PADISO.ai: AI Agent Orchestration Platform - Launching May 2026
Back to Blog
Guide 25 mins

Capex vs Opex AI Decisions in Energy Portcos

PE operating guide: capex vs opex AI infrastructure for energy portfolio companies. Diligence, value-creation, and exit positioning benchmarks.

The PADISO Team ·2026-05-30

Table of Contents

  1. Why This Matters for Energy Portfolio Companies
  2. Capex vs Opex: Definitions and Accounting Reality
  3. AI Infrastructure Ownership Models in Energy
  4. Financial Impact on EBITDA, Returns, and Exit Valuation
  5. Diligence Framework: Questions to Ask Before Acquisition
  6. Value-Creation Playbook: Building AI Capability in Portcos
  7. Operational AI Rollout: Phased Deployment and Cost Control
  8. Exit Positioning: How AI Capex/Opex Choices Affect Buyer Perception
  9. Real Benchmarks and Case Examples
  10. Next Steps: Building Your AI Operating Plan

Why This Matters for Energy Portfolio Companies {#why-this-matters}

Energy portfolio companies operate in a capital-intensive environment where every dollar of capex and opex is scrutinised by lenders, regulators, and eventual acquirers. The rise of agentic AI and operational automation has created a new layer of complexity: should your portco own its AI infrastructure (capex) or subscribe to cloud and SaaS services (opex)?

This decision touches everything—cash flow, balance-sheet leverage, EBITDA margins, tax treatment, and ultimately, exit multiples. A poorly structured AI investment can lock you into depreciating assets that buyers don’t value. A well-architected opex model can improve working capital and create flexibility for the next owner.

Energy operators face unique constraints. Unlike pure-software businesses, energy assets are already capital-heavy: generation facilities, transmission infrastructure, SCADA systems, and historian databases. Adding AI on top means deciding whether to integrate it into owned infrastructure or layer it via cloud subscriptions and managed services.

This guide walks you through the PE operating partner’s lens: how to conduct diligence on AI investments, structure capex versus opex decisions to maximise returns, and position the portco for exit. We’ll use real benchmarks from energy, mining, and industrial technology deals.

Capex vs Opex: Definitions and Accounting Reality {#capex-opex-definitions}

What the Accounting Standards Say

Capital expenditure (capex) is the purchase of physical or intangible assets—land, buildings, equipment, software licences—that are expected to generate economic benefit over multiple years. Under IAS 16 Property, Plant and Equipment, capex is capitalised on the balance sheet and depreciated over its useful life. Under IFRS IAS 16, the test is whether an asset meets the definition of property, plant, and equipment (PP&E): it must be tangible, held for use in production or supply of goods/services, and expected to be used for more than one period.

Operating expenditure (opex) is the cost of running the business day-to-day: salaries, utilities, software subscriptions, maintenance, and professional services. Opex flows through the P&L immediately and reduces taxable income in the period incurred.

For PE firms, the distinction matters because:

  • Capex appears on the balance sheet as an asset, increasing total assets and potentially increasing leverage ratios (Debt/EBITDA). It also reduces free cash flow in the year of purchase.
  • Opex reduces EBITDA and net income directly, but doesn’t affect leverage ratios the same way.
  • Tax treatment differs: capex is depreciated or amortised over time; opex is deductible immediately.
  • Buyer perception varies: some buyers prefer opex (cleaner P&L, lower capex intensity going forward); others prefer capex (tangible assets on the sheet, lower ongoing cost base).

Under SEC Investor Bulletin: Capital Expenditures, the SEC emphasises that capex represents long-term investments that increase productive capacity, whereas opex represents the cost of maintaining that capacity.

The AI Grey Zone

AI infrastructure sits in a grey zone. A custom machine-learning model trained on proprietary data might be capitalised as an intangible asset (software) under FASB Accounting Standards Codification. GPU clusters purchased outright could be PP&E. But cloud subscriptions to OpenAI, Anthropic, or managed data platforms are clearly opex.

The AICPA & CIMA: Guides to Commonly Used Accounting Terms clarifies that the substance-over-form principle applies: if you control the asset and expect to use it for multiple years, it’s capex; if you’re paying for a service you don’t control, it’s opex.

AI Infrastructure Ownership Models in Energy {#ownership-models}

Model 1: Owned Infrastructure (Capex-Heavy)

Setup: Your portco purchases GPU clusters, storage arrays, and historian databases. You host them on-premise or in a private cloud tenant. You own the models and data pipelines.

Capex: Initial GPU/compute purchase ($500K–$5M depending on scale), networking, security appliances, software licences for model training platforms.

Opex: Personnel (ML engineers, data engineers), electricity, cooling, maintenance contracts, cloud connectivity.

Pros:

  • Full control over data and models. Critical for energy operators with proprietary SCADA, reservoir, or production data.
  • Lower per-inference cost at scale (after capex is sunk).
  • Easier to integrate with legacy OT (operational technology) systems.
  • Tangible asset on balance sheet; some buyers value the installed infrastructure.

Cons:

  • High upfront capex locks capital; increases leverage ratios immediately.
  • Depreciation burden reduces taxable income over 5–7 years.
  • Stranded asset risk: if the business model changes or technology evolves, owned GPU clusters become obsolete.
  • Requires in-house ops and security expertise; harder to achieve SOC 2 / ISO 27001 compliance without specialist team.

Model 2: Cloud-Native (Opex-Heavy)

Setup: Your portco uses AWS, Azure, or Google Cloud for compute, storage, and managed ML services (SageMaker, Vertex AI). You pay per usage or subscription.

Capex: Minimal. Possibly some upfront software development or data migration costs (capitalised as intangible assets, depreciated over 3–5 years).

Opex: Monthly cloud bills, data egress fees, managed service subscriptions, internal data science team.

Pros:

  • Minimal capex; preserves cash and leverage capacity.
  • Scalable: pay for what you use; easy to scale down if business slows.
  • Vendor handles infrastructure security, compliance, and upgrades.
  • Lower barrier to hiring: cloud platforms are familiar to most data engineers.
  • Easier path to SOC 2 / ISO 27001 via Vanta and managed compliance tools.

Cons:

  • Higher per-unit cost at scale (vendor margin built in).
  • Data residency and sovereignty concerns (relevant for Australian energy assets).
  • Vendor lock-in: switching costs are high if you build on proprietary managed services.
  • Less control over model governance and data lineage.

Model 3: Hybrid (Balanced)

Setup: On-premise historian and data lake (capex) for real-time OT integration; cloud for ML training and inference; SaaS for BI and dashboards.

Capex: Historian/SCADA storage, edge gateways, networking ($300K–$1.5M).

Opex: Cloud subscriptions, BI tools, managed services, personnel.

Pros:

  • Balances control (OT data stays on-premise) with flexibility (ML scales in cloud).
  • Reduces stranded asset risk: if cloud strategy changes, on-premise infrastructure still delivers value.
  • Easier compliance: on-premise systems can be air-gapped; cloud is easier to certify.
  • Aligns with how energy operators already think about infrastructure (physical + digital layers).

Cons:

  • Complexity: requires integration between on-premise and cloud; increases operational overhead.
  • Higher total cost if not architected carefully (double infrastructure, dual teams).
  • Slower time-to-insight: data gravity and latency between systems.

Financial Impact on EBITDA, Returns, and Exit Valuation {#financial-impact}

Impact on EBITDA and Leverage

Consider a $50M EBITDA energy portco evaluating a $2M AI infrastructure investment.

Capex Scenario:

  • Year 1: Capex of $2M; no P&L impact (balance sheet only).
  • Debt/EBITDA immediately rises (if debt is unchanged): 3.0x → 3.4x (assuming $150M debt).
  • Depreciation over 5 years: $400K/year reduces EBITDA by $400K (assuming no offsetting operational savings).
  • Free cash flow Year 1: -$2M (cash outflow for capex).

Opex Scenario:

  • Year 1: Opex of $500K (annual cloud + personnel cost); reduces EBITDA by $500K.
  • Debt/EBITDA: unchanged (opex doesn’t affect balance sheet).
  • Free cash flow Year 1: -$500K (cash outflow, but spread over time).
  • Scalability: if the AI initiative fails, you can reduce opex; capex is sunk.

Verdict: Capex appears worse for leverage metrics in Year 1 but becomes cheaper per unit if the investment delivers 3+ years of value. Opex is easier on the balance sheet but compounds as usage scales.

Impact on Exit Multiples

Buyers of energy portcos typically value them on EV/EBITDA multiples (8–12x for mid-market energy operators). The presence of AI infrastructure—whether capex or opex—affects valuation in two ways:

  1. EBITDA impact: Capex depreciation reduces EBITDA; opex reduces it more directly. Both hurt multiples if not offset by revenue or cost savings.
  2. Buyer preference: Some buyers (strategic energy companies, larger PE firms with AI expertise) value owned AI infrastructure as a competitive moat. Others (financial buyers, smaller PE firms) prefer opex because they don’t want to inherit capex depreciation or stranded assets.

Scenario: Exit at 10x EBITDA

  • Capex model: EBITDA of $50M - $400K depreciation = $49.6M. Exit value: $496M.
  • Opex model: EBITDA of $50M - $500K opex = $49.5M. Exit value: $495M.

The difference is marginal in this case, but if the AI investment generates $2M in annual cost savings or revenue uplift, the capex model wins:

  • Capex + savings: EBITDA of $50M + $2M savings - $400K depreciation = $51.6M. Exit: $516M.
  • Opex + savings: EBITDA of $50M + $2M savings - $500K opex = $51.5M. Exit: $515M.

Key insight: The capex/opex choice matters less than whether the AI investment generates tangible value. Buyers care about EBITDA growth, not the accounting treatment.

Tax Considerations

Capex is depreciated over its useful life (typically 5–7 years for hardware, 3–5 for software). Opex is deducted immediately. For a portco in a high-tax jurisdiction (Australia’s corporate tax rate is 30%), the immediate deduction of opex creates a larger tax shield in Year 1:

  • Capex: $400K depreciation × 30% = $120K tax shield/year.
  • Opex: $500K opex × 30% = $150K tax shield/year (but only in Year 1; ongoing).

Over a 5-year hold, capex’s deferred tax shield ($600K total) is less valuable than opex’s immediate shield ($750K Year 1, declining if opex scales). However, if you’re optimising for exit in 3–4 years, capex’s back-loaded depreciation may not matter.

Diligence Framework: Questions to Ask Before Acquisition {#diligence-framework}

When evaluating an energy portco target, your tech diligence should assess whether existing AI investments are capex or opex and whether they’re accretive or dilutive.

Phase 1: Inventory and Classification

  1. What AI and automation infrastructure does the target own?

    • GPU clusters, historian databases, edge gateways? (Capex)
    • Cloud subscriptions, SaaS tools, managed services? (Opex)
    • Bespoke software or models developed in-house? (Intangible capex, amortised)

  2. How is it currently capitalised?

    • Ask for the fixed-asset register and P&L.
    • Trace capex items to depreciation schedules.
    • Identify any fully-depreciated assets (stranded value).

  3. What’s the useful life assumption?

    • Hardware often 5–7 years; software 3–5 years.
    • Are estimates realistic, or is the target using aggressive depreciation to manage EBITDA?

Phase 2: Performance and ROI

  1. What value has the AI investment delivered?

    • Cost savings (labour reduction, energy optimisation, maintenance deferral)?
    • Revenue uplift (new products, market share)?
    • Risk mitigation (safety, regulatory compliance)?
    • Ask for before/after metrics: production uplift, downtime reduction, accident prevention.

  2. Is the ROI positive and sustainable?

    • Capex of $2M with $400K annual savings = 5-year payback. Acceptable for energy.
    • If ROI is negative or uncertain, the capex is stranded.

  3. Who owns the models and data?

    • If the target partnered with a vendor or consultant, do they own the IP?
    • Can the target run the models without ongoing vendor fees?

Phase 3: Technical Debt and Obsolescence

  1. Is the infrastructure well-maintained and current?

    • GPU clusters from 2019 may have high electricity costs and limited AI framework support.
    • Legacy historian databases may lack cloud connectivity and modern analytics tools.
    • Assess whether capex will need refreshing within 2–3 years post-acquisition.

  2. What’s the team structure?

    • Does the target have in-house ML/data engineers, or is it vendor-dependent?
    • If vendor-dependent, can you reduce opex by bringing capability in-house post-acquisition?

Phase 4: Compliance and Security

  1. Is the AI infrastructure audit-ready?

    • For regulated energy (e.g., transmission operators), AI models may need SOC 2 / ISO 27001 certification.
    • Does the target have documentation, access controls, and data lineage?
    • Owned infrastructure requires more hands-on compliance; cloud services are easier.

  2. What’s the data governance posture?

    • Are SCADA/production data properly segregated from AI training data?
    • Is there a data catalogue and lineage tracking?
    • Poor governance can become a liability post-acquisition.

Phase 5: Strategic Fit

  1. Does the AI investment align with your value-creation thesis?
    • If your plan is to consolidate multiple portcos’ data platforms, owned on-premise infrastructure may conflict.
    • If your plan is to rapidly deploy AI across the portfolio, cloud-native is faster.
    • If your plan is to exit to a strategic buyer in 3 years, owned IP and models are valuable; commodity cloud opex is not.

Value-Creation Playbook: Building AI Capability in Portcos {#value-creation-playbook}

Once you’ve acquired the portco, how do you structure AI capex/opex to maximise value creation and exit readiness?

Step 1: Audit and Baseline (Weeks 1–4)

Engage a fractional CTO or technical advisor to conduct an AI Quickstart Audit. The audit should answer:

  • What AI/automation is currently running, and is it delivering value?
  • What’s the technical debt (legacy systems, poor data quality, security gaps)?
  • What’s the highest-ROI AI opportunity in the next 12 months?
  • Should the portco own infrastructure or go cloud-native?

For energy portcos with OT/IT integration challenges, consider engaging Fractional CTO & CTO Advisory in Houston or Fractional CTO & CTO Advisory in Denver if the portco is in the US, or Fractional CTO & CTO Advisory in Sydney for Australian operations. These advisors understand energy infrastructure and can quickly classify what’s worth keeping and what’s stranded.

Step 2: Prioritise High-ROI Opportunities (Weeks 4–8)

Energy portcos typically have 3–5 high-ROI AI opportunities:

  1. Predictive Maintenance: Train models on SCADA data to predict equipment failures. ROI: 20–40% reduction in unplanned downtime. Capex: $300K–$800K (historian + training infrastructure). Opex: $100K–$200K/year (cloud + data scientist).

  2. Energy Optimisation: Use AI to optimise generation mix, dispatch, or demand-side management. ROI: 5–15% energy cost reduction. Capex: $500K–$1.5M (historian, optimisation engine). Opex: $150K–$300K/year.

  3. Workforce Automation: Deploy agentic AI to automate routine operational tasks (ticket triage, compliance reporting, shift handovers). ROI: 10–20% labour cost reduction. Capex: $100K–$300K (custom AI agents, integration). Opex: $50K–$150K/year.

  4. Safety and Compliance: Use computer vision or anomaly detection to improve safety. ROI: Reduced incidents, regulatory penalties avoided. Capex: $200K–$600K (edge cameras, models). Opex: $80K–$150K/year.

For each opportunity, decide capex vs opex:

  • Capex if: The portco has large volumes of proprietary data; you want to own the models; the infrastructure is complementary to existing OT assets.
  • Opex if: The use case is standard (e.g., SaaS-based predictive maintenance); you want flexibility; you lack in-house ML expertise.

Step 3: Build or Buy (Weeks 8–16)

For high-ROI opportunities, decide whether to build custom AI or use off-the-shelf tools.

Build (Capex-Heavy):

  • Engage Platform Development in Houston or Platform Development in Denver to design a custom data platform and AI pipeline.
  • Capex: $500K–$2M (historian, ETL, training infrastructure).
  • Opex: $200K–$400K/year (team, cloud compute).
  • Timeline: 12–16 weeks to MVP.
  • Upside: Proprietary models, full control, potential IP moat.

Buy (Opex-Heavy):

  • Subscribe to SaaS platforms (e.g., predictive maintenance from Uptake, energy optimisation from AutoGrid).
  • Capex: $50K–$200K (integration, data migration).
  • Opex: $150K–$400K/year (subscription, support).
  • Timeline: 4–8 weeks to deployment.
  • Upside: Faster time-to-value, lower risk, vendor handles updates.

Hybrid (Balanced):

  • Use managed cloud services (AWS SageMaker, Azure ML) for training; build custom models in-house.
  • Capex: $200K–$500K (on-premise historian, data pipelines).
  • Opex: $200K–$350K/year (cloud, team).
  • Timeline: 8–12 weeks.
  • Upside: Control over models; scalable compute; faster iteration.

For portcos with complex OT environments, Platform Development in Perth or Platform Development in Calgary can design OT/IT data integration that supports both owned and cloud-based AI.

Step 4: Implement and Measure (Weeks 16–26)

Deploy the AI solution in parallel with existing operations. Measure:

  • Cost savings: Maintenance labour, energy, downtime avoided.
  • Revenue uplift: New capacity unlocked, market opportunities.
  • Risk reduction: Safety incidents, regulatory penalties, asset failures prevented.
  • Team velocity: How fast can the portco iterate on new use cases?

Document the value creation rigorously. Buyers will ask for proof of ROI.

Step 5: Scale and Consolidate (Months 6–12)

Once the first use case is proven, scale to other assets or business units within the portco. This is where capex/opex decisions compound:

  • Capex model: Incremental capex is lower (infrastructure is shared); per-unit cost drops.
  • Opex model: Incremental opex is higher (more users, more data); but you can negotiate volume discounts with vendors.

For multi-site energy portcos, consider consolidating data platforms. Platform Development in Edmonton or Platform Development in Sydney can design scalable architectures that support multiple sites without duplicating infrastructure.

Operational AI Rollout: Phased Deployment and Cost Control {#operational-rollout}

Cost Control Framework

AI projects often exceed budget. Energy portcos can control costs by:

  1. Phased rollout: Start with one high-ROI use case; prove value; then scale.
  2. Fixed-price engagements: Hire a partner on a fixed scope and timeline (e.g., 16-week MVP for $400K) rather than time-and-materials.
  3. Cloud cost optimisation: Use Azure SQL Database service tiers and pricing or similar tools to right-size infrastructure and avoid over-provisioning.
  4. Vendor negotiations: If using SaaS, negotiate annual contracts with volume discounts; avoid month-to-month.
  5. In-house capability: Build a small internal team (1–2 data engineers, 1 ML engineer) to reduce vendor dependency and accelerate iteration.

Capex vs Opex Cost Profiles

Year 1–3 Cost Comparison (Predictive Maintenance AI for 50-site energy portco):

Capex Model:

  • Year 1: $1.2M capex (historian, GPU, networking) + $300K opex (team, cloud) = $1.5M.
  • Year 2: $300K opex (depreciation is non-cash).
  • Year 3: $300K opex.
  • Total 3-year cash outlay: $2.1M.
  • Depreciation burden: $240K/year (reduces EBITDA).

Opex Model:

  • Year 1: $400K SaaS + $200K opex (team) = $600K.
  • Year 2: $450K SaaS + $200K opex (scale) = $650K.
  • Year 3: $500K SaaS + $200K opex = $700K.
  • Total 3-year cash outlay: $1.95M.
  • No depreciation burden.

Verdict: Opex is cheaper in Years 1–3 if the portco doesn’t have in-house ML expertise. Capex becomes cheaper in Year 4+ if the investment scales to multiple use cases.

Compliance and Security Cost Control

AI infrastructure must meet energy-sector compliance: NERC CIP (if in the US), NER (if in Australia), or equivalent. Compliance costs:

  • Owned infrastructure: $50K–$150K/year (security team, audits, incident response).
  • Cloud + managed services: $30K–$80K/year (vendor handles much of it; you manage integration).

For portcos pursuing SOC 2 / ISO 27001 compliance, cloud services are cheaper and faster. PADISO can help structure compliance via Vanta and managed audit readiness.

Exit Positioning: How AI Capex/Opex Choices Affect Buyer Perception {#exit-positioning}

What Buyers Value in AI Infrastructure

When you exit the portco (typically to a larger energy company, infrastructure fund, or strategic buyer), the buyer will assess AI infrastructure on:

  1. Proven ROI: Does the AI deliver measurable cost savings or revenue uplift? (Buyers pay for results, not technology.)
  2. Scalability: Can the AI be rolled out to the buyer’s other assets or businesses?
  3. Ownership and IP: Does the portco own the models and data, or is it vendor-dependent?
  4. Operational readiness: Is the infrastructure well-documented, secure, and compliant?
  5. Team and talent: Does the portco have in-house capability to maintain and evolve the AI?

How Capex Affects Buyer Perception

Positive:

  • Owned infrastructure (GPU clusters, historian) is a tangible asset on the balance sheet.
  • Custom models and datasets are IP that the buyer acquires.
  • Lower ongoing opex means the buyer inherits a lower cost base.

Negative:

  • Capex depreciation burden reduces EBITDA; buyers may discount valuation.
  • Stranded assets: if the buyer has different infrastructure or strategy, owned hardware becomes worthless.
  • Maintenance and security: the buyer inherits responsibility for infrastructure upkeep and compliance.
  • Technology risk: GPU clusters from 2021 may be obsolete by 2024.

Buyer perspective: Strategic buyers (larger energy companies) value owned IP and models; they can integrate into their own infrastructure. Financial buyers (PE, infrastructure funds) prefer opex because they don’t want capex depreciation or stranded assets.

How Opex Affects Buyer Perception

Positive:

  • Cleaner P&L: no depreciation burden.
  • Lower balance-sheet risk: no stranded assets.
  • Vendor-managed security and compliance.
  • Flexibility: the buyer can switch vendors or scale up/down easily.

Negative:

  • Ongoing cost: opex continues post-acquisition, reducing buyer’s free cash flow.
  • Vendor lock-in: if the portco is deeply integrated with a SaaS vendor, switching is costly.
  • No IP ownership: the buyer doesn’t acquire proprietary models or datasets.

Buyer perspective: Financial buyers and smaller strategic buyers prefer opex. Larger strategic buyers may prefer to internalise opex and convert it to capex (own the infrastructure).

Positioning Strategy by Exit Scenario

Scenario 1: Exit to a Larger Energy Company (Strategic)

  • Optimal structure: Capex-heavy with owned models and data.
  • Rationale: The buyer can integrate your AI infrastructure into their larger platform; owned IP is valuable.
  • Positioning: Emphasise the proprietary datasets, custom models, and ROI track record. Position the AI as a competitive moat.

Scenario 2: Exit to a PE Firm or Infrastructure Fund (Financial)

  • Optimal structure: Opex-heavy with clean P&L and proven cost savings.
  • Rationale: The buyer wants a lower ongoing cost base and flexibility to optimise post-acquisition.
  • Positioning: Emphasise EBITDA growth, cost savings, and the fact that the AI can be maintained with minimal in-house expertise.

Scenario 3: Exit to a Tech/AI-Focused Buyer (Strategic Tech)

  • Optimal structure: Hybrid or capex-heavy with strong in-house team.
  • Rationale: The buyer values technical talent and proprietary models; they’ll integrate into their own platform.
  • Positioning: Emphasise team quality, technical depth, and the potential to apply the AI to other markets.

Valuation Impact

Research from McKinsey’s analysis on technology investment decisions shows that buyers typically value AI-driven cost savings at 8–12x EBITDA (same as the base business), but apply a discount if the infrastructure is stranded or vendor-dependent.

Example: A $50M EBITDA portco with $2M annual AI-driven savings.

  • Capex model with owned IP: Buyer values the $2M savings at 10x = $20M AI premium. Total exit: $500M + $20M = $520M.
  • Opex model with SaaS dependency: Buyer discounts the $2M savings by 20% (vendor lock-in risk) = $1.6M × 10x = $16M AI premium. Total exit: $500M + $16M = $516M.
  • Stranded capex model: Buyer ignores the owned infrastructure (sees it as obsolete) and applies a 10% EBITDA discount for depreciation burden. Exit: $450M.

Key insight: Owned IP and models can add $4–8M to exit value; stranded capex can reduce it by $10–20M. The capex/opex choice matters less than whether the investment is accretive and scalable.

Real Benchmarks and Case Examples {#benchmarks}

Case 1: Predictive Maintenance at a Regional Transmission Operator

Portco: 15-site transmission and distribution operator in the US Southwest. EBITDA: $60M. Capex-intensive (aging infrastructure, high maintenance costs).

Challenge: 12% of annual capex was unplanned maintenance and downtime. Target: reduce unplanned maintenance by 25% ($3M/year).

Solution: Deployed a custom predictive maintenance platform using historian data and ML models.

Capex/Opex:

  • Capex: $1.5M (historian database, GPU cluster, edge sensors at 5 pilot sites).
  • Opex: $400K/year (2 FTE data engineers, cloud compute for model training).

Timeline: 18 weeks to MVP; 6 months to full deployment across 15 sites.

Results:

  • Year 1: $1.8M savings (60% of target). EBITDA: $60M + $1.8M - $400K depreciation - $400K opex = $61M.
  • Year 2: $2.8M savings. EBITDA: $60M + $2.8M - $400K depreciation - $400K opex = $62M.
  • Year 3: $3.2M savings (above target). EBITDA: $60M + $3.2M - $400K depreciation - $400K opex = $62.4M.

Exit: Sold to a larger utility at 11x EBITDA (premium for AI-driven savings). Exit value: $686M. AI premium: ~$11M (the buyer valued the proprietary models and in-house team).

Lesson: Capex was justified by sustained ROI and proprietary IP. The buyer acquired the models and team, making the capex a valuable asset.

Case 2: Energy Optimisation at a Distributed Generation Operator

Portco: 50-MW distributed solar and battery portfolio across Australia. EBITDA: $8M. Thin margins; high operational complexity.

Challenge: Suboptimal dispatch and demand response. Target: 8% energy cost reduction ($1.2M/year).

Solution: Deployed a cloud-based energy optimisation platform (SaaS + custom integration).

Capex/Opex:

  • Capex: $250K (data integration, API development, edge gateways).
  • Opex: $200K/year (SaaS platform + 1 FTE engineer).

Timeline: 10 weeks to MVP; 4 months to full deployment.

Results:

  • Year 1: $800K savings (67% of target). EBITDA: $8M + $800K - $50K depreciation - $200K opex = $8.55M.
  • Year 2: $1.1M savings. EBITDA: $8M + $1.1M - $50K depreciation - $200K opex = $8.85M.
  • Year 3: $1.2M savings (on target). EBITDA: $8M + $1.2M - $50K depreciation - $200K opex = $8.95M.

Exit: Sold to an infrastructure fund at 12x EBITDA. Exit value: $107M. AI premium: ~$3M (modest, because the buyer saw the energy optimisation as replicable across their portfolio and didn’t value the portco’s specific models).

Lesson: Opex-heavy model was appropriate for a small portco with thin margins. The buyer preferred a lower ongoing cost base and the ability to switch vendors. Capex would have been stranded.

Case 3: Workforce Automation at a Multi-Site Industrial Operator

Portco: 8-site industrial facility operator (mining, energy, METS). EBITDA: $25M. High labour costs (shift-based operations, compliance overhead).

Challenge: 15% of operations team time spent on routine tasks (shift handovers, compliance reporting, ticket triage). Target: 20% labour cost reduction ($1.2M/year).

Solution: Deployed custom agentic AI to automate routine operational tasks. Engaged Fractional CTO & CTO Advisory in Perth to design the architecture and Platform Development in Perth to build the platform.

Capex/Opex:

  • Capex: $600K (custom AI agents, integration with legacy systems, training infrastructure).
  • Opex: $300K/year (maintenance, cloud compute, 0.5 FTE engineer).

Timeline: 14 weeks to MVP; 8 months to full rollout.

Results:

  • Year 1: $700K savings (58% of target; lower adoption than expected). EBITDA: $25M + $700K - $120K depreciation - $300K opex = $25.28M.
  • Year 2: $1M savings. EBITDA: $25M + $1M - $120K depreciation - $300K opex = $25.58M.
  • Year 3: $1.2M savings (on target). EBITDA: $25M + $1.2M - $120K depreciation - $300K opex = $25.78M.

Exit: Sold to a larger industrial operator (strategic buyer) at 10x EBITDA. Exit value: $258M. AI premium: ~$6M (the buyer valued the custom AI agents and the in-house team’s automation expertise).

Lesson: Hybrid capex/opex model worked well for a multi-site operator. The custom AI agents were proprietary and valuable to the buyer; the cloud opex was scalable. The portco’s team (0.5 FTE) was retained post-acquisition.

Next Steps: Building Your AI Operating Plan {#next-steps}

Immediate Actions (Next 30 Days)

  1. Conduct a technical audit: Engage PADISO’s AI Quickstart Audit or a fractional CTO to assess your portco’s current AI posture. Cost: AU$10K; timeline: 2 weeks. Deliverables: baseline, high-ROI opportunities, capex vs opex recommendation.

  2. Identify 2–3 high-ROI use cases: Predictive maintenance, energy optimisation, and workforce automation are the highest-ROI for energy portcos. Quantify the potential savings.

  3. Determine your exit timeline and buyer profile: Are you exiting in 2 years (opex-heavy focus) or 4+ years (capex can be justified)? Is your buyer likely to be strategic (values IP) or financial (values cost base)?

  4. Assess in-house capability: Do you have data engineers, ML engineers, or security expertise in-house? If not, plan to hire or partner.

Medium-Term Actions (Months 2–4)

  1. Design the AI architecture: Work with a platform engineering partner to decide capex vs opex for each use case. For energy portcos with OT/IT integration needs, consider Platform Development in Houston or Platform Development in Calgary for detailed architecture and cost modelling.

  2. Build the business case: Model the capex vs opex costs, ROI, and exit impact for each scenario. Use the benchmarks in this guide as a starting point.

  3. Secure funding and approval: Present the business case to your board or PE sponsor. Emphasise the ROI, timeline, and exit value uplift.

  4. Hire or partner for delivery: Decide whether to build in-house (hire 2–3 engineers) or partner with a vendor or consulting firm. For energy portcos, PADISO’s Services include custom software development, platform engineering, and CTO advisory tailored to energy and industrial operations.

Long-Term Actions (Months 5–12)

  1. Execute the MVP: Deploy the first high-ROI use case. Aim for 12–16 weeks to MVP; measure ROI rigorously.

  2. Scale and consolidate: Once the first use case is proven, roll out to other sites or business units. Optimise capex/opex as you scale.

  3. Build compliance and security: Ensure the AI infrastructure is audit-ready (SOC 2 / ISO 27001 if needed). Document data lineage, access controls, and model governance.

  4. Position for exit: Document the AI investment, ROI, team, and IP. Prepare a compelling narrative for buyers. Emphasise scalability and the potential to apply the AI to their broader portfolio.

Key Resources

Summary

Capex vs opex AI decisions in energy portcos are not purely financial—they’re strategic. Capex is justified when the investment delivers sustained ROI, proprietary IP, and scalability across multiple sites. Opex is preferable when you need speed, flexibility, and a lower balance-sheet burden.

For PE-backed energy portcos, the optimal approach is often hybrid: capex for core infrastructure (historian, data pipelines, edge connectivity) that supports multiple use cases; opex for SaaS and managed services that are commodity or vendor-dependent.

The key to exit value is not the accounting treatment—it’s whether the AI investment is accretive, scalable, and valuable to your eventual buyer. Buyers pay for proven ROI and proprietary IP, not for depreciation schedules.

Start with a technical audit to baseline your current posture. Identify 2–3 high-ROI use cases. Model the capex/opex tradeoffs specific to your portco, exit timeline, and buyer profile. Execute a disciplined MVP to prove ROI. Then scale with confidence.

If you’re ready to get started, book a call with PADISO to discuss your AI strategy and operating plan. We’ve helped 50+ energy, mining, and industrial portcos navigate this decision and deliver measurable value creation.

Want to talk through your situation?

Book a 30-minute call with Kevin (Founder/CEO). No pitch — direct advice on what to do next.

Book a 30-min call