What a top-tier data center market looks like, according to Bernstein

Bernstein says U.S. data center demand remains strongest in Major and Minor Metro markets, where Digital Realty Trust and Equinix are viewed as most insulated from buildout risks. Northern Virginia has about 8GW of supply, roughly 5GW under construction, seven-year power queues, and pricing near $217/KW/mo with vacancy below 1%; Atlanta pricing is about $175/KW/mo with vacancy around 2%. Bernstein rates both DLR and EQIX Outperform, with price targets of $232 and $1,222, respectively, and says their major-metro build strategy carries limited execution risk.

Analysis

The important second-order takeaway is that scarcity is no longer just about power; it is about entitlement quality. In the hardest-to-build markets, incumbents with operating campuses and deep utility relationships will see a widening moat as incremental supply gets pushed out multiple years, which should keep pricing power intact even if new construction headlines slow. That favors the large-scale colocation landlords over pure-play developers and should compress the opportunity set for smaller, more levered entrants that need fresh supply to re-rate.

This also changes where the value accrues in the AI infrastructure stack. Latency-sensitive inference and enterprise workloads prefer the same constrained metros, so edge-oriented demand becomes a de facto support for incumbent REIT cash flows rather than a separate growth theme. The risk is that investors overpay for the obvious winners while missing that hyperscale demand can be cyclical at the margin; a 6-12 month pause in cloud capex would hit rural/speculative pipelines first, but would barely dent the premium metro landlords.

The main contrarian point is that the market may be underestimating how long pricing can stay elevated in the top metros because vacancy is already too tight for a meaningful correction. That said, this is not a clean secular long if rates stay high: cap rates and financing costs can offset rental gains, so the trade works best if the Fed eases into 2025 while power bottlenecks persist. The asymmetric setup is strongest where existing assets, balance-sheet strength, and execution in constrained locations intersect.

Key Decisions for Investors

• Long EQIX vs short a basket of smaller data-center developers with heavy speculative build exposure over 3-6 months: the pair isolates metro scarcity premium while hedging broader sector valuation risk.
• Add to DLR on pullbacks and use a 6-12 month horizon; risk/reward is favorable as constrained-market pricing should support same-store NOI even if new starts slow.
• Avoid chasing rural/speculative data-center names until there is evidence of power delivery and pre-leasing conversion; the downside is a capital-intensive air pocket if hyperscale demand normalizes.
• If we want convexity, buy medium-dated calls on EQIX or DLR into any rate-driven selloff; the trade works best if lower funding costs arrive before new metro supply materially clears.