Miner Weekly: $33B in Bonds Is Fueling the AI Megawatt Arms Race

If you want to understand the AI and data center boom in 2026, don’t just look at GPUs, megawatts, or backlog announcements.

Look at the bond market.

And the spread between who pays 4% and who pays 9% tells you almost everything about how the market is underwriting the data center arms race.

At the high-yield end of the spectrum, the capital isn’t cheap.

CoreWeave printed:

Applied Digital: $2.35B at 9.25% (Nov 2025)

TeraWulf: $3.2B at 7.75% (Oct 2025)

Cipher’s February deal is interesting. In just three months, its pricing improved by a full percentage point even as it doubled down with a $2 billion issuance. That suggests there’s still appetite for “compute-backed” credit, especially when colocation leases and power contracts are in place.

But zoom out and compare this to regulated utilities and power producers.

Now look at the other side of the ledger.

Vistra Corp.: $2.25B at 4.70% and 5.35%

Same macro environment. Same Treasury curve. Different credit pricing.

If you line these issuers up by coupon, you get a rough risk ladder:

Regulated utilities and established power producers tend to sit in the investment-grade universe, with long operating histories, predictable (often regulated) cash flows, and deep institutional demand for their paper.

On the other side, the newer “compute” names — especially the ones still scaling, still building, or still proving the durability of their customer base — are typically borrowing as high-yield / speculative-grade credits. Even when they have real contracts, the market still prices in execution risk, refinancing risk, and the reality that capex eats cash before it creates cash.

Utilities are openly revising capital plans upward. Southern now expects a $78.1 billion investment plan through 2030, with $15.9 billion in 2026 alone — explicitly citing projected load growth from data centers. Dominion similarly flagged billions in anticipated long-term debt issuance (between $6B and $9.5B in 2026) to support infrastructure expansion driven by large new data center customers.

On the AI side, the logic is simpler: secure power first, figure out monetization later.

That’s a bigger question hanging over all of this.

If AI demand holds, these coupons may look entirely rational. Debt gets refinanced lower. Assets appreciate. Power scarcity becomes the bottleneck.

Remember: most of these maturities cluster around 2030–2036. That’s not far away in infrastructure time. This isn’t just a power story anymore. It’s a balance sheet story.