Key Takeaways
- Vik Uppal, who built real estate assets as a Minnesota teenager and now runs $2 billion credit shop Mavik Capital Management, has rejected every AI data center and power deal shown to his firm.
- Uppal draws a direct parallel to the late 1990s telecom boom, where traffic projections were accurate yet dark fiber assets still traded down to pennies on the dollar after capital structures broke.
- AI infrastructure sponsors are raising capital because liquidity is abundant, not because the long-term underlying asset values can be underwritten with safety.
- Credit investors who take on technology obsolescence risk accept an asymmetric setup where upside is capped at coupon payments while downside exposure equals total loss.
- Mavik is deploying capital into commercial real estate recapitalizations, where tangible assets and verifiable replacement costs offer clear downside protection.
Telecom Echoes in the Data Center Rush
Lenders and equity sponsors are pouring billions into specialized facilities and dedicated power generation to feed artificial intelligence clusters. Uppal refuses to participate. The frenzy looks identical to the late 1990s buildout of long-haul fiber optic networks. Back then, telecom sponsors raised vast sums on projections that internet data transmission would explode. The volume forecast was correct. The debt structures still collapsed.
“If you look at during the internet age, it was the same discussion with fiber,” (play from 38:04) Uppal observed. “And then what happened with fiber? All these companies that had all these wacky capital structures and then black fiber was sold for pennies on the dollar.” (play from 38:09)
When hardware efficiency improved and competitors overbuilt parallel routes, fiber capacity turned into a cheap commodity. The original debt holders were wiped out long before data demand peaked. Today, similar mechanics threaten AI infrastructure. Rapid changes in chip power draw, cooling architecture, and model training efficiency mean a specialized data center built today could become obsolete before a ten-year credit facility matures.
Raising Money Because It Exists
The volume of capital chasing AI power generation and compute real estate is driven by liquidity rather than clear recovery math. When an asset class gains momentum, institutional allocations follow the trend. Managers launch dedicated funds simply because market appetite guarantees quick fee generation.
“It's really interesting when you look at what's happening with AI and then all the AI infrastructure is you do see how manias and crazes start to get developed because you can't have a conversation that someone is not talking about that,” (play from 36:46) Uppal noted. “A lot of that is people raise capital, not necessarily when and where it makes sense, but because they can.” (play from 37:13)
Mavik manages flexible special situations credit. For a private credit lender, the math on AI compute facilities does not work. If the borrower succeeds, the lender collects a fixed coupon. If technological evolution renders the facility uncompetitive or tenants leave for cheaper capacity elsewhere, the lender takes the keys to a specialized shell with massive ongoing power liabilities. Without predictable residual asset value, credit underwriting turns into pure speculation. Rather than stretching for yield in fashionable infrastructure, Uppal directs Mavik toward distressed commercial real estate recapitalizations where baseline property values and replacement costs can be audited today.
Why It Matters
When credit funds decline high-yielding paper in the market's hottest sector, it signals an expanding disconnect between liquidity and asset-level security. The rush into AI compute mirrors past infrastructure cycles where technical obsolescence destroyed debt tranches despite undeniable top-line demand growth. Sophisticated capital is separating structural macro tailwinds from underwritable credit basis, leaving crowded momentum trades to managers who confuse capital availability with margin of safety.