Why Are Major Energy Companies Building AI Internally?

Even When Specialized Vendors May Be Better Positioned

By: Lee S. Gliddon III, President/COO

Across the energy industry, some of the largest utilities in America are investing heavily in internal Artificial Intelligence initiatives.

From predictive maintenance and operational analytics to outage management and Project Controls automation, utilities are racing to position themselves as “AI-enabled enterprises.”

This seems logical. These organizations own enormous amounts of operational data and have substantial IT budgets.

But a deeper question is beginning to emerge across the industry:

Why are utilities attempting to build many of these AI solutions internally instead of relying on specialized vendors that already possess decades of industry knowledge, operational data expertise, and Project Controls experience?

The answer lies in a mix of strategy, culture, organizational politics, and a widespread disconnect of what industrial AI requires.

Executive Fear of Falling Behind

The AI wave has created enormous pressure inside large corporations. Boards of directors and executive leadership teams increasingly ask: “What is our AI strategy?”

No major utility wants to appear behind competitors in the race toward automation and intelligent operations. This creates internal urgency to launch AI initiatives quickly, often by establishing internal AI teams or “innovation” centers.

From an executive perspective, internal development feels safer because the company controls the roadmap. Sensitive operational data stays in-house along with the intellectual property. IT governance is centralized, making cybersecurity concerns appear easier to manage. In highly regulated industries like nuclear power and energy infrastructure, those concerns are amplified.

However, many organizations underestimate the complexity of building operational AI systems that work reliably in industrial environments.

AI Is Not the Hard Part, The Data Is

One of the biggest fails in enterprise AI is the belief that AI itself is the primary challenge.

Modern AI models and frameworks are increasingly accessible. What separates successful industrial AI implementations from failed ones is the quality, structure, history, and context of the underlying operational data. Utilities often, erroneously, assume: “We already own the data, so we can build the AI.”

Owning the data is very different from possessing normalized, contextualized, historically trusted, cross-system integrated, operationally meaningful data. Just as important—having an AI-ready data architecture. This is something many organizations still do not fully understand or know how to implement.

As modern AI began emerging, the utility industry rushed toward the concept of the Data Lake as the ultimate solution. However, experienced Project Controls professionals who deeply understood the importance of structured operational data, including Bechtel’s retired “100 Million Dollar Man,” Jim Melvin, argued that relying solely on Data Lakes would ultimately produce less meaningful AI outcomes. When organizations lose structured relationships within the data, such as schedule logic ties, budget structures, and resource-loaded scheduling information, the resulting AI becomes far less effective. That realization helped drive the evolution toward the Data Lakehouse, which preserves both the scale of modern data platforms and the structured relationships critical to producing high-value industrial AI outcomes.

Most utility environments are fragmented across dozens of disconnected systems: Primavera, Maximo, SAP, Access databases, Excel spreadsheets, SharePoint sites, engineering repositories, outage systems, document management platforms, custom legacy applications, the list goes on and on.

This creates one of the biggest hidden barriers in industrial AI.

A large language model can summarize text or generate reports quickly. But generating accurate operational outcomes, especially in nuclear operations and Project Controls, requires far more than generic AI capability.

Industrial AI Requires Deep Domain Expertise

There is a massive difference between building a chatbot and building reliable operational intelligence.

For example, AI outcomes such as:

• Identifying outage schedule risk drivers
• Automatically generating work scopes
• Building schedules from scope data
• Optimizing resource allocation
• Forecasting earned value performance
• Predicting execution bottlenecks
• Identifying work available windows
• Supporting fleet-level outage benchmarking

…are not generic technology problems. They are highly specialized operational problems.

They require a deep understanding of:

• Outage execution behavior
• Work package relationships
• Schedule logic
• Resource constraints
• Fleet operations
• INPO performance expectations
• NRC sensitivities
• Maintenance planning
• Project Controls methodologies
• Operational risk tolerance

This is where specialized vendors frequently possess a major advantage. Many niche vendors in the energy and nuclear space have spent decades building operational context, historical metrics, workflow expertise, and trusted calculations. That expertise is difficult to replicate internally, even with large budgets.

Organizational Politics Play a Bigger Role Than Many Realize

Another major factor is internal organizational protection. Large utility IT organizations are often incentivized to retain budget authority and expand internal teams. With that, they look to control architecture decisions to reduce vendor dependence, thereby owning strategic transformation programs. An outside vendor delivering a superior solution can unintentionally threaten internal roadmaps, consulting contracts, organization influence, existing staffing models, and perceived technology ownership.

This dynamic is rarely discussed openly, but it exists across many industries, especially in highly regulated enterprise environments. The result is that utilities sometimes spend years attempting to internally recreate capabilities that already successfully exist externally.

Nuclear Industry Culture Favors Control and Caution

The nuclear industry, in particular, operates within a culture where caution is both understandable and necessary. Reliability, cybersecurity, regulatory compliance, and operational safety are treated as absolute priorities. As a result, utilities often favor internally governed solutions that move through long testing cycles, gradual implementation phases, and layers of centralized oversight.

Those priorities are entirely reasonable given the critical nature of nuclear operations. However, they can also significantly slow innovation. In many cases, specialized vendors have already solved complex challenges involving data normalization, outage analytics, dashboard automation, schedule intelligence, predictive analytics, cross-fleet benchmarking, and AI-ready data architectures. Yet despite those advancements, procurement models and internal governance structures frequently steer organizations toward slower, internally driven development paths rather than leveraging proven external expertise.

History Suggests Specialized Vendors Often Win

This pattern is not unique to Artificial Intelligence. Large enterprises have historically attempted to develop many strategic technologies internally before eventually turning to specialized external vendors. The same cycle has played out repeatedly with ERP systems, cybersecurity platforms, cloud infrastructure, outage management tools, business intelligence software, data visualization platforms, and advanced analytics solutions.

The pattern is remarkably consistent. An enterprise launches an ambitious internal development effort, believing its size, resources, and access to data will provide an advantage. Over time, however, complexity expands rapidly. Integration challenges begin surfacing across disconnected systems, costs rise, timelines slip, while the original scope becomes increasingly difficult to manage. Meanwhile, specialized vendors have spent years refining focused solutions and continue outperforming internal efforts in both capability, speed, and most importantly, cost.

By the time leadership recognizes the initiative is no longer on track, the organization has invested enormous amounts of money, time, and political capital into the project. The sunk costs become difficult to walk away from, accountability pressures intensify, and the individuals most closely associated with the failed initiative leave the organization or are replaced. Industrial AI appears on track to follow this same historical trajectory.

The Future Belongs to Hybrid Models

The utilities that achieve the greatest success with Artificial Intelligence will not be the ones attempting to build every capability in-house. Instead, the strongest long-term performers will maintain ownership of their operational strategy, governance frameworks, cybersecurity requirements, and core data assets, while strategically leveraging specialized vendors for highly technical and domain-specific capabilities.

These external specialists will increasingly play critical roles in areas such as AI-driven workflows, operational analytics, Project Controls intelligence, as well as predictive modeling and data engineering, outage optimization, and schedule intelligence.

This hybrid model allows utilities to retain strategic control over their operations and data while simultaneously benefiting from decades of specialized industry expertise, proven technologies, and operational lessons. Rather than reinventing complex solutions internally, organizations can accelerate innovation by combining internal governance with external domain mastery.

In Energy Operations, Trust Matters More Than Hype

One of the biggest realities in industrial AI is that operational trust matters far more than flashy demonstrations.

A generic AI system may produce impressive presentations or summaries. But if the outputs are operationally unreliable and inaccurate, adoption will fail, every time. In environments like nuclear power generation, outage management, and fleet operations, accuracy and trust are everything. That is why deep operational expertise remains one of the most valuable assets.

And the companies that truly understand both the data and the operational environment may ultimately be the best positioned ones to lead the next generation of AI-enablement.

A practical next step for readers who want to explore this model further is the Knowledge Relay solutions overview, the article on AI and unified data in modern plant operations, and recent case studies showing how specialized expertise supports utility performance improvements.