Today, about 74% of enterprises have AI in production, yet roughly half of them cannot demonstrate clear ROI. The core issue isn’t deployment—it’s measurement and governance.

Three main gaps show up again and again:

• No clear business metrics: AI projects often launch without being tied to specific, named KPIs (e.g., cost per ticket, churn rate, order cycle time). That makes it hard to prove value later.
• Weak integration with core systems: When AI runs in a silo instead of being embedded into core workflows (ERP, CRM, ticketing, etc.), its impact is fragmented and difficult to quantify.
• Limited accountability and governance: Deployment moved faster than governance. Many teams lack a consistent way to track performance, risk, and ownership across AI use cases.

To close the ROI gap, teams are starting to:

• Attach every AI deployment to a business metric before launch—for example, “reduce average handle time by 10%” or “increase self-service resolution by 15%.”
• Instrument the full workflow so you can compare before/after performance, not just model accuracy.
• Review AI initiatives in the same way as other major investments, with regular performance reporting into budget and planning cycles.

The shift underway is from “Are we doing AI?” to “Can we prove it works, and can we explain how it creates value?” Teams that make that shift are the ones able to defend and expand their AI budgets.

Many enterprises are discovering that they’ve effectively “rented” their institutional knowledge to AI platforms they don’t control. As workflows move into vendor models, critical business logic, customer insights, and decision context can end up locked inside someone else’s system.

The main risks are:

• Knowledge lock-in: If your processes and prompts live only inside a vendor’s platform, switching providers can mean rebuilding that knowledge from scratch.
• Unclear data rights: Without explicit terms, it may be ambiguous who owns derived data, fine-tuned models, or logs generated by your usage.
• Opaque model updates: Vendors can update models in ways that change behavior, quality, or compliance posture without you having much say.

To reimagine this relationship, procurement and IT leaders are starting to treat AI vendors like long-term critical infrastructure, similar to ERP or cloud providers. In practice, that means:

• Auditing existing AI contracts for:
  • Data ownership and usage rights
  • Data portability and export formats
  • Audit and transparency rights
  • Exit terms—what happens to your data and any models trained on it if you leave
• Negotiating portability up front so you can move workflows, prompts, and training data if needed.
• Documenting your institutional knowledge outside the vendor platform (e.g., playbooks, prompt libraries, decision trees) so it’s not trapped in one tool.

The goal isn’t to avoid vendors—it’s to avoid a situation where your most valuable operational knowledge is only accessible on someone else’s terms.

Two big shifts are reshaping how enterprises plan AI-related infrastructure and platforms:

1. Industrial AI is being bundled into major OEM platforms

Industrial players like Schneider Electric and Siemens have committed multi-billion-dollar acquisitions to build AI-native capabilities directly into their platforms. For operators, that means:

• AI will increasingly come embedded in automation, building management, and energy systems, not as a separate add-on.
• The AI capabilities you thought you’d source independently in a few years may instead arrive as part of an OEM contract you’re negotiating today.
• Platform dependency can deepen if workflows are tightly coupled to a single vendor’s proprietary AI stack.

Action for your team: when renewing or expanding industrial automation contracts, ask explicitly how AI is bundled, priced, and licensed, and what options you have to control or export the data and logic those AI features generate.

2. Data-center buildout is shifting to new geographies

Community resistance (NIMBY) to data centers in established U.S. markets is pushing hyperscalers to look at more remote industrial land. For example, landowners in the Permian Basin of West Texas are marketing large parcels based on:

• Existing power from oil and gas infrastructure
• Low land costs
• Sparse population density

For enterprise IT and facilities teams, this can affect:

• Latency profiles for cloud regions serving your users
• Power reliability and redundancy for AI-heavy workloads
• Provisioning timelines as capacity comes online in non-traditional locations

Action for your team: include hyperscaler siting and regional capacity plans in your infrastructure reviews, not just as real estate news but as an input into where you place latency-sensitive and AI-intensive workloads.

Taken together, these trends mean AI is becoming a built-in layer of both industrial platforms and cloud infrastructure. Planning ahead now helps you avoid being surprised by lock-in, latency, or licensing constraints later.