#01
CoreWeave
Best for: Largest B200 reference fleet
First neocloud at-scale with GB200 NVL72 racks. Full NVLink Switch System and Quantum-X800 InfiniBand.
Pricing: $5.50/B200-hr · GB200 NVL72 by quote
[Best of 2026] · B200 Providers
Best B200 Providers
The 10 best NVIDIA B200 / GB200 NVL72 providers in 2026 — ranked by available capacity, fabric, and pricing.
[Methodology]Scored on installed B200 count, GB200 NVL72 availability, on-demand $/hr, 1-year reserved $/hr, and fabric (NVLink Switch + Quantum-X800 or EFAv3).
[AI Summary]
The best B200 providers in 2026 are CoreWeave, Lambda, Crusoe, and Nebius — the first neoclouds to deploy NVIDIA Blackwell at scale on InfiniBand NDR or Spectrum-X. GB200 NVL72 racks ship from CoreWeave and Crusoe today; most hyperscalers are still on waitlist. On-demand B200 lands at $5.50–$7.20/GPU-hr; reserved 1-year cuts roughly 35–45%.
#02
Crusoe
Best for: B200 at gigawatt scale
Abilene factory operating B200 capacity dedicated to OpenAI plus merchant inventory.
Pricing: $5.00/B200-hr
#03
Lambda
Best for: Self-serve B200 clusters
B200 1-Click Clusters launched Q1 2026. Strong developer ergonomics on top.
Pricing: $5.99/B200-hr
#04
Nebius
Best for: EU B200 with vertical integration
B200 capacity in Finland and Kansas City with managed Slurm + Kubernetes.
Pricing: $4.50/B200-hr
#05
Oracle Cloud Infrastructure
Best for: Enterprise B200 with Stargate
GB200 NVL72 capacity inside Stargate sites, available to enterprise customers under reserved terms.
Pricing: Enterprise contract
Score
89.7
#06
AWS (p6-b200)
Best for: B200 inside the AWS stack
p6-b200 instances launched 2026 with EFAv3 for B200-grade RDMA networking.
Pricing: ~$8.00/B200-hr on-demand
Score
88.5
#07
Microsoft Azure (ND GB200 v6)
Best for: OpenAI-aligned B200 capacity
GB200 NVL72 capacity reserved across Quincy, Mt Pleasant, and Atlanta AI sites.
Pricing: Azure contract
Score
87.9
#08
Together AI
Best for: B200 inference for open models
B200-optimized inference for Llama-4 and DeepSeek-class models with per-token billing.
Pricing: Per-token APIs · $4.80/B200-hr raw
#09
Google Cloud (A4 / Trillium)
Best for: B200 alongside TPU v6
A4 VMs ship B200 with Google's Jupiter fabric. Pair with TPU v6 Trillium for hybrid stacks.
Pricing: GCP contract
Score
85.0
#10
G42 / Khazna
Best for: Sovereign B200 in MENA
GB200 capacity allocated across Abu Dhabi sites under the Stargate UAE program.
Pricing: Sovereign contract
Score
83.4
[FAQ]
Frequently asked questions
- What is the NVIDIA B200?
- B200 is NVIDIA's 2025-launched Blackwell GPU: 192 GB HBM3e, 8 TB/s memory bandwidth, 20 PFLOPS FP4 dense. GB200 pairs two B200s with one Grace CPU; GB200 NVL72 racks 72 B200s on a single NVLink Switch domain.
- Who has B200 available in 2026?
- CoreWeave, Crusoe, Lambda, Nebius, AWS (p6-b200), Azure (ND GB200 v6), Oracle, and Google Cloud all ship B200 today. Sovereign capacity is allocated through G42/Khazna and Humain.
- How much does a B200 cost per hour?
- On-demand B200 ranges from ~$4.50/hr (Nebius) to ~$8.00/hr (AWS p6-b200). Reserved 1-year contracts typically cut 40–55% off list. Use our B200 Cost Calculator to model your spend.
- Should I use B200 or H200?
- B200 wins for any workload memory-bound on 141 GB HBM3e (H200's cap) — Llama-3 405B inference, 100B+ MoE training. H200 is still cheaper per token for sub-70B inference.
[Knowledge Graph]
Company → Datacenter → GPU → Customer → Industry
Every entity on this site is cross-linked. Follow the graph from operators down to specific facilities, GPU clusters, customers, and sectoral context.
[Rankings]
[Comparisons]
[Best of 2026]
[Pricing & Calculators]
[Infrastructure profiles]
[News & analysis]