The $106 Trillion Classroom
Behind every digital breakthrough is a physical world that has to work. Learning how to improve that world may be one of the most useful things you can do next.
A number big enough to change what you study
The number that stopped me in McKinsey’s Global Private Markets Report 2026 was $106 trillion. That is the estimated cumulative investment needed to meet global infrastructure requirements through 2040. It includes the systems behind everyday life and the systems behind the next wave of technology. [1]
For my students, the useful response is to get specific. Where does a project stall? What information is missing? Who has to make the next decision? What would make a water system, a construction schedule, or an electrical network work better?
Those questions turn an overwhelming number into a place to begin learning.
First, get the comparison right
There is a real story of more than doubling here, but we need to attach it to the right measure. Exhibit 2 shows annual global infrastructure fundraising rising from $83 billion in 2015 to approximately $200 billion in 2025. Using those rounded figures, that is about 2.4 times the earlier level, or a 141 percent increase. It was an uneven decade, including a sharp decline in 2023. [1]
That does not establish that the $106 trillion estimate itself doubled in ten years. The report does not provide a comparable historical series that would support that claim. A separate McKinsey study reports another distinct measure: private infrastructure assets under management increased from about $500 billion in 2016 to $1.5 trillion in 2024—roughly tripling. Assets managed, funds raised, and future investment needs measure different things. [2]
This is your first assignment whenever a large number appears in a headline: identify its unit, time period, scope, and source before explaining what it means.
Need → capital → delivery
A useful project must travel through all three.
What should exist?
A community needs reliable water. A region needs more electrical capacity. A business needs dependable computing.
Who can fund it?
Someone must commit resources, accept risk, and establish how construction and operation will be paid for.
Can it actually work?
Land, approvals, equipment, people, and operating processes must come together at the right time.
The physical world inside your AI lesson
Consider an AI service. A user sees an answer. Behind that answer are chips, a building, electrical equipment, network connections, and a cooling system. Behind those are contracts, construction teams, maintenance work, and decisions about location and capacity.
The International Energy Agency’s 2025 Energy and AI report projected that global data-center electricity consumption would rise from about 415 terawatt-hours in 2024 to around 945 in 2030. It also estimated that grid constraints could put roughly 20 percent of planned data-center projects at risk of delay unless addressed. These are projections and conditional risks, not completed outcomes. [3]
My teaching takeaway is that the ability to generate software quickly makes understanding its dependencies more valuable. A tool can be beautifully built and still be useless to an operator if the necessary data arrives late, the equipment cannot connect, or nobody trusts the output enough to act on it.
McKinsey’s infrastructure chapter describes growing overlap among energy, digital systems, transportation, and services. It also describes AI applications in project scheduling, site selection, permitting analysis, and maintenance. These are examples of work under development, not promises that every deployment will succeed. [1]
Customers want more computing.
That establishes demand to investigate. It does not establish a viable site, a signed customer, or sufficient power.
An investor commits funding.
That makes resources available. It does not put a transformer on site or secure a grid connection.
The facility can serve customers.
Power, cooling, connectivity, staffing, and approvals work together. Measure capacity delivered and reliability, not just money announced.
A need is not automatically a business
The $106 trillion estimate should not be treated as an unfunded balance, an approved spending program, or revenue waiting for any company that adds “infrastructure” to its presentation. Although the report’s introduction uses the word “gap,” the infrastructure chapter describes cumulative investment requirements. Establishing a financing shortfall would require a separate estimate of available funding. [1]
The underlying 2025 research combines historical baselines with economic modeling and additional estimates for less-covered sectors. Its assumptions include moderate inflation and stable trade terms through 2034. A forecast that far out is a planning framework, not a precise schedule. [2]
For a student evaluating an idea, the next questions are concrete: Who experiences the problem? Who controls the budget? Who can approve a change? What must be true for the solution to be adopted? A useful social outcome and a financially viable project can overlap, but you must explain how.
You do not have to own the infrastructure to improve it
My inference from this research is that students should look closely at the work surrounding large assets. A utility, contractor, facility manager, or equipment supplier has workflows that can be studied without raising an infrastructure fund.
Here are four possible learning projects. These are proposed classroom exercises, not documented case studies or claims of proven savings.
Find the missing requirement
Use a public project packet to create a checklist that links every requirement to its source page. Have a knowledgeable reviewer test it. Measure missed requirements and unsupported claims. The goal is reliable retrieval, not automatic approval.
Make a schedule explain itself
Build a small, hypothetical construction schedule. Introduce a late equipment delivery and trace which tasks actually move. Compare the model with a manually checked dependency map. A faster answer is useful only if the dependencies are right.
Turn maintenance records into usable evidence
Use synthetic work orders to classify recurring failures and identify missing information. Measure classification errors against a reviewed sample. Keep the distinction between summarizing a record and diagnosing an actual machine.
Improve a handoff
Map the path from an inspection finding to a repair assignment. Design a tool that carries the location, evidence, owner, and next action together. Test whether a new reader can identify the responsible person and the unresolved issue without another meeting.
Choose one system. Find one delay. Test one improvement.
- Observe: Choose a campus building, public project, or local service. Use public material or information you have permission to use.
- Map: Identify the user, decision-maker, dependencies, and one recurring failure or delay.
- Establish a baseline: Record how the task works today. Choose one measure, such as retrieval time, omissions, or handoff completeness.
- Prototype: Build the smallest tool that addresses that problem. Include source links and a human review step where judgment matters.
- Evaluate: Test against the baseline. Report errors, limitations, and whether the improvement survives review by someone who understands the work.
Your deliverable is a short demonstration and a one-page evidence sheet. “The AI answered” is not a result. “The reviewer found fewer missing requirements in the same packet” is a result you can investigate.
Learn the language of the people doing the work
Students who choose this path need more than familiarity with AI tools. They need enough domain knowledge to ask good questions, enough data discipline to distinguish evidence from inference, and enough operational understanding to see what happens after a recommendation is made.
For an engineer, that might mean learning how procurement decisions shape a project. For a business student, it might mean understanding the physical constraint behind a forecast. For a designer or programmer, it might mean spending time with the person who must use the system under real conditions.
The report’s emphasis on operational improvement is useful here. Buying an asset and improving its performance are different kinds of work. As projects become more complex, the latter demands close attention to how things actually function. [1]
The $106 trillion figure is a reason to look up from the screen and study what the world needs to make possible. Start with one real system. Understand it well enough to make one part of it work better.
Go to the evidence
- McKinsey, infrastructure chapter of Global Private Markets Report 2026. Primary source: the supplied full report, printed pp. 56–66 (PDF pages 57–67). The $106 trillion figure appears on p. 56; fundraising history is Exhibit 2, p. 58; operational and AI applications appear on pp. 65–66. The 2.4× and 141% comparisons are calculations from rounded chart values.
- McKinsey, The infrastructure moment, September 9, 2025. Underlying investment-needs research, methodology, and the separate 2016–24 assets-under-management comparison.
- International Energy Agency, Energy and AI, executive summary, 2025. Independent context on electricity demand and potential grid-related delays. Forecast figures retain their original projection dates.
Interpretation and classroom exercises by John Rector. Figures describe their cited periods; the article does not present them as real-time measurements.