The AI Boom on Borrowed Money: Billions for Data Centers—But Where Is the Cash Flow?

In brief

The AI boom is no longer just a software story. It is built from data centers, chips, cooling systems, grid connections and vast amounts of capital. Oracle, Dell and TSMC demonstrate both how real the demand is and how expensive the build-out has become.

The key question for investors is therefore changing. It is no longer only Who is growing fastest? It is increasingly Who can turn that growth into sustainable free cash flow?

AI requires concrete, copper and billions

A chat window looks lightweight. The infrastructure behind it is anything but.

Modern AI models require specialized processors, high-speed networks, memory, cooling and continuously available electricity. Before a provider earns the first dollar from new data-center capacity, it may have to invest billions.

That is why “the AI boom on borrowed money” is deliberately provocative. Not every expansion is literally funded with debt. Companies use operating cash flow, customer prepayments, bond issuance or a mixture of funding sources. The economic question is the same: Does the money come back quickly enough?

Oracle: rapid growth, even faster investment

Oracle currently provides one of the clearest examples of this tension. Trailing-twelve-month revenue grew by roughly 17%, while capital expenditure over the same period recently reached about $55.7 billion.

In the quarter ending June 2026, Oracle generated approximately $14.6 billion of operating cash flow but spent around $16.5 billion on capital expenditure. Quarterly free cash flow was therefore negative. Over the trailing twelve months, free cash flow stood at roughly negative $23.7 billion.

That does not automatically make the strategy a bad one. Oracle’s cloud infrastructure business is growing rapidly, and large AI contracts create a substantial base of potential future revenue. Customer prepayments can also help fund part of the expansion. But the risk is shifting. The question is becoming less about whether demand exists and more about whether capacity utilization and pricing will eventually produce an adequate return.

Put less politely: a full order book is attractive, but the power company still does not accept it as payment.

Dell: $95 billion is not revenue yet

Dell offers another view of the extraordinary demand. Its backlog of AI-server orders recently reached approximately $95 billion. For fiscal 2027, the company expects around $74 billion of revenue from AI-optimized servers. Total revenue in its latest quarter reached roughly $47 billion.

This confirms that enterprises and cloud providers are ordering physical hardware at real scale. The demand is more than a slide in an investor presentation.

However, backlog represents future business, not revenue or profit already recognized. Delivery speed, chip and memory availability, gross margins and working-capital requirements will determine how valuable those orders ultimately are. Rapidly rising server revenue can produce surprisingly modest returns if components remain expensive and margins come under pressure.

TSMC: the boom is already reaching the income statement

TSMC occupies an unusually strong position in the value chain. It manufactures advanced chips for several competing designers, so it does not have to bet on only one architecture or customer becoming the ultimate winner.

August revenue reached NT$514.8 billion, approximately 53% higher than a year earlier. Its latest reported second-quarter gross margin of about 67.7% also indicates that demand has so far produced not only volume but strong profitability.

TSMC therefore demonstrates that AI infrastructure spending is already generating real revenue and earnings. Yet foundries must plan factories years in advance and commit enormous sums before demand is fully visible. Geopolitical exposure, the cost of international expansion and the possibility of overbuilding remain material risks.

Why negative free cash flow is not automatically bad

Free cash flow is operating cash flow minus capital expenditure. When a company builds a data center today, the expenditure hits cash flow immediately, while the associated revenue may arrive over many years.

Negative free cash flow can therefore be rational when three conditions hold:

• The new capacity is actually utilized.
• Pricing and margins justify the upfront investment.
• Funding remains manageable even if projects are delayed.

The model becomes dangerous when providers build too much capacity, customers postpone projects or new hardware makes older systems obsolete faster than expected. High capital needs combined with falling compute prices can make impressive revenue growth look much less romantic.

Three possible scenarios

Bull case: AI moves from model training into broad, continuous use. Data centers remain highly utilized, customers accept premium pricing and today’s investment wave generates rising cash flows for years.

Base case: Demand stays strong, but competition intensifies. Revenue grows quickly while depreciation, electricity costs and continued investment suppress free cash flow for longer. Good companies still make money, but not every ambitious valuation survives.

Bear case: Customers have reserved more capacity than they ultimately need. Hardware generations turn over faster, compute rental prices decline and financing costs meet weaker utilization. What looked like a moat can become very expensive technology from yesterday.

What investors should monitor

Investors assessing AI-related stocks should look beyond revenue growth. Six indicators matter in particular:

• CapEx relative to operating cash flow: Can the company fund expansion internally?
• Free cash flow: Is the decline temporary or structural?
• Backlog conversion: Are orders delivered and paid for on schedule?
• Gross and operating margins: Does real pricing power survive the growth phase?
• Debt and interest burden: How much flexibility remains if projects are delayed?
• Customer concentration: What happens if one major buyer reduces spending?

SmartDeer makes it possible to examine these layers together: growth, margins, cash flow, balance-sheet strength and valuation. The best AI narrative offers little protection if the market has already priced in perfect utilization until the next ice age.

Perspective

The AI infrastructure boom is real. Oracle is investing at extraordinary scale, Dell has accumulated an exceptional server backlog and TSMC is reporting strong revenue. The central investor question is shifting from “Is the demand real?” to “Who can earn sustainable returns after all the investment?”

The eventual winners will not necessarily be the companies with the largest spending budgets. They are more likely to be those that combine high utilization, pricing power and disciplined financing. Growth attracts attention. Cash flow pays the bill.

This article is for information and analysis only and does not constitute investment advice.