Delivering Energy Infrastructure at Scale Has Become Increasingly Challenging 

As load growth intensifies in the US, the race to build energy infrastructure has fundamentally changed. Data centers, utilities, independent power producers, and industrial companies are all competing for the same equipment, materials, and labor. There are more projects in the pipeline than ever, but they are getting harder to deliver on time and on budget.

Over the past several months, we spoke with EPC (engineering, procurement, and construction) firms*, utilities, developers, and investors to better understand where AI applications can, and cannot, create value across the energy EPC ecosystem.** One theme surfaced repeatedly: the biggest opportunity is unlikely to come from replacing the EPC industry’s existing software stack. Instead, nearly every conversation pointed toward a more specific opportunity set: AI-native applications that solve one painful workflow exceptionally well, integrate with the customer’s existing tools, and generate structured project data that enables expansion into the customer’s adjacent workflows over time.

Building a clean and reliable modern power system requires the deployment of more generation, storage, transmission, industrial, and digital infrastructure than ever before. Meeting that demand will depend not only on continued investment, but on improving the productivity of the organizations responsible for designing, procuring, and constructing these assets. AI will not solve every structural bottleneck facing the infrastructure sector, nor should it be expected to. However, by helping engineering, procurement, and construction teams accomplish more with the people and resources they already have, AI has the potential to reduce the cost and time required to build the critical infrastructure underpinning both the energy transition and the AI economy. 

Key Takeaways 

• The EPC software market is mature, but workflows remain fragmented. The industry has relied on platforms such as Autodesk and Primavera for decades, yet critical project information still moves between disconnected software systems, spreadsheets, PDFs, and teams.

• The biggest barriers to AI adoption are trust and implementation challenges. The prevalence of legacy processes and systems, high switching costs, execution risk, and poor data quality mean that targeted, workflow-specific solutions with significant ROI are more likely to be adopted than broader solutions. 

• The most promising opportunities for AI adoption are concentrated in a handful of high-friction workflows. Detailed engineering, estimation, procurement, and labor coordination are areas where AI applications could meaningfully improve productivity.

• Companies entering this market must earn the right to access project data. By solving one painful workflow well, startups can gradually build trust with customers and integrate into existing processes, produce structured project data, and naturally expand into adjacent use cases over time.

*Note: throughout this article, when we refer to EPCs, we primarily mean large engineering, procurement, and construction firms (e.g., Bechtel, Burns & McDonnell, Black & Veatch, Sargent & Lundy) that assume responsibility for delivering complex infrastructure projects. Smaller engineering firms, contractors, and developers may experience many of the same pain points, and can also be attractive customers, but often have different incentives and purchasing behavior.

**We focused specifically on AI software applications, not physical AI and robotics solutions, which also hold promise in this ecosystem but face meaningfully different adoption dynamics.

If you’re a seed-stage founder developing digital solutions for the energy infrastructure ecosystem, we’d love to hear from you! Contact us.

Mapping the Opportunity

Electrification, the energy transition, and the rapid buildout of AI infrastructure are driving one of the largest infrastructure investment cycles in decades. The International Energy Agency projects global energy investment will reach $3.4 trillion in 2026 with roughly $2.2 trillion of this allocated specifically for the grid, renewables, storage, and other clean energy technologies. At the same time, Goldman Sachs has estimated that global AI infrastructure spend will exceed $1 trillion in 2026.

Turning this generational investment into operating assets that are delivered on time and on budget ultimately depends on the organizations responsible for engineering, procurement, and construction. EPC firms sit at the center of the infrastructure delivery ecosystem, coordinating across numerous stakeholders, including developers, utilities, equipment suppliers, contractors, and owners, to deliver physical infrastructure. Delivering infrastructure is inherently complex and faces significant execution risk that can impact project performance. EPCs with less experience demonstrate a greater risk of cost overruns, and stakeholders throughout the ecosystem emphasize that trust must be earned from demonstrated execution, which is a significant headwind for new entrants. 

Software Isn’t New to EPC

The EPC industry has a long history of software adoption. Firms have relied on digital tools such as AutoCAD (by Autodesk) for more than four decades, and project teams have managed complex capital projects with platforms like Primavera for a similar period. More recently, venture-backed startups have expanded the software landscape further, building products across areas including design automation, estimation, construction management, document intelligence, procurement, scheduling, and field operations. The construction software market alone was $13 billion in 2025 and is expected to reach $32 billion by 2035.

Despite a relatively mature software ecosystem, many of the industry’s most important workflows remain fragmented. Project information is scattered across spreadsheets, PDFs, siloed software systems, emails, internally developed tools, and in the minds of employees. Data generated during one phase of a project often fails to transfer cleanly to the next, forcing teams to repeatedly recreate, validate, or reconcile information throughout the project lifecycle. Fragmentation and poor data quality, not the absence of software, are the defining characteristics of the current EPC software ecosystem.

Several Key Themes Influence AI Adoption in EPC

Our conversations with EPCs and other stakeholders in this segment highlighted four key themes that influence the adoption of AI applications. 

1. Different types of infrastructure projects face different risks and require different software solutions

EPC should not be viewed as a monolith. For example, utility-scale solar is increasingly standardized, with construction processes that more closely resemble manufacturing. By contrast, industrial facilities and firm power plants remain significantly more bespoke and require materially different capabilities across engineering, procurement, and construction. Recent research examining hundreds of energy infrastructure projects confirms that project performance varies substantially by technology type, with standardized assets such as solar and wind generally experiencing lower construction risk than more bespoke projects such as nuclear and hydropower. Different project types therefore face different challenges and often require different software solutions.

2. Adoption is more constrained by implementation challenges than by AI capability

The key question for the EPC industry is less about what AI can do and more about what organizations can realistically adopt.

Legacy software and decades-old workflows remain deeply embedded throughout the EPC industry. Large infrastructure projects leave little room for execution risk, making organizations understandably cautious about introducing new technologies and stakeholders into established processes. Additionally, the time and labor required to prepare project data for AI applications can undermine the expected return on investment and limit adoption. As one expert put it: “Just because AI can, should it?”

3. Project information remains fragmented across the value chain

Stakeholders including engineering firms, suppliers, contractors, EPCs, and developers, each maintain different records of project information, often across disconnected software systems. Even owners who have contractual rights to project data frequently struggle to access complete project records after work has been distributed across multiple contractors. Additionally, proprietary engineering knowledge, especially related to complex, low-volume infrastructure like power plants, can discourage organizations from moving sensitive project information into third-party cloud environments, further increasing data fragmentation. 

4. Productivity is the metric that matters

AI will ultimately be judged by its ability to improve productivity. Several interviewees described engineering organizations spending more labor hours today than they did a decade ago on the same types of projects despite increasingly sophisticated software. Estimating workflows still require weeks or months of manual effort despite software adoption. Many organizations worry that automation gains won’t justify the time and resources required to prepare data and workflows for AI. Unless the potential ROI from productivity gains is extremely high, these concerns present headwinds to adoption. Therefore, targeting use cases with an outsized potential for productivity gains is key. 

Opportunities for AI Applications in EPC 

AI adoption in EPC is unlikely to be evenly distributed across the value chain. We believe the strongest opportunities lie not in replacing established software ecosystems or rebuilding EPC workflows from the ground up, but in solving specific operational pain points that deliver immediate and measurable productivity gains.

Four workflows stood out as particularly manual, fragmented, and labor-intensive, making them compelling targets for AI-native software startups.

Pain Point 1: Detailed engineering remains labor-intensive 

Detailed engineering translates high-level project concepts (Front-End Engineering Design, or “FEED”) into the drawings, models, and specifications required for construction. On large infrastructure projects, it accounts for the vast majority of engineering effort and frequently becomes the critical path before procurement or construction can begin.

A major challenge lies in the number of iterations required. Design changes often cascade across multiple disciplines, requiring engineers to update drawings, documentation, calculations, and downstream deliverables. Even relatively small revisions can trigger significant amounts of repetitive manual work.

Where We See Opportunity

AI can boost productivity by reducing the amount of time spent on repetitive work and allowing engineers to focus on higher-value design decisions and technical review. Additionally, these workflows generate structured engineering data that can support downstream estimating, procurement, and scheduling workflows.

Startups: Candid Intelligence, Torus, Enginira, Endra, Augmenta, Outerport, P1-AI  

Pain Point 2: Estimation is disconnected from the rest of the project lifecycle

Estimation is where engineering decisions translate to project economics. Before work can begin, estimators must interpret engineering drawings, extract quantities, develop material takeoffs, review historical projects, coordinate assumptions, and assemble project costs. This work often takes thousands of hours on large capital projects. Despite advances in engineering software, much of today’s estimation process still depends on manually recreating information that already exists in other fragmented systems. 

Where We See Opportunity

Tools that transform engineering documents into structured project information can provide immediate productivity gains for estimation workflows, and utilizing this structured data to support downstream workflows can create additional long-term value.

Startups: Redwagon, Purplefish, Armeta 

Pain Point 3: Fragmented data impacts procurement 

Procurement processes require constant coordination between numerous teams and external stakeholders. Project information moves between different software environments, document formats, and internal processes, and many of these systems do not talk to one another. Teams spend considerable time locating documents, reconciling inconsistencies, validating assumptions, and unnecessarily recreating information. Data fragmentation increases the risk of delays and reduces opportunities to optimize costs. 

Where We See Opportunity

Breaking down data silos in procurement can improve coordination, reduce schedule risk, and increase opportunities around bundling materials to reduce cost.

Startups: Fluxco, KayaAI

Pain Point 4: Coordinating skilled labor is increasingly critical 

Delivering large infrastructure projects requires coordinating hundreds, even thousands, of skilled tradespeople across long project timelines, often spanning multiple years. As investment in both energy infrastructure and AI infrastructure accelerates, competition for skilled labor is increasing, placing additional pressure on project schedules. For instance, the power sector in the US alone faces an estimated shortage of 500,000 skilled workers by 2030. Software can’t solve labor shortages on its own (and robotics and physical AI present their own exciting opportunities), but it can help organizations make better use of the workforce they already have.

Where We See Opportunity

We see two complementary opportunities emerging.

Planning intelligence. AI can help EPCs, developers, and owners anticipate workforce constraints earlier in the project lifecycle, improving project sequencing, labor selection, resource allocation, and schedule planning before work begins.

Construction intelligence. Once projects are underway, AI can improve crew coordination, scheduling, resource allocation, progress tracking, and schedule execution. Rather than replacing labor, these tools seek to improve the productivity of existing engineering and construction teams. 

Startups: Fyld, Doxel, AI Clearing, Buildots

Powerhouse Ventures’ Perspective: Building from the Workflow Up

AI has the potential to meaningfully accelerate the transition to a modern power system, not by replacing EPCs or other key stakeholders, but by increasing the productivity of the organizations that have a track record of successful execution. 

The most compelling startups will begin with a narrow, high-value workflow that delivers immediate and measurable ROI, integrates into existing processes, and produces structured project data that can naturally expand into adjacent workflows over time.We believe startups can earn adoption not by asking customers to replace trusted legacy systems, but by solving one pain point exceptionally well, building both trust and a durable data advantage in the process.

We’re excited about the potential for AI-native applications to improve productivity and help reduce the time and cost required to deliver the infrastructure powering a more resilient, electrified future. It’s time to build! 

If you’re a seed-stage founder developing digital solutions for the energy infrastructure ecosystem, we’d love to hear from you! Contact us.

Appendix: Startups Mentioned

• Candid Intelligence - AI-native engineering platform for early feasibility through detailed engineering 

• Torus - Using AI to draft complete engineering packages for EPCs 

• Enginira - Automating industrial process plant design and documentation 

• Endra - Automating MEP engineering design 

• Augmenta - Automating electrical routing and generating electrical BIM models 

• Outerport - AI agents generating electrical and control system designs

• P-1 AI - AI mechanical and electrical engineer for industrial teams

• Redwagon - AI engineer for capital project estimation, procurement, and hazard review 

• Purplefish - Automating material and labor takeoff for industrial piping

• Armeta - AI cost estimating and permitting for construction and heavy industry

• Fluxco - Transformer marketplace with AI-optimized specs and procurement

• KayaAI - AI agent managing equipment procurement and delivery

• Fyld - Field operations platform for infrastructure crews

• Doxel - Computer vision progress tracking using hardhat-mounted cameras

• AI Clearing - Drone and AI-based progress tracking for infrastructure construction

• Buildots - Tracking site progress and project risks against BIM and schedules