Global Market Strategist
Credit Research Analyst
The AI capital expenditure cycle has migrated from balance sheet sourced to capital markets, as hyperscaler1 free cash flow is no longer sufficient to fund the buildout. Incorporating data center developers and semiconductor investment, the full data center buildout could cost roughly $5 trillion through 2030, with approximately $2 trillion of that financed in investment grade (IG) credit markets.
Sizing up issuance
Hyperscaler corporate bond issuance has surged in recent years. In 2024, issuance totaled just $17bn as the first years of investment came from operating cash flow. In 2025, issuance catapulted to $109 billion, and this year’s issuance has already reached $194 billion in just the first six months. We estimate 2026 issuance will finish at $279 billion, with another $220–$300 billion likely in 2027. As a result, hyperscalers could approach 10% of the U.S. IG market by 2030, becoming a sizeable share of the credit universe.
It’s more than just the hyperscalers
From a macro lens, all AI-linked issuance pushes these supply figures much higher. A growing share of the data center financing is coming via 144A-for-life and private placement vehicles that are excluded from traditional Bloomberg U.S. Corporate indices2, so the visible index-eligible figures understate the true scale of AI-related leverage accumulating in portfolios.
Moreover, a growing share of data centers currently in construction are financed via bank construction loans or in high yield; as these projects reach completion and begin generating cash flows, many will likely migrate to or be refinanced in the IG market—adding still more supply.
Credit market implications
The pickup in issuance is starting to be reflected in credit spreads. Technology spreads now trade wider than the broad IG benchmark—a reversal from the sector’s long-held premium status; and demand is softening at the margin as investors digest wave after wave of new paper.
To be clear, we do not believe this issuance causes a credit crisis; hyperscaler balance sheets remain among the strongest in the index. But it could cause some indigestion as markets work to absorb the supply smoothly. Spreads have closely tracked trends in hyperscaler free cash flow, and given profitability has come under pressure, the pressure on spreads is no surprise.
Given the multi-year nature of the buildout, hyperscalers are also issuing substantial long-dated paper. That is a sensible strategy in pension-heavy systems like Europe, where liability-matching demand for duration runs deep, but domestic demand for long corporate paper is narrower—reflected in steeper OAS curves for hyperscaler debt.
Foreign demand remains robust as issuance has picked up in other currencies. In February, Alphabet issued a rare £1 billion 100-year bond in the sterling market, the first century bond from a major technology company since Motorola in 1997—which drew nearly 10x oversubscription at roughly 120 basis points over 10-year gilts, anchored by pension funds and insurers.
In the U.S., hyperscaler oversubscription has averaged a healthy 3.5–4.5x in the first half of the year. That said, the first surprise hyperscaler deal in the second half saw just 1.4x oversubscription levels, reflecting a lower appetite as the heavy issuance starts to take its toll. Tapping global markets for incremental demand is smart treasury management; it is also a tacit acknowledgment that dollar IG demand alone cannot comfortably absorb the supply ahead.
Investment implications
Issuance of this scale concentrated in a handful of names, with added complexity from 144A digital infrastructure deals and widening dispersion across sectors, maturities and currencies, strengthens the case for active management. Investors should expect AI-related supply to keep a floor under tech spreads while recognizing that headline figures may understate total exposure. Ultimately, security selection and curve positioning—not passive index replication—will determine who navigates this cycle well.