CoreWeave Leads GPU Neocloud Race as Backlog Hits $104B
Disclosures from top GPU neoclouds like CoreWeave and Nebius reveal a rapidly scaling market where providers are aggressively competing on power capacity and next-generation chip pricing.
The GPU "neocloud" market has entered a phase of massive scale, led by newly public giants CoreWeave and Nebius. CoreWeave reported Q2 2026 revenue of $2.575 billion, up 112% year over year, backed by a $104 billion backlog that grew by another $25 billion in early Q3. The company, guiding for up to $39 billion in 2026 capex, has 1.5 GW of active power and over 4.2 GW contracted across 51 data centers. Meanwhile, Nebius posted Q2 group revenue of $582.3 million, with its AI cloud segment reaching an annualized run-rate of $3.0 billion. Nebius targets 5 GW of contracted power by the end of 2026.
For practitioners, this expansion drives highly competitive on-demand pricing. Lambda offers the cheapest published NVIDIA HGX B200 rate at $6.69 per GPU-hour, though it lacks a spot tier. Nebius undercuts CoreWeave on Blackwell chips, offering B200 instances at $7.15 per hour and B300 instances at $7.85. In contrast, CoreWeave commands a premium, charging $8.60 for B200 and $6.16 for H100, which is about 60% higher than Nebius's $3.85 and Lambda's $3.99 H100 rates. Crusoe offers the cheapest Hopper-class option with the H200 at $4.29 per hour, and is the only provider listing AMD Instinct MI300X on-demand at $3.45 per hour.
The race to deploy next-generation silicon is also accelerating. CoreWeave completed the first bring-up of NVIDIA Vera Rubin NVL72 systems in Q2, while Nebius is validating its own Rubin units. Lambda, under new CEO Michel Combes, secured a multibillion-dollar Microsoft deal that includes GB300 NVL72 systems. Crusoe is expanding with the 1.2 GW Abilene campus for Oracle and OpenAI, alongside a 900 MW campus for Microsoft, bringing its contracted capacity to 4.9 GW.
Even specialized inference providers are pivoting. Groq, which raised a $350 million Series A at a $3.5 billion valuation in August 2026, joined the NVIDIA Cloud Partner program to add GPU capacity alongside its LPU-based GroqCloud. For developers, these shifts mean that while CoreWeave offers premium, Platinum-rated reliability for frontier-scale training, alternative neoclouds now provide highly viable, budget-friendly options for mid-sized training, fine-tuning, and AMD hardware evaluation.
This is our own summary of reporting by MarkTechPost
