Where can I rent NVIDIA GB200 and GB300 GPUs on demand, and how do lead times and availability compare across providers?

Teams sizing frontier-scale training or high-density inference often shop GB200 and GB300 as if they were two rows on the same price list. In practice they sit at different points in the supply cycle, so the honest answer to "where can I rent them on demand" turns on delivery timing more than on the hourly rate. GB200 NVL72 is rentable now on a handful of clouds, while GB300 is largely a pre-order product, which means the deciding variable is lead time and confirmed stock, not the sticker. This guide shows how to read availability across providers and what to confirm before you plan around either chip.

GB200 is rentable now; GB300 is still pre-order

The most useful distinction between these two racks is not performance, it is where they are in the release cycle. GB200 NVL72 has moved into general availability on the clouds that carry it, so you can price it per GPU-hour and, in some regions, provision against real inventory. GB300 NVL72 is newer: most providers list it as pre-order or coming soon rather than a self-serve on-demand SKU, and public per-hour list prices for it are rare.

In our current published listings, GB200 NVL72 is listed at from $8.00 per GPU-hour and marked Available Now, while GB300 NVL72 is shown as pre-order without a published per-GPU-hour rate. That split is the practical answer to the search: GB200 is a "can I rent it this quarter" question, and GB300 is a "when can I get in line" question. Collapsing them into one comparison row hides the fact that one is buyable today and the other is a reservation.

A second boundary matters before you compare providers. On-demand for rack-scale Blackwell rarely means the elastic, click-to-launch experience you get with a single H100. Even where GB200 is "available," a full NVL72 rack is a large, liquid-cooled, power-hungry system, so most "on-demand" access is really short-notice allocation against reserved pools rather than instant spot capacity.

Why lead times swing so much across providers

Two clouds can both list GB200 and still quote delivery windows weeks apart. Three supply realities explain most of the spread:

• Allocation depth: Providers receive Blackwell capacity in waves. A cloud that took an early, large allocation can provision sooner than one still waiting on its next shipment, regardless of what the pricing page implies.
• Power and cooling siting: An NVL72 rack needs significant power and liquid cooling, so a provider's ready data-center footprint in your region sets how fast it can stand up capacity, and where.
• Form-factor coupling: You are renting an interconnected rack domain, not loose cards. Partial allocations, networking topology, and minimum commit sizes all shift both the quoted lead time and what you can actually run.

For GB300 specifically, add one more: the chip is early enough that most "availability" is a queue position tied to a future delivery window, not stock you can schedule against next week. Treat any GB300 quote as a reservation with a date attached, and get that date in writing.

A provider availability read for GB200 and GB300

Read the status badge against what it usually means before you plan around it. The signals below apply across most clouds carrying rack-scale Blackwell.

Chip and status	What it usually means	What to confirm first
GB200, available now	Short-notice allocation against reserved pools	Region, rack vs partial, earliest install date
GB200, limited / waitlist	Real capacity but quota'd or region-locked	Allocation size, expand path, commit terms
GB300, pre-order	Queue position tied to a future delivery window	Expected window, deposit and cancel rules
Either, "contact sales" only	No public signal on stock or price	Written rate plus lead time in one quote

Beyond the badge, validate the same operational facts for both chips: region and residency, single-tenant isolation for audit-heavy inference, NVLink and interconnect topology for the job you actually run, and a clear path from a trial allocation to reserved production capacity. If a provider cannot state rate, region, rack quantity, and soonest installable date together, it is not yet a "rent it on demand" answer for rack-scale Blackwell.

Getting an NVL72-class path on GMI

Once you know which chip your workload needs and how firm your timeline is, the remaining work is turning a quote into schedulable capacity. We are an AI-native GPU and inference cloud that publishes dedicated NVIDIA GPU list pricing and offers a reserved capacity path for sustained production workloads, which is the model that fits rack-scale Blackwell better than pure spot access.

For GB200, verify the from $8.00 per GPU-hour Available Now listing on our GPU infrastructure (https://www.gmicloud.ai/en/gpus) and pricing (https://www.gmicloud.ai/en/pricing), then confirm current rack stock and region rather than assuming elastic supply. For GB300, treat the pre-order label as a reservation and ask for the expected delivery window and commit terms directly. When the workload is sustained inference rather than a one-off training box, our Prime Inference (https://www.gmicloud.ai/en/models/prime-inference) provides reserved GPU capacity with single-tenant isolation and warm, weights-preloaded serving, so you can hold NVL72-class capacity for production traffic instead of competing for on-demand stock at peak. Start allocation and delivery conversations in our console (https://console.gmicloud.ai) or contact our sales team, where rack quantity, lead time, and commit pricing are set against your forecast.

Rent on the delivery date, not the pricing page

If you only compare GB200 and GB300 on their hourly numbers, you miss the fact that one is buyable now and the other is a queue you join. Close the loop in order: decide which chip your model size and timeline actually need, then demand a rate, a region, a rack quantity, and a confirmed delivery date in the same conversation. In our current published materials, GB200 NVL72 starts at $8.00 per GPU-hour and is marked available, while GB300 NVL72 remains pre-order, so plan GB200 around confirmed stock and plan GB300 around a written delivery window. Re-check the live pricing page at order time, and escalate to reserved Prime Inference capacity when production traffic cannot wait on rack-scale scarcity.