McKinsey & Company · Published 2026-08-12

The $106 Trillion Infrastructure Opportunity

privateReview: unreviewedConfidence: high
infrastructureprivate capitalartificial intelligencelimited partnersdata centersdata-centersinference-placementedge-computingsovereign-aigenerative-planningunstructured-datacapacity-forecastingvalue-creation
Open on YouTube ↗

Summary

Overview

  • Speaker: Sean Brown, Al Green, Adrien Couret
  • Channel: McKinsey & Company
  • Main topic: Global Infrastructure Investment and Private Capital Opportunities
  • Purpose: To provide strategic insights into the $106 trillion global infrastructure investment opportunity, highlighting trends in private capital, public sector constraints, AI integration, and LP portfolio allocations. McKinsey & Company experts Sean Brown, Al Green, and Adrien Couret discuss the massive $106 trillion global infrastructure investment requirement through 2040, exploring how private capital is stepping in to bridge the funding gap, the role of AI in transforming infrastructure operations, and LP allocation trends.

Topic Map

The $106 Trillion Infrastructure Investment Gap

  • Explanation: McKinsey's Global Private Markets report shows that more than $100 trillion in investments is needed through 2040 to meet global infrastructure requirements, which is double the amount from the previous 15 years.
  • Key claims:
    • Infrastructure investment is highly correlated with global GDP.
    • The $106 trillion figure is double the investment level of the last 15 years.
    • Public sectors alone cannot raise this capital, creating an opening for private capital.
  • Examples:
    • Replacing aging infrastructure in the US and Europe.
    • Expanding digital, energy, and transport infrastructure globally.
  • Terminology:
    • infrastructure
    • global GDP
    • private capital
    • public sector
  • Why it matters: Understanding the scale of the investment gap highlights the critical need for private capital and innovative financing models.

The Expanding Definition of Infrastructure

  • Explanation: Infrastructure now encompasses far more than traditional transport (bridges, airports, railways) and includes data centers, fiber networks, satellites, agriculture, and defense assets.
  • Key claims:
    • Agriculture, aerospace, and defense are increasingly exhibiting infrastructure-like characteristics.
    • Many investments are happening at the intersection of traditional categories, such as smart parking garages and bridge power for data centers.
    • Public sector appetite for private capital is growing across Europe and the Americas.
  • Examples:
    • Data centers and fiber networks powering the digital economy.
    • Agricultural infrastructure and space-based assets.
  • Terminology:
    • infra-like characteristics
    • digital infrastructure
    • intersections
  • Why it matters: Broadening the definition of infrastructure creates new asset classes and investment opportunities for limited partners.

Geography and Sector Growth Drivers

  • Explanation: The majority of infrastructure investment will occur in Asia due to GDP growth, while North America and Europe face massive costs to replace and modernize aging infrastructure.
  • Key claims:
    • 70 trillion out of 106 trillion of infrastructure investment will happen in Asia.
    • Failing to replace aging infrastructure in the US could cost the economy $10 trillion by 2039.
    • Digital and energy are the single fastest-growing infrastructure sectors, driven by AI and clean energy demand.
  • Examples:
    • Replacing aging water and transport pipes in US municipalities.
    • Expanding renewable energy and data center capacity.
  • Terminology:
    • aging infrastructure
    • digital sector
    • energy investment
  • Why it matters: Identifying geographic and sectoral hotspots helps investors target high-growth infrastructure opportunities.

The Role of AI in Infrastructure

  • Explanation: Artificial intelligence is driving unprecedented demand for data centers and energy while dramatically improving operational efficiency, project scheduling, and asset management in infrastructure.
  • Key claims:
    • AI is both a root driver of infrastructure expansion and a tool to improve project efficiency by 10% to 25%.
    • Generative scheduling and design can compress project timelines and reduce costs significantly.
    • Data consolidation using AI tools helps investors and operators manage unstructured infrastructure assets.
  • Examples:
    • Using AI-based scheduling to manage complex $200M+ construction projects.
    • Satellite infrared imaging to detect heat-venting building inefficiencies.
  • Terminology:
    • generative scheduling
    • generative design
    • edge data centers
    • sovereign AI
  • Why it matters: AI adoption is transforming infrastructure from a laggard asset class into a technology-driven value creation engine.

LP Allocation Trends and Value Creation

  • Explanation: Limited partners are increasingly allocating capital to infrastructure, prioritizing value creation, operational efficiency, and portfolio diversification over passive asset holding.
  • Key claims:
    • Infrastructure closed-end funds reached nearly $200 billion in fundraising in the last calendar year.
    • Top criteria for LPs include performance, investment team quality, and value creation strategy.
    • LPs are moving beyond just buying and holding assets to actively improving operational performance.
  • Examples:
    • Allocations to infrastructure reaching nearly $2 trillion in AUM.
    • Internal LP teams establishing standard playbooks for capex deployment and procurement.
  • Terminology:
    • limited partners (LPs)
    • general partners (GPs)
    • assets under management (AUM)
    • value creation strategy
  • Why it matters: LP behavior dictates which general partners succeed and how infrastructure assets are managed for maximum yield.

Key Points

The $106 Trillion Requirement

  • Explanation: Global infrastructure needs through 2040 require $106 trillion, double the spending of the previous 15 years.
  • Evidence: McKinsey's 2026 Global Private Markets report.
  • Practical implication: Private capital must partner with governments to bridge the massive funding gap.

AI as a Catalyst

  • Explanation: The AI boom is driving explosive demand for data centers, power generation, and grid updates while offering operational efficiencies.
  • Evidence: Observations on data center growth and generative scheduling reducing project costs by 10-25%.
  • Practical implication: Investors must factor AI demand and efficiency tools into all infrastructure underwriting.

Evolution of LP Preferences

  • Explanation: Limited partners are increasing allocations to infrastructure and prioritizing active value creation over passive yield.
  • Evidence: Infra fundraising hitting all-time highs and LP surveys ranking value creation strategy in top three criteria.
  • Practical implication: GPs must demonstrate operational excellence and value creation capabilities to secure LP capital.

Frameworks, Models & Processes

Infrastructure Asset Classification Framework

  • How it works: Categorizes infrastructure into traditional core assets and expanding new verticals based on risk-return profile and societal importance.
  • Components:
    • Core Infrastructure (transport, utilities, toll roads)
    • Core+ / Value-Add (telecom, fiber, data centers)
    • Emerging Verticals (smart parking, microgrids, waste-to-energy)
  • When to use: When evaluating portfolio diversification and risk-adjusted returns for private market investments.

Examples & Case Studies

AT&T's gigapower fiber rollout partnering with private equity.

  • Illustrates: How large-scale infrastructure deployment relies on private capital to reach millions of homes.
  • Lesson: Private equity can accelerate capital-intensive telecom and digital infrastructure projects.

Using satellite infrared imaging to detect building heat loss across large portfolios.

  • Illustrates: How AI and remote sensing tools can dramatically improve building efficiency audits.
  • Lesson: Technology enables rapid identification and remediation of infrastructure inefficiencies.

Actionable Takeaways

  • Immediate:
    • Monitor growth in digital and energy infrastructure sectors driven by AI demand.
    • Incorporate AI-based generative scheduling tools into project management to compress timelines.
  • Strategic:
    • Recognize infrastructure as a maturing asset class requiring active value creation.
    • Align investment strategies with LP preferences for performance, team quality, and sustainability.
  • Questions to investigate:
    • How will government budget constraints impact public-private partnerships in different geographies?
    • What is the exact return potential of edge data centers versus traditional hyperscale facilities?

Claims Worth Verifying

  • Global infrastructure investment required through 2040 is estimated at $106 trillion. (statistical)
  • Failure to replace aging US infrastructure could cost the economy $10 trillion by 2039. (predictive)
  • AI-based generative scheduling can achieve 10 to 25 percent cost and schedule compression. (empirical)

Notable Quotes

"Infrastructure is having a moment." (at 0:33) "The number we're looking at, $106 trillion, that's actually double what it was the last 15 years." (at 0:18) "Without value creation, not only can you not make the most of the asset, but it's actually hard to prevail as the one who gets to buy the asset." (at 71:59)

Compressed Summary

  • Global infrastructure needs reach $106 trillion through 2040, doubling past spending.
  • Private capital is essential to bridge public sector funding gaps.
  • AI is both driving demand for data centers and optimizing project execution.
  • LPs are increasing infrastructure allocations and demanding active value creation.
  • Keywords: infrastructure, private capital, artificial intelligence, limited partners, data centers
  • Core insight: The $106 trillion infrastructure investment opportunity requires private capital and AI-driven operational efficiency to meet unprecedented global demand.

Core insights

6
Mental Modelmedium noveltymoderate evidence

AI plays a dual role in the same asset class: it is simultaneously a root driver of demand and a tool for improving project execution efficiency, which means capacity planning for AI systems cannot treat demand as an exogenous input.

Why it matters

If efficiency tools raise adoption and therefore consumption, then forecasting inference/data-center demand requires modeling a feedback loop rather than a static growth curve; the optimizer changes the curve it is optimizing.

Generalization

For any AI system that optimizes a resource it also consumes (compute, storage, human review time), demand forecasts must include an induced-demand term, not just a savings term.

AI is both a root driver of infrastructure expansion and a tool to improve project efficiency by 10% to 25%.
Open source video
Digital and energy are the single fastest-growing infrastructure sectors, driven by AI and clean energy demand.
Open source video
Practicemedium noveltymoderate evidence

Generative scheduling and generative design are being applied to constraint-dense, high-value construction programs, with claimed timeline compression and cost reduction at $200M+ project scale.

Why it matters

This is a concrete instance of model-driven planning under hard physical and contractual constraints — closer to constrained agent planning than to text generation, and it suggests where planning agents may first prove measurable ROI.

Generalization

Planning agents earn trust in domains with explicit constraints, expensive mistakes, and measurable schedules (construction, logistics, capacity rollout), not in open-ended creative tasks.

Generative scheduling and design can compress project timelines and reduce costs significantly.
Open source video
Using AI-based scheduling to manage complex $200M+ construction projects.
Open source video
Mechanismlow noveltyweak evidence

A prerequisite for AI value in asset-heavy operations is consolidating unstructured asset data into a managed substrate before higher-level analysis or agents are layered on top.

Why it matters

It names an ordering constraint: data consolidation is the enabling layer, and skipping it produces agents operating over fragmented, ungoverned source material.

Generalization

In any legacy enterprise, build the consolidation/grounding layer first; agent reasoning quality is bounded by whether asset records live in one addressable substrate.

Data consolidation using AI tools helps investors and operators manage unstructured infrastructure assets.
Open source video
Architecturemedium noveltyweak evidence

Edge data centers and sovereign AI are treated as first-class infrastructure categories, implying deployment locality — not just scale — is becoming a design axis for AI systems.

Why it matters

Locality constraints (jurisdiction, latency, grid proximity) push inference placement into the serving stack as a routing and policy concern rather than a pure capacity concern.

Generalization

Systems serving regulated or latency-sensitive workloads need a placement/policy layer decoupled from model logic, so the same model can be routed to different physical or jurisdictional tiers.

edge data centers
Open source video
sovereign AI
Open source video
Practicemedium noveltymoderate evidence

The unit of scaling in a mature capital-intensive domain is the standardized internal playbook for deployment and procurement, not the individual transaction or tool.

Why it matters

Playbooks are the operational memory that makes repeated deployments cheap; the same abstraction applies to rolling out AI systems across many sites or teams.

Generalization

Codify deployment, procurement, and evaluation into reusable runbooks — the artifact that converts one-off pilots into repeatable infrastructure.

Internal LP teams establishing standard playbooks for capex deployment and procurement.
Open source video
LPs are moving beyond just buying and holding assets to actively improving operational performance.
Open source video
Mechanismmedium noveltyweak evidence

Remote sensing is used as a detection layer whose output is an operational remediation decision (locating heat-venting inefficiencies), not an analytic report.

Why it matters

The valuable pattern is sense-detect-prioritize-act over physical assets, where the model output is consumed by a work queue rather than by a human reading a dashboard.

Generalization

Agentic monitoring of physical or production systems should terminate in a ranked, actionable work item; detection without an action binding produces no measurable value.

Satellite infrared imaging to detect heat-venting building inefficiencies.
Open source video

Deep dives

4

Induced demand vs efficiency rebound in AI-optimized infrastructure

Research question

Do AI-driven efficiency gains in infrastructure project execution and operations lead to a net increase in total resource consumption (induced demand) or a net reduction, and how should capacity forecasts model the feedback loop?

Why

If efficiency tools raise adoption and consumption, then forecasting inference/data-center demand requires modeling a feedback loop rather than a static growth curve; the optimizer changes the curve it is optimizing.

Digital and energy are the single fastest-growing infrastructure sectors, driven by AI and clean energy demand.
Open source video
the role of AI in transforming infrastructure operations
Open source video
Source video

Deferred replacement of aging infrastructure as a fragility multiplier for AI-layer deployments

Research question

How do degraded physical infrastructure assets (water, transport, energy) constrain or increase the total cost of AI-layer deployments and data center operations built on top of them?

Why

Deferred maintenance compounds into economy-scale costs, and AI-layer deployments built on degraded physical infrastructure inherit that fragility, potentially increasing downtime and operational costs.

Failing to replace aging infrastructure in the US could cost the economy $10 trillion by 2039.
Open source video
Replacing aging water and transport pipes in US municipalities.
Open source video
Source video

Category ambiguity in the expanding definition of infrastructure

Research question

How does broadening the definition of infrastructure to include agriculture, aerospace, defense, and data centers affect risk-return classification, underwriting comparability, and asset management models?

Why

Category ambiguity at the intersections makes risk-return classification and underwriting comparability harder, which can lead to mispricing and inefficient capital allocation.

Agriculture, aerospace, and defense are increasingly exhibiting infrastructure-like characteristics.
Open source video
Many investments are happening at the intersection of traditional categories, such as smart parking garages and bridge power for data centers.
Open source video
Source video

Sensitivity of the Asia infrastructure investment concentration to GDP and policy assumptions

Research question

How sensitive is the forecast that $70 trillion of $106 trillion in infrastructure investment will occur in Asia to GDP growth revisions, policy changes, and supply-chain shocks?

Why

Heavy concentration in a single region creates correlated exposure to regional policy, grid, and supply-chain shocks, so understanding the sensitivity of this forecast is critical for risk management.

70 trillion out of 106 trillion of infrastructure investment will happen in Asia.
Open source video
The majority of infrastructure investment will occur in Asia due to GDP growth, while North America and Europe face massive costs to replace and modernize aging infrastructure.
Open source video
Source video

Article ideas

4

The AI Infrastructure Feedback Loop: Why Capacity Models Must Include Induced Demand

AI's dual role as both a driver of infrastructure demand and an efficiency tool creates a feedback loop where efficiency gains induce additional consumption, so capacity planning models that treat demand as exogenous will systematically under-provision.

Angle

Arguing that forecasting must include an induced-demand term, with engineering consequences for provisioning and cost models.

Source video

Generative Scheduling Is the First Proven Agentic Planning Domain

Planning agents earn trust in domains with explicit constraints, expensive mistakes, and measurable schedules—like $200M+ construction projects—rather than in open-ended creative tasks, and the 10-25% efficiency claim should be tested against historical baselines.

Angle

Positioning constrained construction scheduling as the beachhead for agentic planning, and calling for benchmarking against historical project data.

Source video

The Placement Layer: Why Edge Data Centers and Sovereign AI Force Routing Out of Model Logic

Locality constraints (jurisdiction, latency, grid proximity) make inference placement a per-request routing and policy concern, requiring a serving layer decoupled from model logic so the same model can be routed across edge and jurisdiction-bound tiers.

Angle

Architectural argument for a placement/policy layer, with implications for observability and compliance.

Source video

Playbooks as Operational Memory: The Real Unit of Scaling in Capital-Intensive AI Deployments

Standardized deployment and procurement playbooks, not individual models or transactions, are the scalable unit in capital-intensive domains; encoding them as executable decision graphs is the next step to convert one-off pilots into repeatable infrastructure.

Angle

Argues that operational memory artifacts are the leverage point, and that automation should target playbook execution.

Source video

Project ideas

4

Induced-Demand Instrumentation for AI-Optimized Operations

beyond-evals

Deploying AI optimization in a resource-consuming operation (e.g., data center cooling, construction scheduling) increases total resource consumption relative to baseline due to induced demand, despite per-unit efficiency gains.

Proof of concept

Instrument a simulated or real operation with an AI optimizer; track total consumption (energy, cost, materials) pre- and post-deployment, alongside adoption metrics, to test for rebound effects.

Measurement

Ratio of post-deployment total consumption to baseline, controlling for scale; magnitude of rebound effect; correlation between efficiency gains and increased adoption.

Source video

Generative Scheduling Benchmark Against Historical Projects

new

Generative scheduling reduces project timeline by at least 10% and cost by 5% on $200M+ projects compared to historical baselines, but claimed gains shrink when measured against a control group of non-AI-scheduled projects.

Proof of concept

Build a benchmark suite of anonymized historical construction schedules; apply generative scheduling models; compare predicted vs actual outcomes and against a control set.

Measurement

Mean absolute percentage error in timeline and cost predictions; realized vs modeled savings; statistical significance of deltas.

Source video

Region-Aware Inference Routing Prototype

gatehouse

A placement/policy layer for inference routing across edge and sovereign tiers can meet data residency requirements with less than 5% latency overhead and less than 10% cost increase relative to centralized serving.

Proof of concept

Prototype a request router that dispatches requests to edge or region-bound inference endpoints based on jurisdiction and latency; measure performance and compliance against a centralized baseline.

Measurement

P50/P99 latency, cost per request, cache-hit rates, compliance violations, and routing decision accuracy.

Source video

Data Substrate Gating Experiment for Agent Reasoning on Asset Data

beyond-evals

Building a consolidated asset data substrate before layering agents improves decision accuracy by more than 20% compared to agents operating over fragmented, heterogeneous asset records.

Proof of concept

Set up two conditions: agents querying fragmented asset data vs. agents querying a consolidated substrate; measure task success on maintenance prioritization or asset management tasks.

Measurement

Task accuracy, retrieval precision, time-to-resolution, and error type distribution (retrieval vs. reasoning).

Source video

Architectural implications

3

Edge data centers and sovereign AI appear as distinct infrastructure categories alongside traditional digital infrastructure.

Before

Inference is assumed to be centralized in a small number of hyperscale regions.

After

Inference placement becomes a per-request routing and policy decision spanning edge and jurisdiction-bound tiers.

Consequence

Serving stacks need a placement/policy layer decoupled from model logic, plus per-region observability for latency, cost, and data-residency compliance.

Source video

AI is described both as the consumer of infrastructure and the tool that improves infrastructure project execution.

Before

Demand for compute and capacity is forecast as an independent variable in planning models.

After

Efficiency gains from AI feed back into adoption and consumption, so planning models must include an induced-demand term.

Consequence

Capacity and cost models that assume linear savings will systematically under-provision; feedback-aware forecasting becomes an engineering requirement, not a finance nicety.

Source video

Data consolidation is positioned as the enabler for managing unstructured infrastructure assets.

Before

Analysis and decision tooling is built directly against fragmented, heterogeneous asset records.

After

A consolidation and grounding layer is built first, with higher-level reasoning tooling layered on top of it.

Consequence

Dependency ordering changes: data substrate work gates agent work, and skipping it shifts failure from model quality to retrieval quality.

Source video

Tradeoffs and failure modes

4

Unverified efficiency ranges entering investment decisions

Benefit

A 10-25% efficiency claim is concrete and directly useful for underwriting and prioritization.

Cost or risk

The summary provides no measurement methodology, baseline, or scope, so the range may be a modeled estimate treated as an empirical result.

AI is both a root driver of infrastructure expansion and a tool to improve project efficiency by 10% to 25%.
Open source video
Source video

Geographic concentration of investment

Benefit

Concentrating capital where GDP growth is fastest improves returns and speeds deployment.

Cost or risk

Heavy concentration in a single region creates correlated exposure to regional policy, grid, and supply-chain shocks.

70 trillion out of 106 trillion of infrastructure investment will happen in Asia.
Open source video
Source video

Deferred replacement of aging assets

Benefit

Deferring capital expenditure preserves near-term cash and avoids disruption.

Cost or risk

Deferred maintenance compounds into economy-scale costs, and AI-layer deployments built on degraded physical infrastructure inherit that fragility.

Failing to replace aging infrastructure in the US could cost the economy $10 trillion by 2039.
Open source video
Source video

Broadening the definition of infrastructure

Benefit

New verticals (data centers, space assets, agriculture, defense) expand the investable universe.

Cost or risk

Category ambiguity at the intersections makes risk-return classification and underwriting comparability harder.

Many investments are happening at the intersection of traditional categories, such as smart parking garages and bridge power for data centers.
Open source video
Source video

Open questions

4

How was the 10-25% project efficiency range measured, and over what baseline and sample?

Why unresolved

The summary states the range without method, scope, or comparison group.

Research direction

Build a benchmark of generative scheduling against historical project schedules with cost and timeline deltas, separating modeled gains from realized gains.

Source video

Do AI-driven efficiency gains reduce total resource consumption or induce additional demand?

Why unresolved

The summary asserts AI as both demand driver and efficiency tool but does not state whether the two effects net out or compound.

Research direction

Instrument deployed AI optimization tools and track post-deployment consumption against a pre-deployment baseline to test for rebound effects.

Source video

What concrete design constraints do edge data centers and sovereign AI impose on model serving and training pipelines?

Why unresolved

These appear only as terminology in the summary, with no mechanism, latency figure, or compliance rule specified.

Research direction

Prototype region-aware request routing with per-jurisdiction data handling and measure the latency, cost, and cache-hit penalties relative to centralized serving.

Source video

Can the standardized capex deployment and procurement playbook be encoded as an executable agent workflow?

Why unresolved

The summary notes LPs establishing playbooks but does not describe their contents or structure.

Research direction

Extract a representative playbook as a decision graph with approval gates, then measure where automation is blocked by data or policy rather than capability.

Source video

Key claims

8
factualVerification needed

More than $100 trillion in infrastructure investment is needed through 2040, double the level of the previous 15 years.

Evidence

The $106 trillion figure is double the investment level of the last 15 years.

Question

What baseline period, sector scope, and deflator were used to produce the $106 trillion estimate?

Source video
causalVerification needed

AI improves infrastructure project efficiency by 10% to 25%.

Evidence

AI is both a root driver of infrastructure expansion and a tool to improve project efficiency by 10% to 25%.

Question

Is the 10-25% range from measured deployments or from modeled projections, and which project phases does it cover?

Source video
causalVerification needed

Generative scheduling and design compress project timelines and reduce costs significantly.

Evidence

Generative scheduling and design can compress project timelines and reduce costs significantly.

Question

What magnitude of timeline and cost reduction was observed, and against what comparison baseline?

Source video
predictionVerification needed

$70 trillion of the $106 trillion in infrastructure investment will occur in Asia.

Evidence

70 trillion out of 106 trillion of infrastructure investment will happen in Asia.

Question

What growth and policy assumptions drive the Asia concentration, and how sensitive is it to GDP revisions?

Source video
predictionVerification needed

Failing to replace aging US infrastructure could cost the economy $10 trillion by 2039.

Evidence

Failing to replace aging infrastructure in the US could cost the economy $10 trillion by 2039.

Question

Which cost categories are included in the $10 trillion figure, and is it a counterfactual or observed projection?

Source video
factualVerification needed

Infrastructure closed-end funds raised nearly $200 billion in the last calendar year.

Evidence

Infrastructure closed-end funds reached nearly $200 billion in fundraising in the last calendar year.

Question

Which fund universe and reporting period are included in the $200 billion fundraising figure?

Source video
comparativeVerification needed

Limited partners now prioritize value creation and operational improvement over passive asset holding.

Evidence

LPs are moving beyond just buying and holding assets to actively improving operational performance.

Question

Does LP survey data show a measured shift in allocation criteria, or is this an interpretive observation?

Source video
opinionVerification not requested

Public sectors alone cannot raise the capital required, creating the opening for private capital.

Evidence

Public sectors alone cannot raise this capital, creating an opening for private capital.

Source video

Connections

5