Why Nvidia-Backed Nebius Group Stock Skyrocketed Thursday Morning

Nebius plans to raise on-demand Nvidia GPU compute prices by roughly 20% from Oct. 1, citing exceptionally strong AI-processing demand; individual GPU hourly rates will rise 17%-21%. The company reported Q2 revenue growth of 454% year over year to $582 million and cut its adjusted net loss 64% to $33 million, while maintaining midpoint full-year revenue guidance of $3.2 billion and an $8 billion annualized run-rate target. Shares rose as much as 11.2% intraday before trading up 2.1%, though the stock's 43x sales valuation leaves substantial growth expectations priced in.

Analysis

The key underwriting question is whether higher list rates translate into realized revenue per GPU-hour rather than merely repricing scarce spot capacity. If utilization remains near capacity after implementation, Nebius can improve gross-margin absorption quickly because depreciation, power commitments and data-center labor are largely fixed over a 1-3 month horizon. Conversely, enterprise customers with committed capacity or multi-cloud flexibility may shift marginal training workloads to CoreWeave (CRWV), hyperscalers, or internal clusters, making the change a weaker signal than headline pricing implies.

NBIS's valuation leaves little tolerance for an execution gap: the market needs both sustained utilization and capex discipline to avoid a cycle in which incremental revenue is consumed by accelerated GPU, networking and power spend. The more relevant KPI at the next result is revenue growth relative to deployed compute/capex, plus gross-margin progression and contracted backlog conversion—not another high-level demand claim. A guide-down in annualized revenue, deterioration in cash conversion, or evidence of discounting in reserved instances would likely cause disproportionate multiple compression within days.

NVDA has limited near-term earnings sensitivity to one customer's pricing, but the broader implication is constructive for accelerator demand and, more importantly, for networking/content per cluster if neocloud economics support continued build-outs. The contrarian view is that a spot-price increase may be the last phase of supply tightness rather than proof of durable pricing power: new Blackwell capacity, competing cloud builds and customer-owned clusters can normalize GPU-hour rates over 6-18 months, compressing neocloud returns before hardware demand weakens.

Key Decisions for Investors

• Do not chase NBIS on the initial reaction; wait for post-implementation utilization and realized revenue-per-GPU-hour disclosure. Initiate only if utilization holds and gross margin expands sequentially; exit on guidance reduction or evidence that reserved-instance discounts offset the list increase.
• Use a 1-3 month relative-value expression: long NVDA / short NBIS in matched beta-adjusted notional. NVDA captures continued infrastructure demand with less dependence on cloud pricing power; the principal risk is NBIS demonstrating sustained operating leverage faster than consensus.
• Monitor CRWV and other GPU-cloud proxies for competitive responses, especially discounted committed-capacity offers. A broad pricing increase without competitor discounting validates industry scarcity; a widening NBIS-to-peer rate premium is a warning that workload substitution may cap realized pricing.
• For an existing NBIS long, consider trimming into strength and retaining upside only through defined-risk calls after the next earnings report. The asymmetric downside comes from valuation compression if capex rises faster than contracted revenue, while the upside case requires independently verifiable margin and cash-flow inflection.