Hook
$3 billion. That’s the price tag for a data center network most AI users have never heard of. TPG is in advanced negotiations to acquire Netrality Data Centers, a portfolio of carrier hotels scattered across second-tier US cities like St. Louis, Kansas City, and Philadelphia. The stated goal: "accelerate AI infrastructure growth." But if you strip away the AI narrative, what you’re left with is a real estate acquisition levered against the assumption that AI demand will never cool. I’ve seen this movie before — in 2017, when every ICO whitepaper promised "decentralized compute" and bought server racks at 10x multiples. Code doesn’t care about your feelings, but PE firms care about spreadsheets. Let me show you where the real risk sits.
Context
TPG Capital, a private equity giant with a growing digital infrastructure portfolio (DataBank, Cirion), is circling Netrality Data Centers. The transaction values Netrality at roughly $3 billion, including assumed debt. Netrality’s assets are not the hyperscale campuses in Northern Virginia; they are multi-tenant carrier hotels with deep fiber connectivity, mostly in Midwestern markets. Think of them as "last-mile" interconnection nodes rather than pure AI heavyweights. The pitch is simple: as AI workloads explode, even secondary locations will fill up with cost-sensitive training and batch inference tasks. The deal fits a pattern — Blackstone bought QTS for $10B, KKR took CyrusOne private for $15B. But this particular price, around $8–10 million per MW, sits above the industry median, demanding a premium that only sustained AI growth can justify.
Core
Let’s dissect the financial engineering, because that’s where the true alpha signal lives. A $3 billion enterprise value, assuming a 60% loan-to-value ratio, means TPG puts up only $1.2–$1.5 billion in equity. Their target IRR? Likely 15–20%, driven by a combination of rent escalation, EBITDA expansion, and eventual exit via an REIT IPO or sale to a yield-chasing infrastructure fund. This is a classic "yield capture" play — leveraging cheap debt to buy an asset that produces predictable cash flows. But here’s the catch: those cash flows are contractually fixed in nominal terms, while electricity costs, which make up 50–60% of a data center’s operating expenses, are floating. In a bull market for energy — which we are in — the landlord’s margin gets squeezed. You can pass costs to tenants, but only if vacancy stays low. If AI demand falters, tenants have negotiating power.
I’ve audited enough DeFi lending protocols to recognize this as overcollateralized lending: TPG is lending against future AI compute demand, with the collateral being steel and concrete. The liquidation risk is a sudden drop in lease rates. In 2022, when Ethereum switched to Proof-of-Stake, dedicated mining farms saw their asset values halve overnight. The same mechanism applies here: if an AI model release fails to deliver or regulation caps compute, the "yield" on this real estate evaporates. Yield is the bait, rug is the hook.
Contrarian
The consensus narrative is that data centers are a scarce, irreplaceable bet on AI’s secular rise. I disagree on three grounds. First, scalability is a double-edged sword. Netrality’s assets are in secondary markets specifically because hyperscale providers (AWS, Azure, GCP) have already saturated prime locations. That means Netrality’s competitive advantage isn’t technology — it’s geographic arbitrage on power prices. But power price arbitrage is not a durable moat; utilities can raise rates, and next-generation small modular reactors could make rural sites equally competitive. Second, the supply side is catching up. Over the next 24 months, thousands of MW of new capacity are coming online in markets like Phoenix, Las Vegas, and Ohio. If supply eases, lease rates will compress. Third, the liquidity of these assets is poor. If TPG needs to exit quickly during a downturn, they’ll sell at a discount. Panic sells, liquidity buys — and right now, TPG is the liquidity buyer, mopping up an asset class that might look less attractive when AI T-shirts are no longer being printed.
Takeaway
Watch the power purchase agreements, not the press releases. If TPG locked in 10-year electricity hedges at sub-4 cents per kWh, this deal has a strong floor. If they didn’t, the spread between rent and power could shrink faster than a token liquidity pool. The real question isn’t whether AI will grow — it’s whether the infrastructure market is pricing in a permanent growth rate that ignores mean reversion. In DeFi, we rebalance positions based on on-chain data, not marketing. For data center investors, the on-chain data is the PJM wholesale electricity price and the vacancy index of the Top 10 data center markets. Ignore those at your own IRR.