AI’s biggest infrastructure builders (Alphabet, Amazon, Meta, Microsoft, and Oracle) are in the middle of one of the largest borrowing sprees in corporate history. They’re raising enormous sums to fund data centre buildouts, and they’re doing it in the public bond markets.

The obvious question is: can the AI-related borrowing binge actually break the bond market?

So far, the answer looks like “no,” but the early deals are starting to show where the pressure points might be.

The first wave of AI mega-deals

In 2025, AI hyperscalers dominated U.S. investment grade bond issuance.

• Oracle sold US$18 billion in bonds in September.
• Meta followed in October with a record US$30 billion deal—the largest investment-grade transaction ever that wasn’t tied to an acquisition.
• Alphabet and Amazon joined in November with US$17.5 billion and US$15 billion deals, respectively.

Activity has accelerated in 2026.

Oracle (rated Baa2/BBB) returned in February with US$25 billion of senior unsecured bonds across eight tranches, including a floating-rate note geared to bank portfolios. Management has framed a broader plan to raise roughly US$45–50 billion via debt, equity-linked, and common equity to fund AI data centres.

Alphabet (Aa2/AA+) chose a different route, coordinating a multi-currency financing program:

• US$20 billion across seven U.S. dollar tranches on Monday, February 9th
• £5.5 billion across five sterling tranches the next day, including a £1.0 billion 100 year bond, and
• CHF 3.1 billion in Swiss-franc bonds launched the same day.

What these deals tell us about market capacity…so far

Despite the size, the deals cleared easily. That’s the first important signal concerning capacity. 

• Order books were deep. Oracle reportedly drew over US$127 billion of demand for its US$25 billion deal. Alphabet’s U.S. dollar tranches attracted more than US$100 billion of orders. The 100 year sterling tranche saw demand close to 10 times its £1.0 billion size.
• Pricing looked like a concession, not distress. Even at US$25 billion, Oracle borrowed out to 30 and 40 years at about +180 bps and +195 bps over Treasuries (1.80–1.95 percentage points). That’s wider than you’d see for a lighter capex profile, but it doesn’t look like a failed distribution.
• Structure broadened the buyer base. Oracle’s floating-rate note created a natural home in bank portfolios and among floating-rate buyers, widening the pool of eligible investors.
• Alphabet executed a “textbook” capacity arbitrage. By tapping three currencies (USD, GBP, CHF), setting records in the sterling market, and placing ultra-long duration (including a 100 year bond), Alphabet illustrated how an issuer can:
  • avoid saturating any single buyer segment, and
  • place very long-dated bonds with investors that actually need them (U.K. pension funds and insurers with long-term liabilities).

In spread terms, the market response looked healthy.

•  Alphabet’s U.S. dollar bonds priced between roughly +27 bps and +95 bps over Treasuries.
• Lower-rated Oracle needed wider spreads—about +95 bps to +195 bps on fixed-rate tranches—but was still able to place 30  and 40 year paper.

Recent century and ultra-long bonds from these names (Alphabet’s 2126 sterling tranche and 2075 U.S. dollar tranche; Oracle’s 2066 tranche; Amazon’s 2065 tranche) underline that there is still demand for perceived high-quality tech issuers at the long end of the curve.

In other words, the technical plumbing has held up so far. Deals cleared, books were oversubscribed, and spreads were consistent with strong, if price-sensitive, demand.

The looming wave: volume, cadence, and where things could crack

If these transactions were the entire story, the recent issuance would already be impressive. In reality, they’re just a preview.

• Reuters reports that dealers expect U.S. corporate bond issuance to approach record levels in 2026, with Barclays projecting US$2.46 trillion in total supply and US$945 billion in net issuance.
• The five most active hyperscalers issued US$121 billion of bonds in 2025—more than four times their US$28 billion annual average between 2020 and 2024.

So far in 2026, Oracle and Alphabet have financed only part of their stated capital expenditure:

• Alphabet has debt-financed roughly 15% of its US$180 billion 2026 capex plan.
• Oracle has financed about 50% of its US$50 billion plan.

If Amazon, Meta, and Microsoft debt-finance even 30% of their 2026 capex guidance (Amazon: US$200B, Meta: US$125B, Microsoft: US$105B) and if Alphabet or Oracle return to the market, incremental issuance could exceed US$150 billion this year from this cohort alone—a roughly 24% year-over-year increase, with more to come if capex keeps growing.

The concern isn’t that the global bond market is too small—it exceeds US$100 trillion across government and corporate bonds in dozens of currencies. The concern is how concentrated and how fast the supply arrives, and in what form.

If multiple US$30–60+ billion hyperscaler deals hit within weeks, especially at ultra-long maturities (40–100 years), the marginal clearing price could start to move and cause:

• wider new-issue concessions,
• heavier use of CDS hedges, and
• more day-to-day credit spread volatility, potentially spilling over into the broader market.

Both Reuters and the Financial Times have already noted that some investors are questioning the payoff timelines for AI capex, and they’ve linked modest spread widening and increased hedging activity directly to AI-related bond supply.

It’s not just “how much” they issue. It’s what and where

Three structural factors will shape how much capacity remains at current spread levels:

1. Credit quality still matters

• Alphabet placed 32 year and 100 year sterling bonds at about +60 bps and +120 bps over gilts, tight for that kind of duration.
• Lower-rated issuers see steeper curves. Alphabet’s 30 year U.S. dollar tranche only needed about 12 bps more spread than its 10 year. Oracle’s 30 year needed an additional 25 bps over its 10 year.

Put simply: ultra-long capacity exists, but it’s more price-sensitive for lower-rated credits.

2. Currency and market segmentation

Alphabet’s choice to sequence issuance across USD, GBP, and CHF is itself a capacity management strategy:

• it reduces dependence on any single buyer base, and
• it can lower all-in cost if one currency market is relatively “cheaper” at the time.

A similar multi-market, staggered approach from Amazon, Meta, or Microsoft could help the market absorb additional supply without major dislocation.

3. Dealer balance sheets and liquidity conditions

Even when end-investor demand is there, dealer syndicates have to warehouse risk and support secondary trading. Alphabet’s U.S. dollar deal gives a sense of the scale with three global coordinators, 17 joint lead managers, and 17 co-managers with 20 underwriters participating for the first time.

That kind of syndicate is a signal: the market can handle large deals, but doing so consumes balance sheet and human capital. If several similar-sized deals arrive in close succession, that intermediation capacity—not the abstract size of “the bond market”—will be tested.

So, how close are we to breaking the bond market?

At today’s spread levels and with the structures we’ve seen so far, the market is absorbing AI-related supply surprisingly well. The ICE fixed income index tracks more than US$100 trillion of government and corporate bonds, so even an extra US$150+ billion of hyperscaler issuance in a year is not, in itself, a showstopper.

The stress points are more subtle:

• Cadence and clustering: too many large, similar deals at once
• Duration: pushing more 40–100 year risk into a narrower investor segment
• Intermediation: how much risk dealers are willing and able to hold while matching sellers and buyers

We don’t yet know exactly where the marginal buyer will start to demand meaningfully wider spreads, but the Oracle and Alphabet transactions suggest that we have not hit that point yet.

In the coming months, each new hyperscaler financing will be another live experiment in how far investors are willing to extend in size and maturity for the AI buildout story.

For now, the answer to “Hey Google, how much can I borrow before I break the bond market?” is: more than you’ve borrowed so far…but the market is quietly measuring your limits.

Post Script:

Since this blog was originally written in mid February, Amazon also returned to the bond market with additional financings. On March 10, the company announced their intention to raise US$25 to US$30 billion in the U.S. market and an additional €10 billion in Europe. By the time the dust settled, Amazon raised over US$53 billion in aggregate, printing US$37 billion across eleven tranches in the U.S. and €14.5 billion across eight tranches in Europe. The U.S. dollar financing included multi-billion 30-year, 40-year and 50-year tranches, and the Euro denominated bond issue set a new record for amount raised by a corporate issuer in that market.