Big Tech is traditionally a low-leverage sector. The dominant companies (Amazon, Meta, Alphabet, Microsoft, etc.) make heaps of money, have robust operating cash flows, and haven’t needed to take on debt.
Way back in 2024, hyperscalers funded data center construction largely with their own money, keeping corporate bond issuance at a modest $16.7 billion, according to LSEG. Since then, things have changed. By August 10th of this year, the same group had issued roughly $220 billion.
Goldman Sachs puts global AI-related investment at $1 trillion in 2026, a figure that includes private companies and firms outside the U.S. Morgan Stanley separately expects around $570 billion of AI-related debt issuance this year. Goldman’s own estimate is that debt will cover 33% of AI capital expenditure in 2026, rising to 35% in 2027.
In addition to private credit, the AI space has turned to giant multi-tranche public bond sales. Currently, all Big Tech companies maintain investment-grade (IG) credit ratings, meaning they are given high marks by major credit rating agencies like Moody’s and S&P Global Ratings. These ratings, in turn, grant Big Tech access to capital markets on the cheap.
The range within that band is wide. S&P rates Microsoft AAA and Oracle BBB with a negative outlook, which puts Oracle two notches from losing investment grade altogether.
A Sector That Never Needed the Bond Market
In the words of Amanda Lynam, Chief Credit Strategist at Goldman Sachs:
“The tech sector broadly has been very highly rated and very under-levered for a period of time. And so there’s ample runway to add debt to the capital structures for a strategic need.”
However, for a sector that traditionally hasn’t taken on debt because it hasn’t needed to, it’s a notable change and one that’s affecting the broader economy, especially the U.S. Treasury bond market. Nomura estimates that roughly $200 billion of borrowing by the largest tech companies is equivalent to about 25% of the Treasury’s net issuance of notes and bonds to private investors, five times its share in 2025.
What are the underlying conditions of the funding strategy? And what are the risks?
Anything But Selling Stock
The rapid expansion of data centers around the world has been accompanied by skyrocketing costs. These costs require hundreds of billions in upfront cash, amounts that are increasingly outpacing Big Tech’s immense earnings.
In lieu of issuing stock to cover costs, which would dilute shareholder value and lessen the voting power of founders and major investors, hyperscalers are relying on the bond market.
Interest paid for corporate bonds is often tax-deductible, which lowers the overall cost. Additionally, investor demand for corporate bonds in the AI space has been high, which has kept spreads compressed even as underlying rates have climbed.
However, there are indications that investor demand is softening. S&P Global analysts describe hyperscaler spreads widening with each new debt infusion, including for Microsoft, which has not recently issued. Lynam of Goldman Sachs reports that many clients are already exposed to the tech sector via equities, which could throttle demand for hyperscaler or tech debt via the bond market.
Institutional investors, including foreign investors, pension funds, and insurance companies, are the buyers absorbing the corporate bonds funding the AI infrastructure boom. Managers of these funds may be reaching their limits in regard to tech exposure.
Keeping It Off the Books
Hyperscalers have historically owned their largest self-built campuses outright. However, new data centers are increasingly being funded by special purpose vehicles (SPV), shell companies specifically set up to finance and own a site-specific data center.
Within this arrangement, the hyperscaler is typically a 20% owner, while the remaining 80% is made up of private credit or a pool of bond investors. The hyperscaler acts as a long-term tenant, and the bondholder is paid out of the lease payments.
Meta’s Hyperion campus in Louisiana is the template. Blue Owl-managed funds own 80% of the joint venture, and Meta owns 20%. Morgan Stanley arranged roughly $27 billion of SPV debt, anchored by Pimco, and Meta leases the finished campus back. None of that debt appears as a Meta liability. What does appear, in a footnote, is a residual value guarantee obligating Meta to compensate the joint venture if the campus is worth less than a set threshold when a lease ends.
SPVs enable Big Tech to keep billions of dollars in construction debt off its main corporate balance sheets. In turn, Big Tech firms are able to maintain investment-grade credit ratings.
Shadow Borrowing and the Footnote Problem
It’s a curious arrangement. Big Tech’s strong credit rating underpins the stability of the SPV arrangement, which ultimately convinces bondholders and private credit to invest. However, the SPV arrangement in itself actually creates structural risk that is hidden in the footnotes of Big Tech’s corporate accounting.
Critics refer to the strategy as shadow borrowing and point to ballooning obligations far outpacing revenue. A Wall Street Journal analysis of filing footnotes at nine tech companies found roughly $3 trillion in off-balance-sheet commitments tied to AI. About $1.2 trillion of that is leases that have not started, and $1.9 trillion is purchase commitments. The total runs to roughly five times the $600 billion of capital expenditure the same companies reported over the past year.
The IPO Is the Warm-Up Act
Anthropic filed a confidential S-1 on June 1st and is targeting an October listing on the Nasdaq, following a $65 billion Series H that valued the company at $965 billion. OpenAI filed a week later but said it had not decided on timing, and most forecasts now put its listing in 2027. As evidenced by SpaceX, neither company will stop at raising money through the IPO process.
SpaceX listed on the Nasdaq on June 12th under the ticker SPCX, raising about $75 billion. It had already told investors it had investment-grade ratings lined up from three agencies before the offering priced, despite posting a net loss in 2025. Weeks later the company priced a $25 billion bond in five tranches, moving past its IPO windfall to raise more money for its CAPEX needs.
Anthropic and OpenAI can be expected to do the same, raising billions of dollars directly, as opposed to relying on circular financing deals with their partners. Currently, neither carries a public credit rating from the major agencies.
Due to this reality, they rely heavily on their hyperscaler and chip partners (Amazon, Nvidia, etc.) who act as guarantors, landlords, and at times customers. But with the massive influx of fresh capital via a successful IPO, both Anthropic and OpenAI could be deemed investment grade, which would further open the floodgates for borrowing.
Nobody Has Tested Which Way It Breaks
The borrowing itself is unremarkable. Investment-grade companies issue bonds; that is what the market exists for, and the hyperscalers have more headroom on their balance sheets than almost any issuer around. JPMorgan Asset Management figures the group could add $1.5 trillion more and still be comfortable.
The harder question is what happens to the part that never lands on a balance sheet. Meta moved $27 billion of Hyperion construction debt into a joint venture it owns a fifth of, leased the campus back, and guaranteed the residual value in a footnote. The rating agencies see a lease. The bondholders see Meta. Both are looking at the same building.
That is the trade the whole structure rests on. Big Tech’s credit rating makes SPV paper sellable, and the SPVs keep the credit rating clean. Each side props up the other, which is another way of saying nobody has tested which one gives first.
The market has started charging for the uncertainty. Oracle’s five-year credit default swaps have gone from about 40 basis points a year ago to north of 200, and Société Générale puts implied default odds for the hyperscaler group near 7% against roughly 4.5% for investment grade broadly. Reuters notes the tech CDS market is thin enough that those spreads may not mean much yet. They have also only moved in one direction, and the first quarter that AI revenue misses will be the one that tells us what they were pricing.
Author: Tim Tolka, Senior Reporter
#Crypto #Blockchain #DigitalAssets #DeFi
The editorial team at #DisruptionBanking has taken all precautions to ensure that no persons or organizations have been adversely affected or offered any sort of financial advice in this article. This article is most definitely not financial advice.
See Also:
America’s AI Bubble Has the Same Cracks That Just Broke South Korea | Disruption Banking
Are Hedge Funds Abandoning the AI Trade After Four Weeks of Selling? | Disruption Banking
