The Self-Build Surge Will Not Kill Colocation. It Will Split It.
The self-build vs. colocation debate frames the wrong question. The structural issue is not whether the colocation market survives alongside hyperscaler-owned campuses it does. The issue is which segment of the colocation market compounds at premium economics over the next decade, and which segment begins to compress.
For infrastructure investors, these are not equivalent positions.
The mechanism
Microsoft's cancellation of approximately 2 GW of US and European leases in 2025 was widely interpreted as a signal that self-build threatens the broader colocation market. The more precise read is narrower. Microsoft exited early-stage standard leases in markets where its product could be replicated by self-build. It continued leasing in markets where time-to-power was the binding constraint. The cancellations were not a statement about colocation. They were a statement about undifferentiated colocation.
As hyperscalers scale toward gigawatt-class demand, their negotiating leverage over undifferentiated wholesale operators increases proportionally. A 12-year fixed wholesale contract signed at today's compressed vacancy premiums does not guarantee the same pricing power at renewal. Operators whose product is replicable square footage will face that renewal on the hyperscaler's terms. Operators whose product is a pre-approved grid connection and a completed permitting stack will not because no capital can recover the time already spent in the entitlement queue.
Grid connections in top US markets now require 24–36 months of advance planning. In 2025, 25 planned data center projects were cancelled due to regulatory opposition, quadrupling from the prior year. Pre-entitled, power-ready land is the genuinely scarce resource. Operators who control it are selling time-to-power, not square footage. The pricing logic is fundamentally different.
Two sub-segments are structurally protected
Interconnection-anchored colocation carries a network-effect moat that hyperscaler campuses cannot replicate. No self-built AWS facility can offer simultaneous private peering to Azure, Google Cloud, and 50 carriers. Equinix's xScale utilization exceeded 90% across key markets through 2025. The reported $15 billion joint venture with GIC and CPP Investments, closed in February 2026, targets 10–15-year wholesale leases for new hyperscale campuses in Tokyo, Sydney, and Frankfurt. Interconnection density generates platform-effect pricing power that commodity square footage cannot command.
BTS and powered-shell operators controlling pre-entitled land occupy the second protected position. NVIDIA's DGX Blackwell systems draw approximately 120 kW per rack 10 to 20 times traditional enterprise density. No standard multi-tenant hall accommodates this. BTS is the procurement model for large AI workloads. The $30 billion Blue Owl–Meta joint venture in 2025, funding a reported ~2 GW Louisiana campus pre-leased by Meta, illustrates where premium returns are being structured.
The counterargument
At 1–2% vacancy, every operator has pricing power today. That is correct. Bifurcation is directional, not immediate. But infrastructure funds underwrite against 10–15-year cash flows, not against today's vacancy. If commodity colocation assets are priced at multiples that reflect BTS-equivalent structural durability, the pricing error is embedded in today's acquisition. The correction is felt at exit.
The forward implication for infrastructure investors
Three questions determine which side of the bifurcation a colocation asset sits on. Does the operator control pre-entitled land in a constrained market? Is the asset interconnection-anchored with network-effect density? Does the operator have genuine BTS and powered-shell capability for hyperscaler-grade single-tenant delivery? Assets that answer yes deserve premium multiples. Assets that do not are being valued incorrectly against those that do. That mispricing is present in the market today.
Read the full breakdown: The Self-Build Surge Will Not Kill Colocation. It Will Split It.
Obinna Isiadinso Completely agree with this. The split is real but most people are still underwriting colocation as if it is one category. Where it really gets mispriced is earlier. A lot of sites being marketed as power ready or near infrastructure do not hold up once the real path to power and permitting is defined. By the time capital gets involved the timeline assumptions are already baked in and rarely get challenged. The groups that actually control time are the ones that did the work early, not the ones reacting to it later.
Great article. I’ve personally seen city and state wide network rings deployed in colo environments, and the reality is everything depends on them. Even hyperscalers maintain a colo presence just to tap into that connectivity. My take…Every major company should have a colo strategy for agility. Keep your core network infrastructure in a place that’s secure, fast, and reliable. From there, you can connect private data centers to cloud and AI platforms much more efficiently while reducing reliance on aging on-prem infrastructure. Think of colo like an online marketplace. Put yourself in a position where hyperscalers have to compete for your business because you can easily connect to any of them in colo.
Always great analysis. Thank you.