The term 'neocloud' now encompasses five companies with distinctly different business models. CoreWeave and Nebius are publicly traded, releasing quarterly earnings and SEC filings—CoreWeave on Form 10-Q and Nebius as a foreign private issuer on Form 6-K. Lambda and Crusoe remain private, both inching toward IPOs. Groq, which has operated GroqCloud on its LPU architecture since 2024, pivoted entirely to inference-cloud infrastructure after licensing its technology to NVIDIA in December 2025. This article compares these providers on the factors that matter most to buyers: published pricing, deployed and contracted power, hardware roadmaps, contract structures, and independent quality ratings.
TL;DR
- CoreWeave remains the only Platinum-rated GPU cloud in SemiAnalysis ClusterMAX 2.0, commanding premium list pricing on H100, H200, and B200.
- Nebius is the sole provider here publishing B300 on-demand pricing; Lambda offers the lowest B200 on-demand rate.
- Lambda still holds the cheapest on-demand B200 at $6.69/GPU-hour, but lacks a spot tier.
- Crusoe is the only provider featuring AMD MI300X/MI355X on its rate card.
- Groq is now an NVIDIA Cloud Partner, planning to add NVIDIA GPU capacity alongside its LPU inference cloud.
Company Overview and Key Metrics (as of August 21, 2026)
| Metric | CoreWeave (NASDAQ: CRWV) | Nebius (NASDAQ: NBIS) | Lambda (Private) | Crusoe (Private) | Groq (Private) |
|--------|--------------------------|----------------------|------------------|------------------|----------------|
| Recent Financial Highlights | Q2 revenue $2.575B, +112% YoY | AI cloud ARR $3.0B | B200 on-demand $6.69/hr (lowest published) | 4.9 GW contracted, 40 GW+ pipeline | $350M Series A at $3.5B (Aug 17) |
| Notable Announcements | First Vera Rubin NVL72 bring-up | $5.0B converts priced Aug 19 | Michel Combes as CEO (May 2026) | Only AMD MI300X/MI355X on a rate card | 13 data centers, 54 MW → 200+ MW by 2027 |
Pricing Landscape: What Providers Charge
Published on-demand pricing is the clearest differentiator. CoreWeave sets premium rates across its H100, H200, and B200 offerings, reflecting its Platinum rating. Nebius, meanwhile, is the only provider to publish B300 on-demand pricing, targeting early adopters of next-generation hardware. Lambda's $6.69/GPU-hour for B200 undercuts competitors, making it the most cost-effective option for burst workloads—though the absence of a spot tier may limit savings for interruptible tasks.
Crusoe's inclusion of AMD MI300X and MI355X on its rate card offers an alternative to NVIDIA-centric clouds, potentially appealing to price-sensitive buyers or those diversifying vendor lock-in.
Power and Infrastructure: The Real Constraint
Contracted power is a critical metric for enterprises needing scale. Crusoe leads with 4.9 GW contracted and a pipeline exceeding 40 GW, positioning it as a hyperscale-capable partner. CoreWeave and Nebius, as public companies, disclose power details in filings, allowing for direct comparison. Groq's trajectory—from 13 data centers at 54 MW to over 200 MW by 2027—shows aggressive expansion, albeit from a smaller base.
Hardware Roadmaps: Beyond Current Gen
CoreWeave's early bring-up of Vera Rubin NVL72 signals a commitment to next-generation NVIDIA platforms, while Nebius's B300 pricing suggests early access to Blackwell Ultra. Lambda's focus on B200 at aggressive prices may pressure margins but secures volume. Crusoe's AMD support broadens the ecosystem. Groq's post-NVIDIA-license strategy now includes NVIDIA GPUs, merging its LPU strengths with mainstream hardware.
Contract Structure and Quality Ratings
CoreWeave's Platinum ClusterMAX 2.0 rating underscores operational excellence, justifying its premium pricing. Contract terms vary: public entities like CoreWeave and Nebius offer standardized agreements, while private ones like Lambda and Crusoe may negotiate more flexibly. Groq's transition to an inference-cloud pure play adds a specialist option.
Conclusion
Choosing a neocloud in 2026 depends on priorities: CoreWeave for top-tier reliability, Nebius for cutting-edge B300 access, Lambda for budget B200, Crusoe for AMD alternatives and massive power reserves, and Groq for inference-focused workloads. With pricing and power dynamic, buyers must weigh published rates against long-term contracts and hardware roadmaps to secure the best fit.
via MarkTechPost