When AI Borrows from Wall Street: The "CapEx Cycle" of Tech Giants Accelerates Financialization

Over the past two years, when Wall Street discussed AI, the most important metrics have consistently been GPUs, data centers, and CapEx.

But by 2026, another number has begun rapidly entering the market's field of view: debt.

Once upon a time, the most profitable companies in Silicon Valley were accustomed to using their own cash to purchase GPUs and build data centers. Today, as the scale of AI infrastructure continues to march toward hundreds of billions or even trillions of dollars, even companies with the strongest cash generation capabilities globally—Alphabet, Amazon, Meta—are increasingly turning to the bond market.

Morgan Stanley projects that global AI-related debt issuance could approach $570 billion in 2026, more than double last year's figure; as of the end of May, issuance had already reached approximately $236 billion, four times the level seen during the same period last year.

Meanwhile, Alphabet, Amazon, Microsoft, and Meta are expected to spend approximately $700 billion on related expenditures this year, and by 2027, hyperscaler capital expenditures could even surpass $1 trillion.

And the financing methods are continuing to expand outward.

• On August 10, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, aiming to mobilize over $500 billion in third-party capital into AI infrastructure.
• Around the same time, Microsoft, Meta, Oracle, Amazon, and Alphabet disclosed future lease payment commitments not yet in execution, already reaching the trillion-dollar scale;

When even the wealthiest companies on the planet begin changing their financing methods, it signals that the AI story is entering a new phase. That raises the question: these companies used to sit on tens of billions in cash—why are they suddenly so fond of borrowing?

1. AI's Cash Burn Scale Is Becoming Increasingly Extreme

Alphabet is the best example for understanding this shift.

From a business perspective, its latest quarter was almost undeniably strong—Q2 revenue reached $119.8 billion, up 24% year-over-year; Google Cloud revenue hit $24.8 billion, a remarkable 82% year-over-year increase.

But on the other side, Alphabet's quarterly capital expenditures have already reached approximately $44.9 billion, resulting in a situation where even with business still growing rapidly, the company posted its first quarter of negative free cash flow, with Q2 free cash flow falling to -$5.9 billion. At the same time, Alphabet raised its 2026 capital expenditure guidance to $195 billion–$205 billion.

This is precisely the biggest difference between AI capital expenditures and the traditional software era.

Large tech companies in the software era were essentially cash machines: after the initial R&D phase, the marginal cost of adding each additional user was limited, and a significant portion of revenue could settle into free cash flow.

In other words, AI is making tech companies "heavy" again.

GPUs, servers, high-speed networking, data centers, substation infrastructure, cooling systems, and land all require massive cash outlays before revenue is actually generated.

Amazon CEO Andy Jassy once explained that data centers typically begin generating construction expenses about two years before they become operational, while revenue cannot be realized until the facilities actually come online.

This creates a natural maturity mismatch of capital: cash must be spent today, but revenue will only trickle in over many future years. Under these circumstances, even if a company holds substantial cash on its balance sheet, relying entirely on internal cash flow for financing may not be the most rational choice.

Alphabet completed approximately $31.5 billion in global bond financing in February this year, including a rare 100-year bond; in August, it issued another $25 billion in investment-grade USD bonds. Amazon has been even more aggressive—raising approximately $37 billion in the US bond market in March, followed by a €14.5 billion bond issuance the very next day, totaling close to $54 billion, and then another $25 billion in USD bonds in July.

Meanwhile, Amazon has raised its 2026 capital expenditure plan to $220 billion. AWS revenue grew 37% year-over-year in the latest quarter, the fastest pace in over four years, but free cash flow over the past 12 months has still deteriorated from positive $18.2 billion a year ago to -$7.6 billion.

Meta is showing the same trend. In April this year, Meta completed a $25 billion bond issuance; Q2 revenue still grew 28% to $60.8 billion, but free cash flow plummeted from $8.55 billion in the same period last year to just $784 million. Its 2026 CapEx guidance has been raised to $130 billion–$145 billion.

These companies haven't suddenly lost their ability to generate profits—they're still earning plenty—it's just that the cash they can freely allocate is shrinking.

This is why, in the AI era, free cash flow is becoming a metric that is harder to ignore than EPS alone.

2. Borrowing Is the Same, But Google and Oracle Are Entirely Different Stories

But debt itself doesn't necessarily mean danger.

For companies like Alphabet and Amazon, debt is more of a capital structure tool.

They have massive core businesses, stable cash flows, and high credit ratings. Given the extreme front-loading of capital expenditures, spreading construction costs into the future through long-term bonds represents normal maturity matching.

What truly requires attention is whether, when capital expenditures begin persistently exceeding a company's own cash generation capacity, financing is optimizing the balance sheet or starting to put pressure on it.

Oracle is one of the most extreme examples right now.

Through the end of fiscal year 2026, Oracle's full-year capital expenditures reached approximately $55.66 billion, while operating cash flow was only about $32 billion, driving full-year free cash flow to -$23.69 billion. Meanwhile, the company completed approximately $43 billion in debt financing and $5 billion in equity financing during FY2026, with total future principal on all borrowings reaching approximately $130.1 billion as of the end of May.

On July 9, S&P Global Ratings downgraded Oracle's long-term credit rating from BBB to BBB-—the lowest tier within investment grade, with the next notch down entering speculative grade.

So, while both are burning cash on AI, Alphabet is leveraging its balance sheet, whereas Oracle has begun challenging its balance sheet. This is precisely the new analytical framework that must be established for the next phase of the AI trade.

Previously, the market's primary focus was on AI revenue growth, order growth, and cloud business growth rates. Going forward, additional questions need to be asked: How much did companies spend to achieve this growth? How much additional CapEx is required for every $1 of incremental revenue? How much free cash flow remains? How much debt needs to be raised? Once interest, depreciation, and leasing costs all flow into the income statement, how much profit is ultimately left?

At the end of the day, what truly determines valuations is not growth alone, but the capital efficiency of that growth.

3. The Bigger Change Beyond Debt Issuance: AI Is Becoming a Financial Asset

If Alphabet, Amazon, and Meta issuing bonds merely represents a shift in financing methods, then Nvidia's latest move takes things a step further.

On August 10, Nvidia announced strategic partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, aiming to establish an independent computing financing platform that would mobilize over $500 billion in third-party capital into AI infrastructure over the long term.

Nvidia itself holds an option to provide up to approximately 25%—up to roughly $125 billion—as a backstop for potential deals, although the specific commitments from each institution and capital deployment timelines have not yet been announced.

What's truly noteworthy is how Nvidia defines this model. In its official announcement, Nvidia directly describes AI Compute and AI Factory as a new "investable asset class."

The logic isn't complicated. In the past, a Neocloud needing to purchase billions of dollars in GPUs would first have to raise substantial capital on its own. Going forward, this structure will increasingly resemble traditional infrastructure financing: data centers and GPUs form the assets, customers sign long-term compute leases that generate cash flow, institutions like Apollo, BlackRock, and KKR provide long-term capital, and projects repay financing costs using future compute revenue.

As a result, a GPU is no longer just a chip sold and forgotten. Instead, the entire AI Factory—comprising GPUs, data centers, power, and long-term compute contracts—is being packaged as an infrastructure asset capable of generating long-term cash flows, one that can be priced and financed by capital markets.

The significance of this is enormous.

Because once AI infrastructure enters the investment scope of pension funds, insurance capital, private credit, infrastructure funds, and asset managers, the capital pool AI can draw upon is no longer limited to tech company cash.

This could allow the current AI infrastructure buildout to last longer than the market expects.

But at the same time, it will change the shape of risk. After all, the biggest risk in the first phase of the AI cycle was that nobody was using AI. But in this latest second phase, the more concerning risk might be that AI is clearly being used, yet compute prices are falling too fast—revenue growth can't keep up with debt, depreciation, and financing costs.

Especially when GPU iteration speeds remain extremely fast, today's asset return models—designed around five-year or even longer cycles—must rest on a critical assumption: that these devices can maintain sufficiently high utilization rates and economic value over the coming years.

This is the new variable the AI industry must confront as financial capital truly begins to enter.

Beyond this, there's another capital commitment that's easier to overlook.

According to Reuters' analysis of company filings, Microsoft, Meta, Oracle, Amazon, and Alphabet have disclosed approximately $1.09 trillion in future lease payment commitments not yet in execution—with Microsoft at approximately $329.1 billion, Meta at approximately $279 billion, Oracle at approximately $260 billion, Amazon at approximately $137.2 billion, and Alphabet at approximately $85.2 billion. Meta subsequently signed approximately $68 billion in new data center lease agreements in July, pushing the known total further up to approximately $1.16 trillion.

Of course, these numbers shouldn't be simplistically interpreted as "tech companies already owe $1.16 trillion." Many contracts span more than a decade and have not yet formally begun execution, so they haven't fully entered balance sheets as lease liabilities. Amazon's disclosures also include warehouses, offices, aircraft, and vehicles—not all of which belong to AI data centers.

But it still reveals something important: a significant portion of AI infrastructure investment over the next few years has already been locked in ahead of time. As long as compute demand continues to grow, these commitments serve as the foundation for future revenue growth. However, if model efficiency improves rapidly, unit compute prices keep declining, or enterprise AI monetization underperforms expectations, then the long-term capacity locked in today could become fixed costs that are difficult to cut quickly in the future.

And this is precisely the biggest difference between the AI financing cycle and a purely technological cycle.

Final Thoughts

It's still difficult to simply interpret these changes as negative signals.

Quite the contrary.

The entry of bond markets, private credit, pension funds, insurance capital, and global asset managers is likely to further expand the pool of capital AI can command, allowing this infrastructure buildout to last longer.

As of August 12, US equities remain not far from their historical highs. Following the latest earnings reports, Amazon rose nearly 9% after hours on the strength of AWS's 37% growth; Microsoft was also handsomely rewarded by the market after demonstrating cloud growth and cash generation capability; Alphabet, meanwhile, came under pressure after announcing continued CapEx increases, while Meta faced selling after free cash flow plunged 91%.

The market hasn't begun rejecting AI investment—it's simply transitioning from a pure "demand trade" to a more demanding "return on capital trade":

• Phase one was about who dared to spend the most;
• Phase two is about who can secure more compute capacity at lower capital costs while generating sufficient cash flow from every dollar invested;

And when AI evolves from a tech company's capital expenditure into an asset that Wall Street can buy, finance, and price, the question this race ultimately needs to answer will inevitably become: "Whose buildout will actually end up making money?"

And that might be the true dividing line determining valuation gaps between Google, Amazon, Meta, Oracle, CoreWeave, and even Nvidia over the next year or two.