AI is emerging as an operating layer of connected infrastructure

Organizations are shifting from asset optimization to system optimization

For many infrastructure organizations, predictive maintenance was a gateway AI use case and provided the first clear proof of its value. This was not surprising since it offered a clear business case to infrastructure leaders by reducing downtime, extending asset life, and intervening before failures became costly.

As infrastructure becomes more connected, leaders are increasingly looking to move beyond asset-level optimization to more systems-level decision-making by applying AI across entire systems to improve coordination, resilience, and operational performance (figure 1).

What the survey tells us:

• Infrastructure organizations are planning to focus on a broader set of operational use cases over the next three years, including energy demand forecasting (44%), grid management (35%), traffic management and mobility optimization (38%), and AI-enabled workforce optimization (38%).
• The projected decline in predictive maintenance implementation—from 61% to 29%—is best interpreted as the use case maturing and organizations expanding into broader operational intelligence, rather than as waning interest.

The more connected the infrastructure becomes, the more valuable AI becomes in managing it. Over time, this shift could move some infrastructure environments along a maturity curve from “staffed” operations to “assisted,” “dim,” and, in a few cases, “dark” operations (figure 2). This maturity path is unlikely to unfold uniformly across infrastructure environments. However, in controlled, mission-critical environments such as logistics hubs, energy assets, water treatment facilities, tunnels, and ports, it could become a benchmark for operational efficiency, resilience, and workforce safety.

AI moves upstream into infrastructure design and delivery

Some of the biggest barriers to infrastructure delivery are not technical but institutional, including permitting, approvals, compliance, procurement, and resilience planning. They often determine whether a project moves from concept to construction or remains stuck in the pipeline.

To help reduce these barriers, infrastructure leaders are increasingly leveraging AI in their infrastructure delivery processes, deploying it earlier in the infrastructure life cycle across planning, design, compliance, scheduling, and project delivery (figure 3).

What the survey tells us:

• According to survey respondents, as AI capabilities mature, they plan to increase adoption in several higher-value build-stage use cases over the next three years, including streamlining permitting and compliance (from 36% to 42%), supply chain logistics (from 32% to 40%), and weather-related modeling for resilient design (from 32% to 40%).

This marks a shift from using AI primarily to improve operational efficiency toward using it to address institutional friction and externalities. AI may not eliminate regulatory or procurement challenges, but it can improve the speed and quality of the decisions that surround them.

A number of US cities are using AI to streamline permitting and reduce approval times. In Austin, AI-assisted building permit reviews help accelerate processing and allow staff to focus on more complex applications.¹ Boston is taking a different approach: It is using AI to analyze 25 years of permit data and transform unstructured application records into around 260 common permitting scenarios, such as replacing a boiler or building a deck.² For each scenario, the city maps the required permits, expected timelines, and inspections, giving residents and builders a clearer, more predictable path through the permitting process.³

Australia is rethinking building regulations using AI, partnering with states and territories to speed up housing approvals and cut red tape. The effort also includes exploring how smart technology can help builders and other users navigate the country’s three-volume, 2,000-page codebook, and establishing a strike team in the environment department that uses AI to fast-track and streamline housing assessments and approvals.⁴

Managing intelligent infrastructure requires a broader technology foundation

Intelligent infrastructure requires more than AI models. It depends on a broader technology stack: sensors to capture data; connectivity to move it; cloud, edge, and data systems to process it; cybersecurity to protect it; and digital twins and analytics to translate information into better decisions. AI should not be treated as a standalone modernization agenda.

In a connected infrastructure system, the weakest layer of the stack can limit the value of every AI use case it supports. Model performance matters, but so do data quality, interoperability, latency, resilience, and trust.

As these layers become more connected, infrastructure organizations will also need stronger command-center models. The goal is to move from visibility to coordinated action.

Digital twins are one of the clearest examples of this shift because they link the physical asset, the data environment, and AI-enabled decision-making. When powered by AI, they can evolve from static visual representations into living systems of intelligence that simulate design choices, monitor construction progress, predict operational performance, and guide interventions across the life cycle.⁵

Figure 4 shows that infrastructure leaders across regions recognize the breadth of technologies required for modernization. However, differences in regional priorities point to different modernization pathways shaped by geography, digital maturity, energy needs, security concerns, and the condition of existing infrastructure.

What the survey tells us:

• Globally, respondents are looking beyond AI to a wider ecosystem that also includes clean/green energy (47%), data security (39%), batteries and storage (37%), 5G (31%), and cloud computing (28%).
• North American respondents prioritize clean/green energy (55%) and data security technologies (49%), signaling that modernization is as much about resilient, secure infrastructure as it is about AI adoption.
• In addition to clean/green energy, Asia-Pacific, European, and Latin American respondents prioritize battery storage, likely reflecting energy security, supply-chain exposure, and decarbonization priorities.
• Middle East/Africa respondents expect the greatest impact from clean/green energy (48%) and 5G connectivity (43%), reflecting the importance of both energy transition and connectivity foundations.
• APAC (32%) and North American (25%) respondents show a much higher interest in digital twins than respondents from other regions.

AI strategies should therefore account for electricity, water, land, network, and embodied-carbon requirements, as well as the location-specific opportunity costs of allocating scarce infrastructure capacity to computational demand.

Infrastructure leaders are building diversified AI portfolios

Generative AI continues to attract significant attention across almost all regions. But infrastructure leaders are moving beyond a single-technology view of AI. They are assembling portfolios of capabilities that support different needs across the infrastructure life cycle (figure 5).

This points to a more mature strategy for AI adoption. Infrastructure leaders are matching the technology to the problem: using traditional AI for classification and automation; predictive analytics for forecasting; digital twins for life-cycle intelligence; gen AI for knowledge work and decision-making support; and agentic AI for emerging orchestration use cases.

What the survey tells us:

• Globally, 50% of survey respondents selected gen AI in the AI and data technology category to have the largest impact on infrastructure plans over the next three years. The figure is even higher among respondents from Latin America (67%) and Europe (62%).
• The regional differences in interest in gen AI can be indicative of the technology’s maturity level. For instance, respondents from North America, a region with deeper expertise and know-how of AI,⁶ show a much more subdued expectation of gen AI than respondents from other regions.
• The survey also points to a more nuanced view of AI adoption in infrastructure. Executives are not betting on a single AI technology: 29% prioritize traditional AI, 21% digital twins, 19% predictive analytics, and 13% agentic AI. Together, these technologies point toward a broadening AI ecosystem.

Talent and governance capabilities could determine AI scalability

Organizations increasingly know what they want to do with AI. The question is whether they have the workforce, governance, and organizational capabilities to execute. Scaling AI requires more than technical expertise.

Governance is a key challenge as AI becomes embedded in mission-critical decisions. Trustworthy AI requires more than technical standards alone. It demands a governance model that continuously governs, maps, measures, and manages AI risks as systems evolve.⁷

The United Kingdom’s nuclear regulatory review clearly illustrates the underlying governance issue, especially in a highly regulated market. The recommendations focused on clearer leadership, stronger coordination across regulators, mechanisms for resolving disputes, and enterprisewide risk management. Although the review was not primarily about AI, it shows that advanced analytical capabilities cannot compensate for fragmented accountability.⁸

Meanwhile, the technology talent gap is emerging as one of the largest implementation barriers globally. Research shows that 52% of organizations lack the in-house expertise needed to manage complex AI infrastructure, while only 9% say they have strong internal AI talent.⁹ These organizations require collaboration across engineering, data science, cybersecurity, procurement, operations, finance, and leadership.¹⁰ This reality is clearly reflected in Deloitte’s infrastructure survey (figure 6).

What the survey tells us:

• 34% of executives globally cite technology talent gap (particularly in AI and cybersecurity) as a major barrier to implementation.
• Regions that are still early in the AI learning curve, like Latin America and the Middle East/Africa, identify technology talent as a major barrier at 68% and 51%, respectively.
• By contrast, only 19% of North American respondents identify lack of technology talent as a major barrier, which indicates that this region may be further ahead on the AI adoption and maturity curve than the other regions.
• Private companies are more likely to identify technology talent as a barrier (41%) due to more active use of gen AI and related technologies, which exposes capability gaps sooner.

Infrastructure organizations need “bilingual” talent—people who can tread between infrastructure and AI expertise. This means hiring talent with know-how of physical assets, delivery constraints, and operational risk, as well as an understanding of AI-enabled decision systems across the infrastructure life cycle.

The good news is that organizations appear to recognize this gap. Infrastructure executives are investing in AI training and development programs (65%), building AI vision and implementation road maps (59%), and using pilots to demonstrate business value (57%) (figure 7).

The City of San José, California, is building AI capabilities through hands-on workforce development. Since 2024, more than 1,000 employees have completed an AI Upskilling Program developed with San José State University, creating tools to address operational challenges. By linking training to real departmental needs, the city is strengthening AI adoption, workforce capability, and service delivery.¹¹

Building the foundation for connected infrastructure

Infrastructure leaders are increasingly rethinking how infrastructure is planned, delivered, and operated. Rather than deploying AI as a standalone technology, they are embedding it within connected infrastructure systems to improve coordination, resilience, and life-cycle performance. It is no longer about automating individual tasks, but about creating infrastructure that can continuously monitor, learn, adapt, and respond to changing operating conditions.

The opportunity is especially significant since only about 25% of the infrastructure required by 2050 actually exists today.¹² This means much of tomorrow’s infrastructure can still be designed around more intelligent, adaptive operating models from the outset. While full system autonomy may still be some distance away, allowing routine operational decisions to move closer to machine speed could make the infrastructure more responsive while freeing scarce human talent to focus on higher-stakes judgments.

The organizations that scale AI effectively will likely not be those with the most pilots, but those with the strongest foundations built on trusted data flows, resilient technology architecture, clear governance, and talent that understands both infrastructure realities and AI-enabled decision-making.¹³