The AI Borrowing Spree
In previous weeks, I reflected on the direction of hyperscaler issuance and whether the scale of AI investment could eventually lead to a crisis. However, continued reliance on leverage has revealed new ways in which investors are balancing the trade-off between participating in the AI buildout and associated credit risks.
What triggered this reassessment? In the last week alone, the big tech hyperscalers returned to the bond market with significant new offerings1. The deals reveal differentiated investor responses: steeper , wider and a growing divergence between the issuers’ official and the spreads investors are demanding.
Alphabet issued €6bn across seven , Microsoft entered a $25bn bond deal, and Meta priced $15bn, followed by a €8.25bn transaction. Oracle also raised $10bn in debt alongside ~$9.9bn in equity2.
The scale of demand remains extraordinary with Microsoft attracting a $115bn for its $25bn offering. This deal has attracted significant attention given that it was its first major bond offering since 2017. Meta attracted $58bn against its current ~$15bn dollar .
The dilemma for credit markets going forward is: how much debt big tech can take on to finance the AI buildout before the cost of staying ahead begins to undermine the credit quality that made them so attractive to begin with?
The Curve
The Price of AI
Arguably, one of the most interesting developments across the recent issuance is the shape of the curves. Short-dated bonds are priced relatively close to credit ratings, spreads widen progressively as maturities extend, suggesting that investors are demanding greater compensation for longer-term exposure.
Alphabet’s second euro deal of 2026 illustrates this clearly. Spreads over treausries range from 60bps at the two-year to 207.5bps at the 19-year point, a considerable increase from its May deal2. Microsoft’s curve, by comparison, is flatter, with 10 tranches ranging from 34.7bps to 132.5bps across maturities of 2-40 years. Despite substantial , its longer-dated tranches still steepen at the .

What stands out is how quickly spreads widen with and the pace differs by issuer. Meta’s spread premium rises about 100bps from 5 to 30 years, with the biggest jump between 5-to-7-year tranches (110bps to 145bps). Alphabet’s euro curve is steepest, though euro deals typically steeper curves than dollar ones. Oracle’s 5-year implies , a above its BBB- rating despite paying a 30-35bp concession2.
The broader implication is that markets are not simply pricing the creditworthiness of today’s , but the uncertainty of their cash flows too.
The Ratings
Are Investors Buying the AI Story?
indicate a divergence amongst hyperscalers, and the major players are treated heterogeneously. The rating puts these companies in a clear order, but the new-issue pricing doesn’t follow it3. Thus, investors are pricing each company’s ability to finance AI, and more crucially, generate returns from it.

Microsoft offers the strongest credit profile, supported by and a cloud business that can directly monetise AI infrastructure. Alphabet retains a strong balance sheet but faces scrutiny over its scale of capital commitments, as well as its ability to monetise its AI-powered search engine (without cannibalising its key advertising business.) For Meta, its 7-30-year tranches priced like BBB- despite an AA- rating. With no external cloud business, its AI return comes back only indirectly through advertising and engagement, thus explaining why it paid the steepest concession relative to its credit rating. Oracle, however, is priced at or inside BBB- rating, the ~$9.9bn of equity raised alongside reads as discipline to bondholders.
None of this is a downgrade. Wider long-dated spreads reflect , currency differences, and new issue concessions. More notably, they signal a changing credit story: as AI absorbs FCF, investors are looking beyond today’s balance sheets to the future trajectory of and the visibility of AI returns.
Each new-issue spread is matched to the closest average spread for US investment-grade bonds in each rating bucket. That average is not matched by maturity and the new-issue spread includes the concession, so the long-dated gaps overstate the pure credit concern. Alphabet's euro spreads, quoted over mid-swaps, are compared with a dollar ladder.
Fair Value
The Capex Backdrop: Scale and Duration
Investors view the recent spread widening as technical rather than fundamental4: a higher supply of bonds requires greater yields for supply absorption. Free cash flow is also a key credit variable to anchor thinking, and the figures are staggering. Gawrelski estimates across the Big-4 (Microsoft, Alphabet, Amazon and Meta), FCF is forecast to reach -$53 billion in 2027, as capex growth averages to 62% YoY, outpacing revenue5. Research by Flanagan expects negative FCF to mostly disappear by 20286, with easing in 2027. The disappearance of negative FCF as well as a slower-than-expected capex pace could allow spreads to tighten – potentially creating a buying opportunity for patient credit investors.
The counterargument is that leverage risks may be understated. In ‘The Gilded Age of Private Credit,’ leases, and can increase a hyperscaler’s leverage: “disclosure does not mean visibility.”7 Other underpriced risks include refinancing requirements, interconnected business models and the risk of obsolescence from single-purpose facilities.
The Outlook
The Return on AI
There’s a bit of a paradox at the heart of the borrowing spree: if hyperscalers continue to expand the buildout, they risk increasing leverage and weakening balance sheets, whereas slowing down could mean surrendering competitive advantage. For bondholders, neither outcome is particularly straightforward.
The real question is whether the returns on AI will justify the enormous amounts of capital being deployed today. If projected returns materialise and free cash flow recovers, then today's concessions could prove attractive. However, if monetisation disappoints, then the debt accumulated could leave a much longer-lasting mark on balance sheets.
For bondholders, the cost of Googling the future may be measured not in computing power, but in the price they are willing to pay to finance it.
- Financial Times, reporting on hyperscaler bond issuance (November 2026): https://www.ft.com/content/6c099dc6-1fb3-44fc-b011-70229e3755cb?syn-25a6b1a6=1
- Bloomberg News, new-issue coverage of the Alphabet, Microsoft, Meta and Oracle bond sales (3 to 5 November 2026)
- Bloomberg, average option-adjusted spreads for US investment-grade corporates by rating, used for the spread-implied ratings (November 2026)
- Daniel Sorid, The High-Grade Beat: Hyperscalers test new wides (research note, 27 October 2026)
- Ken Gawrelski, Supply Chain Inflation Takes Hyperscaler CapEx to $1.1 Trillion in '27 (research note, 28 July 2026)
- Christopher Flanagan, US Fixed Income Strategy: The US Fixed Income Weekly (research note, 2 November 2026)
- Nathalie Lustig, The Gilded Age of Private Credit, The Basis Point (7 July 2026): https://thebasispoint.finance/notes/the-gilded-age-of-private-credit

