Nebius raises GPU prices 20% as supply tightens

By September 16, 2026, the AI neocloud expansion had reached a pricing inflection. BigGo Finance reported that Nebius said it will raise GPU cloud prices by an average of 20 percent from October 1, covering H100, H200, B200 and B300 chips, with B300 rising from $7.85 to $9.50 per GPU-hour. The announcement pushed Nebius shares up nearly 7 percent in after-hours trading.

That increase followed a financing and siting rush that enterprises watched closely. On August 4, 2026, Sharon AI signed a five-year, $373 million contract with a global AI platform customer for AI cloud services, according to tech-insider.org. The first deployment uses 2,048 Nvidia Blackwell Ultra B300 GPUs hosted in NEXTDC data centres in Melbourne and Sydney.

On August 7, 2026, Dentons published an analysis of how such projects get funded. AI-optimized facilities often separate the building shell from the GPU hardware in their capital stacks. Lenders fund the shell, while the chips may be supported through equipment finance, loans secured by the GPUs, or financing from the vendor. The main structural safeguard is a hyperscaler backstop, which turns short-dated neocloud revenue into investment-grade streams. Dentons cited Microsoft's commitment to CoreWeave, which accounted for 62 percent of CoreWeave revenue in 2024 and 67 percent in 2025. Dentons also warned that neocloud counterparties are sub-investment-grade and depend on Nvidia supply.

Data Centre Magazine's August 12 top-10 list noted that Crusoe had raised more than $13 billion in institutional capital and held a multi-gigawatt clean energy pipeline, while CoreWeave had priority access to Nvidia's latest accelerators and long-term deals with Meta and Microsoft.

On August 19, 2026, Nebius priced $5.0 billion in convertible notes, according to a comparison published by MarkTechPost on August 23. The same report said CoreWeave's second-quarter revenue reached $2.575 billion, up 112 percent year over year, while Nebius reported $582.3 million, up 454 percent, with $3.0 billion of AI cloud annualised recurring revenue.

Crusoe reported 4.9 gigawatts of contracted AI infrastructure across five US campuses, with a development pipeline above 40 gigawatts. Sharon AI's total contracted capacity was 132 megawatts, 102 megawatts of which was already under contract, according to tech-insider.org. The same source said Sharon AI planned more than 55,000 GPUs by mid-2027, including a 72-megawatt Nvidia collaboration targeting up to 40,000 GB300 chips.

On September 5, 2026, Tech Times reported that FluidStack closed a $1.5 billion round led by Jane Street at an $18 billion valuation. FluidStack holds no chips itself; it builds modular data centres and runs chip-agnostic clusters for customers such as Anthropic, with Google backstopping billions of dollars in lease obligations at former crypto mining sites.

The latest price signal matters for enterprises purchasing AI compute. Nebius's increase is its second since May, and it follows auction tests where customers paid 15 to 20 percent above the previous maximum listed price for scarce Blackwell capacity. Nebius also said demand visibility now extends beyond 24 months. tech-insider.org reported that committed-contract discounts can reach 60 percent off list price but require multi-year commitments, and recommended hybrid architectures that keep general workloads on hyperscalers while routing GPU-intensive jobs to neoclouds.