GPU Clouds and Neoclouds: CoreWeave, Lambda, Nebius — 2026-09-12
Recent market analyses highlight a divergence in neocloud economics, with CoreWeave and Nebius securing massive power contracts while Lambda leverages private debt to meet Microsoft’s demand. Despite falling H100 spot prices, effective costs for users remain high due to utilization inefficiencies, and investors are scrutinizing the "circular financing" models that underpin the sector's rapid expansion.
GPU Clouds and Neoclouds: CoreWeave, Lambda, Nebius — 2026-09-12
Lambda secures $1B debt facility for Microsoft contracts
On September 8, 2026, reports confirmed that neocloud provider Lambda has secured approximately $1 billion in private, short-dated debt to purchase Nvidia AI infrastructure specifically for leasing to Microsoft. This move underscores the intensifying capital intensity of the GPU rental market, where providers must borrow heavily against future contract revenues to acquire the latest chips. For competitors like CoreWeave and Nebius, this signals that hyperscaler-backed demand is driving a new wave of debt-financed capacity expansion, potentially raising barriers to entry for smaller players who cannot access similar credit lines.
Effective GPU costs remain high despite price drops
A new analysis from September 8, 2026, reveals that while nominal GPU-hour rates have dropped, the effective cost per useful hour remains significantly higher due to low utilization rates. At a measured 5% GPU utilization, a committed rate of $3.44/hr effectively bills out at $68.80 per useful hour for some workloads. This discrepancy matters for CoreWeave and Nebius customers who may see lower sticker prices on platforms like Nebius ($329K for 64 B200s vs $742K on Google Cloud) but fail to achieve the same efficiency if their workloads are not optimized, highlighting that raw price comparisons often mislead on total cost of ownership.
Investors question circular financing and debt loads
Financial analysts are increasingly warning about the risks associated with "circular financing" in the AI infrastructure sector, where chipmakers like Nvidia invest in neoclouds like CoreWeave and Nebius, which then use those funds to buy more Nvidia chips. With CoreWeave holding a $104 billion backlog but also carrying over $35 billion in debt, and Nebius facing similar funding gaps, the sustainability of this model is under scrutiny as interest rates remain elevated. This pressure could force neoclouds to prioritize high-margin contracts or seek equity dilution, impacting stock valuations for all major GPU-rental clouds.
Local view
Tenten (Traditional Chinese Tech Media) Tenten reported on September 11, 2026, that data from Ornn indicates A100 inference costs are now less than half those of H100s. The outlet highlights that while A100 five-year lease prices retain 80.2% of their one-month quoted value, this does not equate to residual value for single cards. Tenten argues that Ornn’s cost and rental data help bring AI infrastructure assessment back to actual workload needs and cash flow rather than speculative hardware appreciation.
Context & numbers
H100 Price Volatility Contrary to earlier trends, H100 rental prices have seen a recent surge. On September 7, 2026, compute trading platform Ornn stated that H100 rental prices rose 22% over the past month to reach $3.28 per hour. This contradicts other reports suggesting H100 prices have halved to $3.38/hr earlier in 2026, indicating significant volatility and potential regional or contract-type disparities in pricing.
Power Capacity Constraints CoreWeave and Nebius have each secured 3.5 GWs of contracted power capacity, but the vast majority remains undeployed. CoreWeave targets 1.7 GW of active power by the end of 2026. This gap between contracted and active power is a critical bottleneck; as neoclouds race to fill their backlogs, the ability to energize these sites quickly will determine who can meet immediate demand versus those who will face delays despite having signed contracts.
On the radar
- Q3 Financials: Investors are awaiting Q3 earnings reports from CoreWeave and Nebius to see how much of their massive backlogs (CoreWeave >$100B) have converted to active revenue and what new debt financing has been undertaken.
- Utilization Metrics: Watch for further industry data on GPU utilization rates. If the "effective cost" gap highlighted by Beri.net continues to widen, customers may shift toward hyperscalers despite higher nominal prices, betting on better managed services and higher uptime.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
