The neocloud market has splintered into five genuinely different businesses wearing the same label. CoreWeave, the only Platinum-rated GPU cloud in SemiAnalysis ClusterMAX 2.0, just posted Q2 revenue of $2.575 billion — up 112% year over year — with a staggering $104 billion revenue backlog that excludes another $25 billion in commitments added early in Q3. That kind of contracted demand explains why the company can command premium list pricing on H100, H200, and B200 while absorbing a $626 million net loss driven largely by $640 million in interest expense. Nobody else in this group is playing the same game at the same scale.
Nebius is the growth outlier, with AI cloud revenue up 514% to $574.9 million and an annualized run-rate of $3 billion. Its 49.7% adjusted EBITDA margin on AI cloud is remarkable for a hyperscaler at this stage. The company raised its year-end contracted power target to 5 GW and plans to deploy over 1 GW annually starting in 2027 — a buildout pace that would put it in the same league as the majors. Nebius also stands alone in publishing B300 on-demand pricing, which tells you something about who actually has next-gen Blackwell Ultra capacity available today.
Lambda is the price aggressor. Its $6.69 per GPU-hour B200 on-demand rate is the cheapest published among all five providers, though the absence of a spot tier means bargain hunters can’t push pricing lower. The company landed a multibillion-dollar Microsoft deal covering GB300 NVL72 systems and brought in Michel Combes as CEO in May 2026 — a move that signals IPO readiness more than product strategy. Meanwhile, Crusoe remains the only provider with AMD MI300X and MI355X on its rate card, and its contracted infrastructure footprint hit 4.9 GW across five US campuses, including the 1.2 GW Abilene site it’s building for Oracle and OpenAI’s Stargate project.
Groq is the wildcard after the December 2025 NVIDIA licensing deal that CNBC pegged at roughly $20 billion and that sent founder Jonathan Ross and much of the senior team to NVIDIA. The remaining company raised nearly $1 billion across two rounds in mid-2026 and joined the NVIDIA Cloud Partner program in August, but its promised GPU capacity isn’t a published service yet. For buyers, the real question is whether Groq’s LPU inference cloud keeps its performance edge once NVIDIA GPU capacity arrives — or whether it becomes just another reseller of the same silicon everyone else is renting. The pricing pages tell one story; the contract terms and actual clearing rates likely tell a very different one.
