantic infrastructure independently, validating the direction.
- The pattern is consistent: when interpretations are shared, costly, and mission-critical, formal semantics pays.
- Energy and industrial operators including Equinor are making the same move; government programs can learn from their patterns.
- Early investment in semantic governance creates durable advantage as AI platforms multiply.
- Operational adoption, not technical design, is the remaining challenge and the next decision.
The next step is to connect this external evidence to the UAE digital transformation agenda: use peer patterns as executive context for data platform and sovereign AI decisions, and make cross-sector validation visible in leadership conversations.
The deck is backed by one public-evidence note per organization.
HTML and PPTX
Organization profiles and evidence notes sit in the UAE gov portal workstream ws5 folder.
One master file
Public source notes are retained internally for every organization profile in this evidence section.
26 files
Detailed source notes are retained internally for every organization profile in this section.
Use in the visit narrative
This evidence layer gives external legitimacy: mature governments and enterprises converge on semantic infrastructure when interpretation errors are costly and AI scale demands shared meaning.