Meta Just Signed a $27 Billion Artificial Intelligence (AI) Deal. Here's the Under-the-Radar Stock That Won.

Meta forecasts up to $135 billion in capex for the year and signed a five-year capacity deal with Nebius worth up to $27 billion (including $12B committed and $15B optional), expected to come online in 2027. Nebius already has a $19.4B agreement with Microsoft and a $2B capital infusion from Nvidia; management had guided 2026 ARR of $7–9B (midpoint implying ~540% YoY growth), and Nebius trades at roughly 3.6x forecasted ARR today, which the author views as an attractive valuation given contracted hyperscaler revenue. Key risks are execution on data-center construction and capital intensity; recommendation is buy-and-hold with dollar-cost averaging for long-term AI infrastructure exposure.

Analysis

Nebius-style capacity aggregators are shifting the locus of pricing power up and down the stack: they can extract rental yield from hyperscalers while simultaneously improving negotiating leverage with accelerator vendors by consolidating demand. That creates a durable margin opportunity for platform operators but also concentrates execution risk (land, grid, transformers, permitting) into a smaller set of build teams — any delay multiplies revenue slippage across multi-year contracts.

A second-order beneficiary set is the physical-infrastructure supply chain — long-lead copper, transformers, substation contractors and firms that can deliver turnkey PPA+storage solutions. Those suppliers will see order books lengthen and pricing power surge, which in turn raises the marginal cost of on-prem alternatives and tilts more demand toward leased capacity arrangements.

Key downside pathways are technological and policy: a material step-change in model efficiency or a credible alternative accelerator (or national export controls) would compress marginal GPU demand and re-rate the whole lease-model. Timeframes matter — market sentiment reacts in weeks, capacity realization happens over 6–24 months, and structural winners/losers crystallize over multiple capex cycles (3+ years). For portfolio construction, treat the space as convex: high upside tied to execution, large downside tied to build/dilution risk, so hedge actively.

Key Decisions for Investors

• Initiate a phased long position in NBIS equity (2% AUM total, 8 equal monthly tranches) to capture secular rent capture; pair each tranche with a 12-month 20% OTM put (cost-target ~0.5% AUM) to cap first-year downside to roughly -20% while retaining upside if execution and pricing remain intact.
• Buy an 18-month NVDA call spread to express continued tightness/pricing power in accelerators (buy 1.25x notional of 18-month 30% OTM calls, sell 0.75x notional of 18-month 60% OTM calls) — lowers premium while keeping 2–4x upside if GPU pricing stays elevated; sizing 1% AUM.
• Small asymmetric hedge against hardware-displacement risk: purchase 18–24 month INTC puts sized at 0.5% AUM to protect the portfolio from a faster-than-expected shift away from legacy CPU-centric architectures; this limits long-book gamma if an alternative accelerator gains share suddenly.
• Monitor two triggers to adjust sizing: (A) any publicized >30 day construction delay or PPA renegotiation at a major site — reduce NBIS exposure by 50%; (B) evidence of sustained model-efficiency gains (training compute per parameter down >20% QoQ) — switch 50% of NVDA call spread notional into protective put spreads.