2026 GPU Neocloud Pricing and Capacity Showdown

Marktechpost released a data‑driven comparison of the five biggest GPU neocloud operators as of August 2026. The analysis covers CoreWeave, Nebius, Lambda, Crusoe and Groq, focusing on published rate cards, active and contracted gigawatt capacity, recent financial results and the hardware roadmaps each company follows.

CoreWeave, a Nasdaq‑listed firm, posted Q2 2026 revenue of $2.575 billion, a 112 percent year‑over‑year increase, and a backlog of roughly $104 billion. Its active power stood at 1.5 GW, while contracted power reached about 3.7 GW, with a public claim of over 4.2 GW across 51 data centers. Nebius, also public, reported $582.3 million in group revenue, a 454 percent jump YoY, and an AI‑cloud run‑rate of $3.0 billion. The company announced a contracted‑power target of 5 GW for year‑end and plans to add more than 1 GW annually from 2027 onward. Lambda, a private player, secured a $1.5 billion Series E round in late 2025 and a $1 billion credit facility in mid‑2026, while Crusoe raised $1.375 billion in a Series E at a valuation above $10 billion and is reportedly courting a $3 billion raise at a $30 billion valuation. Groq, after licensing its LPU inference chip to NVIDIA in December 2025, raised $650 million in June and $350 million in August 2026, operating 13 data centers and targeting 200 MW of capacity by 2027.

The market context is a rapid expansion of AI‑intensive workloads that demand both training‑scale and inference‑scale GPU resources. Public filings from CoreWeave and Nebius provide transparent quarterly data, a rarity among cloud providers, and signal that investors are closely watching the scalability of GPU infrastructure. The emergence of “neocloud” terminology reflects a shift toward specialized AI‑focused data centers that differ from traditional hyperscalers in pricing flexibility and contract structures.

Pricing differences are stark. CoreWeave’s H100 rate of $6.16 per GPU‑hour sits about 60 percent above Nebius and Lambda, and SemiAnalysis estimates a 10‑15 percent premium over other managed clusters. Nebius undercuts CoreWeave on Blackwell‑class chips, offering a B200 price of $7.15 and a unique B300 on‑demand rate of $7.85. Lambda provides the cheapest published B200 instance at $6.69, though its 1‑Click Cluster bundles are priced higher for larger GPU counts. Crusoe stands out as the sole AMD option, listing an H200 price of $4.29 and an H100 rate of $3.90, comparable to Nebius and Lambda. Groq does not publish GPU‑hour rates; instead, its GroqCloud service is billed per token, positioning it as a niche inference offering.

The analysis also highlights limits and risks. Spot‑tier pricing is only available from CoreWeave and Nebius, with Nebius pre‑emptible H100 at $2.15 and CoreWeave spot at $2.46 per GPU‑hour, while Lambda lacks a spot market altogether. Several providers, including CoreWeave and Lambda, have not disclosed pricing for the newest Vera Rubin NVL72 hardware, leaving buyers to negotiate sales contracts. Large‑scale capacity deals, such as Nebius’s $12 billion five‑year order from Meta, create floor utilization but also tie up significant capital, potentially affecting flexibility.

Looking ahead, the competitive landscape will be shaped by upcoming capacity expansions and financial milestones. CoreWeave expects active power above 1.85 GW by year‑end and a full‑year revenue range of $12.4‑$13.2 billion. Nebius aims to deploy over 1 GW per year starting in 2027, while Crusoe’s development pipeline exceeds 40 GW across multiple campuses. Groq’s partnership with NVIDIA’s LPX platform suggests future hybrid offerings that blend LPU inference with GPU acceleration. Investors and customers should monitor IPO filings from Lambda and Crusoe, as well as any shifts in pricing strategy that could arise from the growing demand for AI‑specific hardware.