74% of enterprises run AI in production, but half can't prove it pays off

Seventy-four percent of enterprises now have AI running in production. That figure, reported by Forbes contributor Sandy Carter, sounds like a victory lap. The problem is that roughly half of those same organizations cannot demonstrate what any of it is worth.

That ROI gap is the defining operational tension of 2026. Companies have moved fast enough to deploy, but governance frameworks, integration pipelines, and measurement disciplines have not kept pace. For CIOs and operations leaders, the conversation is shifting from 'are we doing AI' to 'can we prove it works, and who controls the knowledge it generates.'

The ROI problem no one budgeted for

Carter's reporting in Forbes draws on survey data showing that the production-to-proof gap is now one of the most commonly cited leadership concerns. Deployment happened quickly; accountability structures did not. Leaders are being pushed to focus on governance, tighter integration with core business systems, and building the internal metrics that connect AI activity to tangible outcomes.

A separate Forbes piece by Jason Snyder frames the risk in starker terms: enterprises are ceding control of institutional knowledge to AI platforms they do not own. When critical business logic, customer insight, and decision-making context live inside a vendor's model or platform, the enterprise has effectively rented its cognitive infrastructure. The lease terms, data ownership, model updates, and exit rights, are rarely read before signing.

The practical implication for procurement and IT leaders is concrete: contracts with AI platform vendors need the same scrutiny applied to ERP or cloud deals, covering data portability, audit rights, and what happens to proprietary training data if the relationship ends.

Industrial giants consolidate before the window closes

At the infrastructure layer, the market is moving even faster. Forbes contributor Gaurav Sharma reported that Schneider Electric and Siemens have each committed multi-billion-dollar acquisitions aimed at building AI-native capabilities directly into their platforms. The strategic logic is straightforward: operators who run Schneider or Siemens infrastructure at scale will increasingly find AI embedded in the systems they already buy, rather than bolted on through a separate vendor relationship.

That consolidation pattern matters to procurement teams evaluating industrial automation, building management, and energy management platforms. The AI capability a team plans to source independently in two years may be bundled, and potentially locked in, by the OEM contract being negotiated today. Sharma's reporting characterizes the deal pace as a signal that the window for building AI-native capabilities organically is narrowing; incumbents are buying the capability before smaller specialists can establish independent footholds.

The broader industrial AI market is also drawing attention to platform dependency in a different form. As Snyder noted in Forbes, the risk is not just financial, it is operational: companies that build workflows around a specific AI vendor's proprietary models may find themselves unable to migrate without rebuilding institutional knowledge from scratch.

Data-center siting shifts as community resistance grows

The physical infrastructure underpinning all of this AI deployment is running into its own friction. The Wall Street Journal reported that NIMBY opposition to data centers in established markets around the U.S. is opening a new front in the buildout: remote industrial land. Landowners in the Permian Basin of West Texas are actively marketing large parcels to hyperscalers, positioning the region's sparse population density, available power from existing oil and gas infrastructure, and low land costs as advantages over suburban or exurban sites near major metros.

For facilities and infrastructure leaders at large enterprises, the siting shift carries direct supply-chain implications. If hyperscaler capacity expands in non-traditional geographies, latency profiles, power reliability guarantees, and interconnection timelines for enterprise cloud regions could all change. Evaluating where a hyperscaler's next capacity tranche is actually being built is becoming a relevant question for IT infrastructure planning, not just real estate news.

Reuters separately reported that Apple has published guidance enabling Mac users in mainland China to connect Alibaba's Qwen AI model to Siri and Apple's Writing Tools. The move illustrates how platform-level AI integrations are evolving differently across geographies, a dynamic enterprise IT teams operating across regions will need to track as compliance and data-sovereignty requirements diverge.

What this means for your team

• Audit your AI vendor contracts now for data ownership, model update rights, and exit portability. The knowledge your workflows generate inside a vendor platform may not be yours to take when you leave.
• Tie every production AI deployment to a named business metric before the next budget cycle. The Forbes survey data shows half of enterprises cannot prove ROI; having that measurement in place separates defensible spend from budget exposure.
• When renewing or expanding industrial automation contracts with major OEMs, ask specifically how AI capabilities are being bundled, priced, and licensed. Consolidation by players like Schneider Electric and Siemens means the AI layer may arrive as a platform feature, not a separable line item.
• Include hyperscaler siting and regional capacity timelines in your infrastructure roadmap reviews. Data-center buildout shifting to the Permian Basin and similar locations could affect latency, redundancy, and provisioning lead times for cloud-dependent workloads.