Nebius Has 40% Upside in 2026 as Global AI Data Center Shortage Worsens

CBRE says North American data center vacancy has fallen to 0.9%, with Northern Virginia at 0.3% and Dallas-Fort Worth 88% pre-leased on 716.7 MW under construction, underscoring a severe AI infrastructure shortage. Nebius is positioned as a direct beneficiary, reporting $399 million in revenue (+684% y/y), 841% AI cloud growth, a $50 billion backlog, and more than 3.5 GW of contracted power capacity. The article argues this capacity, plus Nvidia’s equity stake and large customer commitments from Microsoft and Meta, supports continued upside for the stock despite valuation risk.

Analysis

The market is underpricing how quickly AI demand converts into a real estate and utilities scarcity regime. Once vacancy gets this tight, the pricing power shifts from compute buyers to capacity owners, and the winners become those who can pre-commit power, land, and interconnects years ahead of delivery. That creates a compounding moat: scarce capacity attracts the best customers, which in turn improves financing terms and makes it easier to secure the next tranche of buildout.

Nebius is the cleanest public expression of that bottleneck trade, but the second-order winners may be the enablers rather than the operator itself. GPU supply relationships, high-voltage equipment, switchgear, cooling systems, and grid-interconnection vendors should see unusually durable order backlogs as hyperscalers and independents race to lock in capacity before 2027. The flip side is that any company dependent on near-term AI demand but lacking secured power may face margin compression as spot capacity costs continue to rise faster than utilization can fill them.

The key risk is not demand disappearing; it is supply response. This setup invites a wave of capital formation into new markets, modular data centers, and power-intensive regions, which could cap pricing power in 12-24 months if permitting and grid timelines accelerate. The other risk is concentration: a handful of large customers can renegotiate or defer if their own capex priorities shift, so the market should watch backlog quality, not just backlog size.

Consensus appears to be treating this as a simple AI beneficiary trade, but the more durable thesis is an infrastructure bottleneck trade with much higher optionality on time. If capacity remains constrained into 2027, the equity market will likely reward owners of scarce assets more than model builders or chip sellers. That argues for owning the picks-and-shovels of scarcity while being selective on names where execution and customer concentration can overwhelm the scarcity premium.

Key Decisions for Investors

• Maintain/accumulate long NBIS on pullbacks over the next 2-6 weeks; the risk/reward favors staying with the scarce-capacity winner, but size it below a full core position given customer concentration and valuation risk.
• Add a basket long in AI infrastructure enablers — NVDA, power equipment, and data-center supply chain names — on any 5-10% market-wide AI selloff; the cleaner expression is to own the bottlenecks rather than the end-demand names.
• Pair trade: long NBIS / short a broad AI software basket over 3-6 months; if the market continues to bid infrastructure scarcity over model monetization, the spread should outperform even if AI sentiment cools.
• Use call spreads instead of outright longs for NBIS into the next earnings cycle; upside remains strong, but implied expectations are elevated and defined-risk convexity is better than cash equity here.
• Watch for policy or permitting catalysts over 6-18 months; if grid approvals accelerate in key markets, take partial profits on infrastructure scarcity trades because the strongest pricing power window may narrow faster than consensus expects.