Key Takeaways

  • Hyperscale campuses require 100 megawatts or more of capacity, with build costs running from $10 million to $15 million per megawatt in the United States.
  • Regional build costs vary across Asia-Pacific: India and Southeast Asia average $7 million per megawatt, Australia reaches $15 million, and Japan peaks at $20 million per megawatt.
  • Hyperscale underwriting targets a 10%-plus yield on cost anchored by 15-year contracts with investment-grade tenants.
  • Urban colocation assets demand mid-teens yields on cost to offset operational complexity, multi-tenant churn, and shorter 4- to 5-year lease terms.

The Hyperscale Model: Scale and Long Leases

Building large-scale computing capacity requires heavy upfront capital and strict site selection. Andrew Thomas outlines the baseline for hyperscale developments: “When you think about the typical hyperscale investment, you're spending somewhere between $10 to $15 million per megawatt in building out a facility. You're generally looking to build a campus that is going to be at least of 100 megawatts size or greater.”

That scale alters the risk profile. A single 100-megawatt campus quickly becomes a billion-dollar commitment before servers arrive. To justify that outlay, sponsors de-risk cash flows through long-term off-take agreements. Thomas notes that sponsors want predictable revenue streams: “And from a return standpoint, we're generally looking to sign 15-year contracts with investment grade counterparties where we're getting a yield on cost that is as close to 10% or above as possible.” These long commitments from high-credit tech firms turn hyperscale assets into credit-like bond proxies.

Regional Cost Divergence and Exit Spreads

Geography dictates construction math. Dev Gupta tracks how local conditions reshape project costs across the Asia-Pacific region: “Build costs in places like India, Southeast Asia are typically in the $7 million per megawatt range. They go up to about 14, 15 in Australia, and they go up to about 17 to $20 million in Japan. Japan becoming the highest cost market.”

High costs in markets like Tokyo do not break returns if rental rates and exit pricing move in tandem. Underwriters anchor their risk tolerance on development spreads rather than raw build figures. As Gupta frames it: “What is my stabilized yield on cost versus what is the cap rate at which I can sell this asset, monetize this asset? And if that spread is sufficient to be able to take that risk, then take that risk.” Strong tenant demand and large lease commitments have protected this spread between development yields and institutional exit cap rates.

Why Colocation Demands Mid-Teens Yields

Urban colocation operates on a different financial model. Instead of a single anchor tenant taking 100 megawatts for fifteen years, a colocation facility leases smaller capacity blocks to dozens of enterprise customers on 4- to 5-year terms.

Shorter leases bring frequent re-leasing cycles, customer churn, and ongoing sales expenses. Managing multiple customer cages also adds direct operating overhead. To offset those frictions, investors demand higher pricing power per kilowatt.

“Tenant quality is critically important, but you have diversity of tenant as well because of the incremental operating risk or operational complexity associated with having multiple customers and because you're able to charge higher price points,” Thomas explains. “We are typically aiming for more of a mid-teens yield on cost on the capital that we put in the ground.” That 300 to 500 basis point spread above hyperscale yields compensates sponsors for the hands-on management colocation demands.

Why It Matters

This divide shows that private equity cannot price data centers as a single asset class. As computing needs expand, capital is splitting between long-dated utility plays and actively managed multi-tenant real estate. Platforms that secure power access and manage construction costs will capture wide monetization spreads as core infrastructure funds compete for stabilized exits.