Jensen Huang Just Said a 1-Gigawatt Facility Is Worth $50 Billion to $60 Billion. Here's What That Means for Neocloud Stocks Like Nebius.

Nvidia CEO Jensen Huang estimated that building a 1-gigawatt AI data-center facility costs $50 billion-$60 billion, implying Nebius' targeted 5 gigawatts of contracted power by end-2026 could represent up to $300 billion of completed infrastructure value. Nebius reported Q2 revenue growth of 454% year over year to $582 million, $3 billion of annual recurring revenue, and a year-end ARR target of $7 billion-$9 billion. Customer contract values rose from $12 million per megawatt annually at the start of the year to more than $20 million in Q2 and above $40 million in certain Q3 negotiations, helping prepayments self-fund 50%-60% of capex and support an $8 billion cash position.

Analysis

The critical valuation distinction is between replacement cost and equity value. A gigawatt site is only financeable at premium multiples if power rights, customer contracts, interconnection queue position, and GPU collateral are transferable; otherwise the asset is a highly specialized shell with rapidly depreciating compute inside it. NBIS and IREN therefore deserve more credit for contracted take-or-pay capacity than for headline gigawatts, and investors should haircut any capacity not backed by disclosed customer duration, pricing escalators, and cancellation protections.

Rising customer prepayments improve near-term liquidity but can conceal a duration mismatch: short-duration AI compute contracts may fund assets with 15-25 year power and facility lives, while GPUs require recurring refresh capital. If spot GPU pricing or utilization softens, prepayments roll off before debt amortizes, creating equity dilution risk precisely when lenders tighten advance rates. The more durable beneficiaries are power-constrained infrastructure vendors and generators—VRT, ETN, PWR, CEG and VST—whose economics are less exposed to AI tenant concentration and GPU residual values.

Consensus is likely overextending peak contracted MW pricing into a steady-state valuation. Pricing above $40m/MW annually implies extraordinary scarcity rents; that is bullish for the next 1-3 months if additional contracts validate it, but it invites hyperscaler self-build, competing capacity, and procurement pushback over 6-18 months. The thesis is falsified positively by disclosed multi-year contracted revenue with investment-grade counterparties and capex coverage above 100% from prepayments plus operating cash flow; it is falsified negatively by falling realized revenue per MW, utilization below plan, or incremental equity issuance.

Key Decisions for Investors

• Do not underwrite NBIS on implied replacement-cost math. Maintain a trading-only long bias over the next 1-3 months only if the company discloses contract tenor, customer concentration, and capex funding coverage; otherwise avoid chasing strength after financing-related headlines.
• Prefer a 6-12 month long basket of VRT, ETN and PWR versus a short basket of capital-intensive neocloud exposure led by NBIS, sized small until relative valuations and borrow are verified. The pair captures AI power-density spend while hedging GPU-utilization and external-financing risk.
• Use CEG or VST as the cleaner 12-18 month power-scarcity expression where incremental data-center load is supported by contracted capacity and regulated/merchant pricing. Key risk is a recession-driven power-demand reset or materially faster renewable/interconnection additions.
• Set an event-driven alert on NBIS and IREN for new debt secured by facilities rather than GPUs. Treat a low advance rate, restrictive covenants, or materially higher borrowing cost as a negative signal on collateral quality; a large unsecured or low-cost facility with long-dated customer assignments would validate the funding thesis.