Artificial intelligence is attracting capital on a scale that few technologies have seen before.
Apollo estimates that roughly $5 trillion will be spent on AI infrastructure between now and 2030. Its more striking number is on the other side of the equation: businesses and consumers would need to spend around $2 trillion a year on AI services to justify that investment.
That does not mean the investment is misplaced. It does highlight a basic principle that can get obscured during periods of rapid technological change.
Eventually, capital has to earn a return. The same test applies whether the asset is an AI data centre, a factory, a logistics network or an established operating business.
AI has already moved far beyond software spending.
It requires chips, data centres, electricity generation, grid connections, cooling systems, networking equipment and construction. McKinsey estimates global data centre investment could reach almost $7 trillion by 2030, with more than $4 trillion directed towards computing hardware.
The numbers continue to grow. In September, McKinsey reported that five major hyperscalers have announced around $800 billion of capital expenditure for 2026 and $1 trillion for 2027, with the majority expected to support AI infrastructure.
BlackRock puts expected AI related capital expenditure through 2030 at between $5 trillion and $8 trillion. Its research identifies the same question now moving towards the centre of the investment debate: whether the revenues ultimately generated by AI can match the scale of the capital being committed.
That is where an investment theme becomes an investment case.
There are already signs that parts of the AI ecosystem are producing strong financial results.
BlackRock reports that companies in its AI infrastructure grouping recorded 54% year on year earnings growth in the second quarter of 2026, compared with 14% for the S&P 500 excluding those companies.
But strong demand at one level of the value chain does not automatically prove that every investment built around that demand will generate attractive returns.
A data centre needs sufficient utilisation. New computing capacity needs customers willing to pay for it. Enterprise AI applications need to create enough productivity or revenue improvement to justify recurring expenditure. Infrastructure financed today needs future cash flows large enough to cover operating costs, financing costs and the original capital invested.
McKinsey identifies real return on investment for enterprise customers as one of the conditions required for its semiconductor growth assumptions to hold. If customers fail to see tangible economic benefits from AI, lower spending could eventually feed back into infrastructure investment.
This creates a chain that investors increasingly need to examine from both ends. Capital expenditure creates capacity. Capacity requires demand. Demand has to create value for customers. That value eventually has to become revenue and cash flow somewhere in the system.
Technological importance and investment quality are different questions.
A technology can transform an industry while some of the businesses financing, supplying or commercialising that transformation produce disappointing returns. Previous investment cycles have repeatedly shown that large markets can attract more capital than the underlying economics can support.
For investors, this makes the quality of the cash flow increasingly important as spending accelerates.
- Who is paying?
- How recurring is the revenue?
- What level of utilisation is required?
- How much additional capital will be needed before the asset becomes productive?
- How long does the infrastructure remain economically useful?
- And after operating expenses, financing and reinvestment, how much cash actually remains for the owners?
These questions become more important as the capital requirement increases.
The AI buildout provides an unusually large example, but the underlying investment discipline is universal.
A business generating £10 million of additional revenue after requiring £100 million of new capital can be less attractive than a slower growing business that generates reliable cash with limited incremental investment.
Likewise, rapid revenue growth can look impressive while producing little shareholder value if maintaining that growth requires constant capital expenditure, additional borrowing or repeated equity funding.
The relationship between growth, capital invested and cash generated ultimately matters more than any one of those numbers in isolation. This is particularly relevant when evaluating established operating businesses and cash generating projects.
Revenue visibility, margins, cash conversion, maintenance requirements and the ability to return capital can provide a much clearer picture of economic value than growth rates alone.
The infrastructure buildout is still accelerating. BlackRock expects investment in data centres and AI chips alone could exceed $700 billion annually by 2030.
The next stage of the discussion is therefore likely to focus increasingly on what all that infrastructure produces. Apollo describes whether AI can earn back the capital being spent on it as one of the defining questions of the current investment cycle.
That question is useful well beyond AI. The technology can change. The scale can change. The source of revenue can change. The fundamental test does not.
How much capital goes in, how much cash eventually comes out and how durable is the return in between?
For investors, that remains one of the clearest ways to separate an exciting market from an attractive investment.
☑️ Apollo, September 2026: Can AI Earn It Back? The $2 Trillion Test and Other Forces Reshaping Markets and Portfolios
☑️ McKinsey, September 2026: The $2.3 Trillion Horizon: How AI Is Rewriting the Semiconductor Story
☑️ BlackRock, 2026: 2026 Fall Investment Directions
☑️ BlackRock Investment Institute: 2026 Global Outlook
☑️ McKinsey, March 2026: The $7 Trillion Data Center Build Out
