Investors are questioning neoclouds' business model but Nebius says AI computing prices are rising

Nebius CRO Marc Boroditsky said AI-computing prices are rising faster than expected and that Nebius could sell out its 2027 capacity if brought online, with similar demand possible in 2028. He said Nebius does not carry debt like CoreWeave; the sector has faced scrutiny over debt-funded data-center investment, and Nebius and peers had fallen sharply over the prior month. Separately, SpaceX reported AI-business revenue of $7.56 billion, up ninefold, while its CFO said pricing was slightly above the high end of a $30–$50-per-watt range; Nebius and other cloud-provider stocks rose in Wednesday trading.

Analysis

The market implication is not simply “AI demand is strong”; it is whether realized rental rates and utilization can outrun the all-in cost of capital and power. Higher prices for older GPUs, if independently confirmed across transactions, would extend useful lives and improve residual values—supporting infrastructure returns while reducing near-term impairment risk. But executive comments and a single auction are weak evidence of durable market-wide pricing: capacity mix, contract duration, utilization, power costs, and customer concentration determine whether price per megawatt converts into cash flow.

NBIS’s relative advantage over CRWV is plausible only if its lower leverage claim holds after accounting for leases, committed capex, and future funding needs. Strong demand can worsen financing risk if operators must build ahead of customer cash receipts. For CRWV, borrowing costs are a direct counterweight to better rental pricing; for IREN and Applied Digital, the key question is whether announced capacity can be funded and brought online at attractive returns. NVDA benefits from sustained GPU utilization, though durable older-chip demand could modestly lengthen replacement cycles at the margin.

Near term, sentiment and price action may lead fundamentals. Over 1–3 months, watch realized auction rates, utilization, debt spreads, and financing disclosures. Over 6–18 months, power availability, delivery timing, and capex discipline will decide who captures returns. Treat SpaceX’s reported pricing and revenue as a demand signal, not proof of broad neocloud economics. The contrarian risk is that scarcity pricing is real but temporary: new capacity, customer efficiency gains, or tighter credit could reverse the economics.

Key Decisions for Investors

• Consider a small, relative-value long NBIS / short CRWV only after verifying NBIS’s debt, lease obligations, committed capex, and liquidity against CRWV’s; the thesis is lower financing sensitivity, not a proven valuation discount. Reassess if NBIS discloses materially larger funding needs or CRWV’s borrowing spreads ease without deterioration in operating metrics.
• Do not chase the sector on executive pricing commentary alone. Set an alert for repeat auction evidence and disclosed realized rates alongside utilization and power costs; improving headline pricing without improving cash conversion is a failed signal.
• Keep NVDA exposure conditional rather than treating older-GPU pricing as unambiguously bullish: durable rental demand supports utilization and residual values, but monitor customer adoption of newer chips and replacement cadence for evidence of delayed upgrade cycles.
• For APLD and IREN, require evidence that capacity additions are financed and contracted before adding exposure. Track funding terms, delivery schedules, and customer commitments over the next 1–3 months; power or financing constraints that delay projects would invalidate the demand-to-earnings thesis.