A Wellington Management fixed-income manager was talking about AI borrowing in the investment-grade bond market. The big five hyperscalers have grown from 1–2% of the index to about 5% in two or three years, and they issue unusually long debt, in some cases 100-year bonds.
…funding needs, they tend to issue much shorter debt. Hyperscalers prefer much longer debt because it fits the business model much better. And so as you think about what we call the duration of the index, it's extending meaningfully because they issue 20, 30, 40, 100-year debt in some instances.
Moran says we've seen the risks of concentration before.
Hearkening back to, you know, when we think about post-GFC and thinking about money center bank concentration, which typically has been the largest source of concentration risk in the market that we've dealt with, the AI hyperscaler trade will easily overtake the U.S. money center banks in terms of concentration over the course of the next year and a half or two years.
Hyperscalers are flooding debt markets with more than just traditional corporate bonds. Special purpose vehicles allow them to issue bonds specifically for data centers, keeping the debt off their balance sheets.
The investment grade markets, as I mentioned, are actually being asked to bear a lot of different types of risk than we do traditionally. So the hyperscalers may actually get to a point where they're very attractive incremental bond market opportunities, whereas data centers are a little bit…
I think where you're seeing more differentiation is in some of the off-balance sheet structures, whether it's the GPU financings, the chip financings, the actual data centers themselves. You've seen in recent days there's been force majeure, at least implied in a couple instances. And so you have…