Enterprise AI Without the Hype: Part 5 – Integration and Infrastructure Architecture

This is part 5 of a blog series on where AI is already creating real value across the different domains of enterprise architecture. The previous articles covered business, process, application, and data and analytics architecture. This final part focuses on the integration and infrastructure foundations that allow those capabilities to operate reliably.

Understanding the role of Integration and Infrastructure Architecture

Integration and infrastructure architecture rarely features prominently in business strategy presentations, but it determines whether AI workloads can run at scale, whether data can move safely between systems, and whether security and governance can keep pace with new demands.

In this final part of the series, we show where AI is changing what enterprise foundations need to deliver, what those changes can cost, and why experienced people remain essential at this technical layer.

Integration and infrastructure: the foundation AI asks more from

AI workloads place different demands on enterprise infrastructure than traditional ERP and analytics workloads. Data transfer volumes can be higher, compute requirements may spike at specific points in a planning or analysis cycle, and response times can become more sensitive to infrastructure performance.

Security architecture also needs to account for model access controls, prompt injection, and the possibility of sensitive data being exposed through external AI services.

For business leaders, the practical implication is clear. Investing in AI at the application or process layer without strengthening the integration and infrastructure beneath it creates operational and security risk. Companies that generate sustained value from AI tend to include these foundations in the AI program from the start rather than treating them as a separate technical workstream.

Where AI gets real within integration and infrastructure architecture

The following examples are drawn from official SAP sources, company disclosures, and published industry analyses. They illustrate what AI-enabled integration and infrastructure transformation looks like when it is operating in production.

Across Pfizer, Shell, and BMW, the pattern is consistent. Predictive maintenance creates operational value when it is supported by reliable, real-time integration and scalable data pipelines.

Pfizer’s PACT program uses IoT sensor data and AWS machine learning to predict equipment failures in continuous manufacturing. This shows that reducing unplanned downtime depends on reliable sensor-to-cloud integration as well as the model itself.

Shell applies AI across more than 10,000 refinery assets and approximately 20 billion data points per week. At that scale, predictive maintenance requires resilient integration, continuous monitoring, and sufficient compute capacity.

BMW’s use of AI-driven predictive maintenance on conveyor systems reinforces the same principle in high-velocity manufacturing. Preventing stoppages becomes valuable when equipment data reaches the analytics environment reliably and quickly enough to support intervention.

Vorwerk and Al Ghurair Iron and Steel illustrate the integration side of the same story. AI becomes easier to scale and maintain when the architecture connects legacy and cloud systems consistently across the enterprise.

Vorwerk used SAP BTP and SAP Integration Suite to connect SAP and third-party back-end systems across its global operations. AI-assisted API mapping and self-healing integration reduced effort, lowered dependence on specialist developers, and helped prevent upstream data quality issues.

Al Ghurair Iron and Steel used SAP BTP to replicate an AI-enabled production planning solution across multiple manufacturing locations rather than leaving it as a one-site pilot. This shows how reusable integration patterns can turn a local use case into an enterprise capability.

Together, these examples show that AI value at this layer depends on infrastructure designed for reliability, reuse, security, and scale.

What business leaders can use AI to do at this layer:

• Use AI-assisted API mapping and integration documentation in SAP BTP and SAP Integration Suite to reduce the time and cost of building connections between SAP and non-SAP systems during migrations and transformations.
• Use anomaly detection on integration event streams to identify data quality failures before they reach planning systems and analytics models.
• Use predictive infrastructure scaling to manage AI compute demand during peak periods, such as S&OP processing or month-end analysis, without maintaining the same capacity throughout the month.
• Use AI-driven predictive maintenance on critical operational assets to identify failure risks earlier and reduce unplanned downtime.
• Use security AI to detect unusual access patterns as copilots and agents gain access to ERP and operational data.

What AI can cost at this level

Computing power can become one of the largest items in an enterprise AI infrastructure budget. Running a private large language model on premises or in a dedicated cloud environment for a mid-sized pharma or CPG company can cost €300,000 to €1 million annually in compute alone.

Commercial AI services can create a more variable cost profile, but they may be easier to monitor and control through consumption limits and usage reporting. The final cost depends on model choice, transaction volume, performance requirements, data sensitivity, and contractual arrangements.

AI functionality in integration platforms can add approximately 15-25% to existing integration platform license costs. Network and security upgrades, including zero-trust controls for model access, AI gateways, and prompt-injection protection, can add €100,000 to €400,000 in project and ongoing license costs.

Organizations also need financial operations, or FinOps, capabilities to monitor and govern cloud AI spending. This operating cost is often underestimated during the first year of an enterprise AI program.

SAP has cited a Forrester Consulting study reporting a 345% return on investment and a 30% increase in integration efficiency for SAP Integration Suite customers. This provides one benchmark for organizations considering whether stronger integration foundations should form part of the AI business case.

These figures are directional. Actual costs depend heavily on the existing architecture, cloud agreements, level of customization, security requirements, and scale of deployment.

The role of people alongside AI

Integration and infrastructure architecture is one of the most specialist areas of enterprise IT. AI can reduce effort in routine activities such as documentation, monitoring, and basic configuration, while creating demand for expertise in new areas.

MLOps engineers monitor model performance and drift. AI security architects manage risks such as prompt injection, excessive data access, and unintended information exposure. FinOps practitioners track consumption and help keep cloud AI spending aligned with the business case.

These roles add capabilities to the existing infrastructure organization. During the first years of an enterprise AI program, many companies will need more specialist capacity to design, secure, monitor, and govern the environment effectively.

Trying to operate AI at scale with the same team structure and skills used before AI can create operational gaps. Sustainable adoption requires clear ownership, new capabilities, and continued investment in the people who understand how the systems connect.

How Bluecrux supports AI in integration and infrastructure architecture

As an SAP Business Integrator, Bluecrux helps clients build the integration and infrastructure foundations that make AI reliable, secure, and governable across the enterprise. This includes SAP BTP integration design for AI-enabled transformations, API management across SAP and non-SAP system landscapes, and security architecture that reflects the expanded risks introduced by AI workloads.

We bring this experience to clients across pharma, biotech, medtech, CPG, specialty chemicals, and industrial manufacturing. In these industries, infrastructure failures can affect production continuity, product quality, regulatory compliance, and customer or patient outcomes.

Bluecrux also connects these foundations to Axon™, Binocs™, and Helion™, ensuring that the integration architecture beneath our analytics and planning technologies supports the operational demands of the teams using them. The goal is dependable AI that supports connected decisions across planning, manufacturing, quality, and commercial operations.