BNP Paribas: CoreWeave Has Catch-Up Opportunity; Nebius Downgraded to Neutral
Claire Weston
BNP Paribas analyst Stefan Slowinski sees a potential catch-up trade in CoreWeave relative to Nebius, while downgrading Nebius to neutral — the stock's rally has outrun its reported fundamentals, and CoreWeave's risk-reward remains attractive in the current pricing environment.
How strong is the pricing backdrop for compute?
Slowinski anchors to SpaceX's recent AI-infrastructure contracts with Anthropic and Google, estimating such deals generate $30–50 billion in revenue per gigawatt.
This means → large customers are locking in long-term premium pricing; the overall backdrop is supportive.
On the demand side, AWS recently raised GPU reservation prices by 20% — a clear sign that compute supply still trails demand.
Nebius rides the right trend — so why the downgrade?
The market is shifting from "token maximization" — throwing maximum compute at inference — to "token optimization," achieving the same output with less compute. This means → enterprise customers are actively seeking lower-cost alternatives.
Nebius's AI cloud stack is well suited to fine-tuned open-source models, fitting this shift neatly.
Yet Slowinski cut Nebius to neutral. The reason is straightforward: the stock has rallied sharply year-to-date, and the thesis has not yet shown up in reported financials. In plain terms = the narrative is ahead of the numbers.
CoreWeave's stock lags — risk or opportunity?
Slowinski sees CoreWeave's underperformance relative to Nebius as a "potential catch-up opportunity."
CoreWeave is scaling capacity aggressively; heavy near-term capex depresses headline margins and obscures its underlying unit economics — how much each unit of compute actually earns.
New contracts already price in component inflation, keeping contribution margins around 25%. This means → new-deal profitability is not being eroded by rising costs.
When does the margin volatility narrow?
CoreWeave's active compute sits at roughly 1 gigawatt, far above Nebius's approximately 200 megawatts.
As new deployments shrink as a share of the installed base, margin swings should tighten. In plain terms = the larger the base, the less each new rack moves the overall numbers.
Whether this plays out hinges on sustained execution improvement in H2 2026 — a test for management, not for the model.
What is the leverage logic for emerging cloud players?
Slowinski notes the sector is still early-stage; companies can accept unlevered returns below those of hyperscalers such as AWS and Azure.
Financial leverage — borrowing to expand — helps deliver "attractive project-level levered IRRs." This means → unlevered returns look modest, but debt amplifies single-project economics to competitive levels.
This reflects a deeper signal: in a window of effective asset-backed financing and declining cost of capital, the leverage strategy itself is a core competitive tool for this class of company.
Content is for reference only, not financial advice.