Goldman Sachs AI datacenter spreads widen to 353bps, raising borrowing costs

By Khal · · 2 min read

Editorial illustration: A steep, narrow bridge climbs across a rocky chasm to a small building containing server racks. A broad, level bridge leads to a much larger datacenter in the background.
AI-generated editorial illustration · AI-generated editorial illustration · Published Sep 14, 2026, 1:04 a.m. (3 hours ago)

In brief

  • Goldman's HY AI datacenter basket trades at 353bps, up from 319bps at July 2026 launch
  • AI-related bond spreads widened as demand for longer-dated tranches cools
  • Independent operators face meaningfully higher financing costs than hyperscaler giants
  • $500 billion in AI debt expected in 2026, roughly 18% of total US investment-grade supply

Spreads Widen Across the AI Debt Landscape

Goldman launched its high-yield AI issuer basket in July 2026, comprising 18 equal-weighted US high-yield names. At inception, the basket carried an average spread of 319 basis points and a yield of 7.45%. The move to 353 basis points represents a 34 basis point jump, signaling deteriorating conditions for the companies that populate it.

The strain extends beyond this single basket. Broader AI-related bond issuance expected across 2025 and 2026 carries an option-adjusted spread of 381 basis points. Goldman's AI leadership basket, tracking investment-grade AI-adjacent credits, has seen its spread widen from 74 basis points to nearly 150 basis points over the past 12 months.

The Real Cost: Higher Financing Burden

For companies issuing this debt, the math is straightforward. Wider spreads translate directly into higher borrowing costs. A datacenter operator that could have financed a billion-dollar facility at 7.45% a few months ago is now facing meaningfully steeper rates.

Of the 23 recent datacenter joint-deal ventures Goldman tracks, 17 are now trading wider than their origination yields. New-issue concessions on large deals have widened by up to 20 basis points, as demand for longer-dated tranches has cooled.

At the scale of AI infrastructure buildout, these moves compound. Even 30-40 basis points of additional spread across hundreds of billions in issuance represents billions of dollars in incremental interest expense over the life of these bonds.

A Two-Tier Financing Market

The real divide is between the giants and everyone else. Hyperscalers like Microsoft, Google, and Amazon can finance datacenter construction off their investment-grade balance sheets at far tighter spreads. Independent operators and joint ventures don't have that luxury.

The basket includes names like CoreWeave alongside various joint venture debt structures. These mid-tier players are bearing the brunt of the widening spreads.

Goldman estimates that close to $500 billion of AI-related debt will be issued in 2026 alone, representing approximately 18% of total US investment-grade supply. If spreads remain elevated, that's a material headwind for the independent operators racing to build out capacity.

Related stories