NVIDIA B200 cloud pricing

Each provider's cheapest on-demand B200 listing, per GPU-hour, cheapest first. Where the cheapest listing is a multi-GPU machine, the price is its hourly price divided by its GPU count.

The lowest B200 list price at each catalog refresh. It hasn't moved since 23 Sep 2026, 14:37 UTC.

For now we keep about a day of history, recorded at each 30-minute catalog refresh.

The NVIDIA B200 is an NVIDIA Blackwell GPU with 180 GB HBM3e of memory. People typically rent it for large-scale training and inference on the newest NVIDIA generation.

• The cheapest B200 costs 5.2× the cheapest H100 ($1.29/GPU-hr) right now.
• One step down in price, the H200 starts at $2.60/GPU-hr (141 GB HBM3e).

Prices come from the public SkyPilot catalog (Apache-2.0), which lists each neocloud's on-demand list prices. We read it every 30 minutes and divide each instance's hourly price by its GPU count, so every figure is per GPU-hour. For each provider we show its cheapest listing; spot and peer-to-peer marketplace listings aren't included.

A list price is not a guarantee: the provider may have no free capacity at that price, and what you can actually launch changes minute to minute. MatMul launches from live supplier inventory and charges the supplier's price plus a markup, so the price you pay on MatMul is not the market floor shown here. The console shows the exact hourly price before you launch.

How much does a B200 cost per hour?

As of 23 Sep 2026, 19:30 UTC, on-demand list prices for the NVIDIA B200 run from $6.69 to $7.15 per GPU-hour across the 3 neoclouds we track, with a median of $6.79. These are list prices from the SkyPilot catalog, not quotes.

Which cloud has the cheapest B200 right now?

Lambda, at $6.69 per GPU-hour (instance gpu_8x_b200_sxm6, europe-central-1). That price comes from a 8-GPU machine, so renting a single B200 can cost more. List prices don't guarantee stock: the provider may have none free at that price.

Can I launch a B200 at these prices on MatMul?

Not necessarily. MatMul launches from live supplier inventory and charges the supplier's price plus a markup, so its price is not the market floor shown here. The console shows the exact hourly price of every GPU you can launch right now, before you launch.

Where do these prices come from?

From the public SkyPilot catalog (Apache-2.0), which lists each neocloud's on-demand instance prices. MatMul reads it every 30 minutes and divides each instance price by its GPU count to get a per-GPU-hour price. Spot and marketplace listings aren't included.

• H100 price per hourfrom $1.29
• H200 price per hourfrom $2.60
• A100 80GB price per hourfrom $0.99
• L40S price per hourfrom $0.85
• A6000 price per hourfrom $0.50

Or see every GPU we track.