Key Questions: Can the Promise of AI Support Today’s Record Capital Spending?

You Have to Spend Money to Make Money

Businesses of all kinds require investment. That is why debt and equity markets exist: to provide the capital needed to grow businesses, expand operations, and foster innovation. The basic math an investor must do is weigh the investment required to launch a venture against the potential profits that venture may generate in the future, a concept commonly summarized through return on invested capital (ROIC). Investors care about ROIC because returns must exceed the cost of capital employed to pursue a given venture. Otherwise, why undertake the investment at all? This logic is central to modern business and investing.

When it comes to investments being made in artificial intelligence (AI) infrastructure, the numbers are truly eye-popping. According to an analysis by Bloomberg Intelligence, cumulative spending through 2032 by the five largest hyperscalers could reach $4 trillion. McKinsey estimates AI infrastructure spending at more than $6.5 trillion between 2025 and 2030. These estimates encompass spending on data centers, power and energy infrastructure, semiconductor manufacturing, and server components.

Many investors are now focused on the other side of the equation. They see the enormous costs and are beginning to ask what returns can ultimately be generated from such a massive investment. We will not attempt to pinpoint a specific number, but there are already clear examples where certain tasks have experienced cost and time reductions exceeding 90%, unleashing tremendous productivity stemming from massive investment.

When considering the impact of this potential AI infrastructure spending, it is important to widen our perspective beyond technology stocks. Consider a data center, where land must be graded with construction equipment, foundations poured, electricity generated and transmitted, and cooling systems installed. Then consider what goes inside the facility: chips, server racks, networking equipment, and miles of cable. Taking it a step further, many of the materials needed to produce cement, cables, power infrastructure, and chips originate from copper, gold, and silver mines; limestone quarries; and oil wells. We could continue tracing the supply chain, but the point is clear: the investment required across industries is both massive and potentially paradigm shifting.

The Risks Underlying the AI Investment Boom

Competition among hyperscalers has become a primary driver of the unprecedented investment cycle surrounding artificial intelligence. Surging demand for AI computing capacity has effectively created a "build now or fall behind" environment for leading tech firms, which has accelerated spending to record levels.

The scale and urgency of AI infrastructure development require substantial upfront capital, creating significant investment risk if such capital does not achieve sufficient returns. Perhaps the most notable risk is that much of this spending is being deployed in anticipation of an end market that has yet to fully emerge. The rationale for expanding AI capacity rests on the expectation that AI services will achieve widespread adoption and generate meaningful economic returns, and while such returns may ultimately materialize, the evidence to date remains mixed.

On the one hand, firms like Amazon and Microsoft have reported impressive results in their AI and cloud compute businesses, suggesting robust demand for these services. On the other hand, a large share of this revenue is being generated by a handful of technology companies rather than by broad-based enterprise adoption. A recent Bloomberg article underscored the narrow customer base supporting today's AI boom, reporting that the majority of Microsoft's AI revenue came from a short list of technology giants, headlined by OpenAI, Meta, and ByteDance. In short, while AI models continue to see rapid improvement and steadily expanding use cases, the current investment cycle remains heavily dependent on future adoption and monetization. As a result, relying on anticipated returns to justify today's elevated capital spending is a significant risk.

Even if artificial intelligence ultimately achieves broad economic adoption, funding the infrastructure necessary to support that vision requires an extraordinary commitment of capital. This raises a second, equally important question: how are companies financing these enormous investments?

Historically, many large technology firms funded expansion primarily through internally generated cash flow. Indeed, most leading hyperscalers have been able to leverage profits from other segments of their businesses to fund AI-related projects. As investment requirements have accelerated, however, a growing share of spending is being supported by equity and debt issuance, leasing arrangements, and other complex financing structures, with capital expenditures now eclipsing the cash flow available to fund them. It is no coincidence that Alphabet raised capital twice this year — $32 billion in bond sales in February and $85 billion in equity issuance in June — ahead of the company's second-quarter earnings report, which revealed the business had generated negative free cash flow for the first time since its 2004 initial public offering.

Issuance is only the tip of the iceberg. Much of the financing supporting the current capital expenditure cycle is now being arranged through complex financial structures that reduce visibility into the true extent of AI-related spending.

For example, in March, CoreWeave raised an $8.5 billion financing facility secured by a combination of GPU inventory and a contract with Meta reportedly worth approximately $19 billion. Had Meta directly funded the underlying compute infrastructure itself, the investment would have added billions of dollars to an already substantial capital expenditure figure.

Instead, Meta can claim it is sourcing compute capacity from CoreWeave, even though CoreWeave is only able to raise the required capital because of Meta's long-term contractual commitment. This structure allows Meta to recognize costs over time through contractual payments rather than as direct infrastructure capital expenditures, even though the underlying economic commitment remains largely unchanged.

This kind of workaround has become common practice for firms leaning into the AI buildout. An analysis from The Wall Street Journal estimated that the top nine technology companies had a combined $3 trillion of off-balance-sheet commitments tied to AI, a number dwarfing their combined capex figures. While the scale and financing of the AI buildout warrant careful scrutiny, the investment case ultimately rests on whether the resulting gains in productivity and profitability justify the upfront costs. Should such gains fall short, the companies raising vast amounts of capital would undoubtedly see their stock prices come under pressure.

Is This Time Actually Different?

Back in 2022, concerns arose regarding the significant increase in capital expenditures across the largest technology companies. The market questioned whether these massive investments would be able to generate sufficient returns. Fast forward to today, cloud platforms such as Amazon Web Services, Microsoft Azure, and Google Cloud have become some of the most valuable business franchises in the world, serving as the foundational infrastructure layer for enterprise computing and artificial intelligence.

Alphabet, Amazon, Meta, and Microsoft collectively invested roughly $360 billion in capital expenditures during fiscal 2025, continuing a trend of aggressive infrastructure spending. Building global cloud platforms requires enormous investments in data centers, networking equipment, semiconductors, and power capacity. Once these platforms reach scale, they benefit from significant operating leverage, recurring revenue streams, and high switching costs, helping them to generate attractive incremental returns over time.  As we look ahead, technology companies are once again deploying unprecedented levels of capital toward artificial intelligence. Unlike prior investment cycles, these expenditures are being made from a position of strength.  What could be viewed as excessive spending is increasingly recognized as the primary vehicle of value creation for hyperscalers.

If AI workloads become as deeply embedded within enterprises as cloud computing has over the past decade, today's investments could represent the next chapter of high-return infrastructure spending. While the magnitude and timing of those returns remain uncertain, history suggests that transformational technology platforms often require years of heavy investment before their economics become fully apparent.  Rather than simply a cost, AI investments today could serve as the foundation for strong long-term value creation for shareholders. Risks are real, however, and thus we continue to recommend a balanced approach by investing in both “AI enablers” and “AI adopters."

For more information, please contact your advisor.