Neocloud Boom: Comparing Nebius, CoreWeave, & IREN

AI hyperscalers including Microsoft, Alphabet, Meta, Amazon and Oracle are projected to spend about $700 billion on AI infrastructure in 2026 and $729 billion in 2027, supporting demand for specialized GPU-cloud providers. Zacks identifies Nebius as the strongest neocloud candidate, with 2026 revenue projected to rise 530.09%, versus 346% for IREN and 151% for CoreWeave; Nebius also holds $8.04 billion of cash, above IREN's $5.9 billion and CoreWeave's $5.52 billion. The upside is tempered by expected ongoing losses, capital-intensive data-center and GPU spending, and reliance on debt financing, although Nebius's 2.3% debt cost is well below CoreWeave's 8.3%.

Analysis

The relevant differentiation is not headline compute demand but contract quality: prepaid capacity commitments shift construction risk from NBIS to customers, while shorter-duration or concentrated commitments leave CRWV and IREN exposed to utilization and repricing. In a GPU-rental market, revenue growth without a rising contracted-revenue-to-capex ratio is value destructive because each incremental deployment requires upfront GPUs, power, and site spend before cash conversion. NBIS therefore deserves a relative premium only if disclosed backlog converts into revenue on schedule and capex per deployed MW falls; the cited estimates are not independently sufficient evidence.

CRWV is the most vulnerable equity if rates remain elevated or GPU residual values disappoint. Its higher cash interest creates a reflexive problem: weaker equity raises the cost of funding expansion, which limits capacity additions and can cause customers to diversify toward hyperscalers or lower-leverage neocloud peers. The second-order winner is NVDA, whose economics are insulated from which lessor owns the GPU, while data-center power and electrical-equipment suppliers should retain pricing power if capacity remains the bottleneck.

Consensus appears too willing to treat neoclouds as asset-light software proxies. They are closer to leveraged infrastructure operators with technology-obsolescence risk: new GPU generations can compress pricing on prior fleets before debt amortizes. Over the next 1-3 months, earnings disclosures on contracted capacity, customer concentration, power availability, and funding terms matter more than reported top-line growth; over 6-18 months, hyperscaler in-house buildouts could cap neocloud utilization and exit multiples.

Key Decisions for Investors

• Maintain a 3-6 month relative-value long NBIS / short CRWV position, sized beta-neutral. Thesis is funding-cost and contract-structure divergence; target a further 15-25% relative move. Exit if CRWV secures materially lower-cost refinancing or NBIS reports a backlog conversion delay, material customer cancellation, or a step-up in capex intensity.
• Do not chase NBIS outright after its strong run. Add only following earnings if management discloses contracted revenue coverage sufficient to fund the next 12 months of buildout without equity issuance; otherwise treat it as a watch item rather than a momentum long.
• Use NVDA as the cleaner 6-12 month AI-infrastructure exposure versus neocloud equities: it captures fleet deployment across NBIS, IREN, CRWV, and hyperscalers without direct utilization or project-finance risk. Reduce if hyperscaler capex guidance weakens materially or if supply lead times normalize faster than demand.
• Avoid unhedged CRWV long exposure until interest expense as a percentage of revenue demonstrably declines and financing terms improve. A negative catalyst is any equity-linked financing, additional high-coupon debt, or customer concentration disclosure; these would likely pressure both FCF expectations and valuation multiple.