BongkarnThanyakij / Getty Images
Meta (META) issued $30B in financing for a Louisiana data center. Meta carries $37B in total debt against over $60B in cash reserves.
Oracle holds nearly $96B in debt after an $18B bond issuance and $38B loan. Oracle’s interest payments could consume a larger share of its $3B quarterly net income.
Tech companies borrowed $75B in September and October for AI data centers. This is more than double the annual average over the past decade.
Some investors get rich while others struggle because they never learned there are two completely different strategies to building wealth. Don’t make the same mistake, learn about both here.
Bank of America just issued research that points to a potentially troubling shift in how major tech companies fund their artificial intelligence (AI) ambitions. It noted borrowing for AI data centers exploded higher in September and October, with firms like Meta Platforms (NASDAQ:META), Oracle (NYSE:ORCL), and others issuing $75 billion in bonds and loans — more double the annual average over the past decade.
This comes as capital expenditures approach limits on what company cash flows alone can support. Consensus estimates show AI capex hitting 94% of operating cash flow minus dividends and share repurchases in 2025 and 2026, up from 76% in 2024. This trend raises questions about sustainable growth amid high valuations, as companies may increasingly rely on debt to build the infrastructure AI models and computing power require.
So far, the tech giants have fueled their AI expansion through strong cash flows from core businesses like cloud services and advertising. However, the scale of data center buildouts now demands more capital than internal resources can provide.
Bank of America’s analysis highlights this pivot, noting a surge in borrowing in recent months. For instance, aggregate values for key players including Amazon (NASDAQ:AMZN), Alphabet (NASDAQ:GOOG)(NASDAQ:GOOGL), Microsoft (NASDAQ:MSFT), and others show capex shares rising sharply. This suggests the self-funded model that drove rapid AI progress may be causing strain, forcing companies to tap debt markets to maintain momentum.
Projections indicate global data center spending could reach $3 trillion by 2028, with a significant portion financed externally. While cash flows remain robust, totaling hundreds of billions of dollars annually across the sector, the pace of investment is outstripping them.
This shift could support continued AI innovation but introduces leverage that wasn’t as prominent in earlier phases.
Story Continues
Many of these companies already carry substantial debt, accumulated from past acquisitions and expansions. Meta, for example, has around $37 billion in total debt, offset by over $60 billion in cash reserves. Its recent $30 billion financing deal, including $27 billion in debt for a Louisiana data center, adds to this but remains manageable given quarterly operating cash flows exceeding $20 billion and low interest expenses under $200 million per quarter. Investors, though, are still worried Meta is plumbing the depths of its metaverse excesses again.
In contrast, Oracle’s debt stands at nearly $96 billion following its $18 billion bond issuance and a recent $38 billion loan in late 2025. With quarterly net income around $3 billion, its interest payments could climb to consume a larger slice of profits if rates rise. Essentially, it is spending money it doesn’t have on facilities that haven’t been built for customers it doesn’t have.
Smaller AI players like Advanced Micro Devices (NASDAQ:AMD) face steeper challenges. With debt under $3 billion currently but potentially having the need for billions more in capex, they might borrow at higher rates due to lower credit ratings, pushing interest costs to eat into slim margins.
Nvidia (NASDAQ:NVDA), however, exemplifies resilience with minimal debt — less than $10 billion — and massive free cash flows from chip sales, allowing it to fund growth internally longer than peers.
Relying on debt could accelerate AI advancements by enabling faster data center deployments, but it heightens risks amid lofty stock valuations. If AI returns disappoint — say, due to slower adoption or technological hurdles — interest burdens could pressure earnings. High valuations, with some firms trading at 40 times earnings or more, assume flawless execution; any debt-fueled slowdown might trigger corrections.
Yet, diversified revenue streams provide a buffer. Microsoft’s enterprise focus and Amazon’s e-commerce base generate steady cash, helping to mitigate risks. Still, systemic issues like a potential debt bubble loom if economic conditions worsen, echoing past tech exuberance.
To that point, The New York Times just revealed that Blackstone (NYSE:BX) is using “exotic financial instruments” to finance the expansion of its digital infrastructure company, QTS Data Centers. Blackstone is about to close on a $3.46 billion commercial mortgage-backed securities (CMBS) offering to refinance debt held by QTS, which it identifies as the biggest player in AI infrastructure. The Times highlights McKinsey data showing $7 trillion in data center investment will be required by 2030 to keep up with projected demand.
The companies — such as Google, Meta, Microsoft, and Amazon — which have spent a combined $112 billion in just the last three months, are using a mix of corporate debt, securitization markets, private financing, and off-balance-sheet vehicles that harken back to the 2008 financial crisis.
For investors, this signals the need for extreme caution before buying AI-driven stocks. Strong balance sheets like Meta’s suggest debt is a tool for growth, not distress, but investors need to watch interest coverage ratios and capex ROI on the stocks they want to buy. Weaker players risk overleverage, potentially eroding returns.
The AI boom will still fuel further growth, but today’s winners won’t necessarily be the ones grabbing the brass ring tomorrow. Sustained AI hype depends greatly on profitable scaling without taking on excessive borrowing costs. Not everyone will thread that needle.