NVIDIA's newest Blackwell generation graphics processors are proving difficult to find on public cloud platforms, with only a small subset of providers offering them at scale. A comprehensive survey of 71 cloud GPU providers reveals that leading-edge chips like the B200 and B300 remain largely unavailable through standard rental channels, forcing AI teams to navigate a fragmented market where access depends heavily on provider tier and geographic location.

Which Cloud Providers Actually Have Blackwell GPUs Available?

The market for Blackwell access splits into two distinct tiers. Among the major hyperscalers, only a few have publicly listed Blackwell inventory. Azure offers B200 through its ND B200 series, while Google Cloud Platform (GCP) includes B200 in its A3 Ultra family. AWS added RTX PRO 6000 Blackwell support in early 2026, initially rolling out the EC2 G7e instance type in US East regions.

The specialist cloud providers, known as neoclouds, have moved faster. CoreWeave, which NVIDIA designated as its first Elite cloud services provider, lists both B200 and B300 in its catalog. Nebius, a European-headquartered provider, also offers leading-edge B200 and B300 capacity. Meanwhile, IONOS in Germany advertises RTX PRO 6000 Blackwell at sub-$2 per GPU per hour, one of the few publicly listed Blackwell options on the European side.

Why Is Blackwell Supply So Constrained Compared to Older Chips?

The scarcity reflects broader production realities. NVIDIA's H100 and H200 chips, which launched years earlier, are now widely available across dozens of providers at competitive rates. The H100 appears in catalogs from hyperscalers like AWS, Azure, GCP, and Oracle Cloud Infrastructure (OCI), as well as neoclouds including Lambda Labs, RunPod, and Crusoe. By contrast, Blackwell's newness means production volumes remain limited, and NVIDIA likely prioritizes direct enterprise relationships over broad cloud availability.

Pricing reflects this constraint. The RTX PRO 6000 Blackwell on IONOS costs under $2 per GPU per hour, but availability is restricted to a handful of single-GPU configurations. CoreWeave's pricing sits closer to hyperscaler rates, reflecting its higher-end enterprise positioning. For teams needing Blackwell capacity at scale, options remain thin compared to the mature H100 market.

How to Find and Compare GPU Availability Across Cloud Providers

• Check Provider Tier First: Hyperscalers like AWS, Azure, and GCP offer broader compliance certifications and cross-service integration but charge 3 to 6 times more than specialist neoclouds for the same GPU. Neoclouds like CoreWeave and Lambda Labs undercut hyperscaler pricing by 50 to 80 percent but offer narrower service catalogs.
• Verify Billing Options Available: Some providers expose only on-demand pricing publicly, while others list spot pricing, reserved instances, or multi-year commitments. RunPod and CoreWeave offer three billing tiers; Lambda Labs publishes on-demand rates only with quote-based multi-week and multi-year options.
• Assess Geographic and Compliance Requirements: European providers like Scaleway, OVHcloud, and IONOS position themselves for EU data residency and GDPR compliance. Scaleway operates entirely on renewable energy across three EU regions. Alibaba Cloud is the only major hyperscaler with mainland China availability, though US and EU enterprises with regulated workloads typically avoid it on jurisdiction grounds.

The fragmentation extends beyond just Blackwell. The survey tracks 71 cloud GPU providers offering more than 2,500 distinct instance configurations across 14 GPU model families, including NVIDIA chips, AMD MI300X and MI355X accelerators, and Intel Gaudi processors. This diversity means no single provider dominates the market the way NVIDIA dominates the chip itself.

Hyperscalers maintain their position through bundled services. AWS remains the largest, spanning 15 GPU model families with H100 through its p5 instance family. Azure covers 10 GPU families including B200 and AMD MI300X. GCP posts the cheapest H100 among hyperscalers at roughly $11 to $14 per GPU per hour, depending on the specific A3 variant. OCI differentiates through bare-metal-first infrastructure and RoCE v2 cluster networking, appealing to teams running tightly coupled multi-node training workloads.

Neoclouds are capturing demand from price-sensitive research teams and startups. Lambda Labs claims to serve over 10,000 research teams with a GPU-only catalog and pre-loaded software stacks including PyTorch, TensorFlow, and CUDA drivers. RunPod operates two tiers: Secure Cloud for dedicated bare-metal and Community Cloud for shared bare-metal at lower rates without service-level agreements. RunPod's instance startup time is sub-minute, the fastest measured in the survey. Crusoe runs data centers on stranded and flared natural gas, funding aggressive H100 and B200 capacity buildout through cost and emissions arbitrage.

The market's fragmentation suggests that Blackwell availability will gradually expand as production ramps, but the initial scarcity is forcing teams to make strategic choices. Those needing immediate access to cutting-edge hardware face a choice between paying hyperscaler premiums for broader compliance and integration, or betting on specialist providers with lower costs but narrower service ecosystems. For teams still using H100 or older chips, the mature market offers abundant options and competitive pricing across all provider tiers.