Cerebrium

Serverless GPUs for real-time AI inference

Founded in Cape Town in 2021 and now headquartered in New York, Cerebrium runs serverless GPU infrastructure for real-time AI workloads: voice agents, image and video pipelines, LLM endpoints, and custom Python apps. Hardware is declared in a cerebrium.toml file and the code deploys as a container that scales from zero, with compute billed by the second.

Example customers include Tavus, Deepgram, Resemble AI. More here.

Cerebrium Homepage

What's good about Cerebrium

Per-second billing, with no charge for idle GPUs
Cold starts of 2-4 seconds via memory and GPU snapshots
SOC 2 Type II, HIPAA, GDPR and ISO 27001 certified
Ten GPU types, from T4 to B200
Region pinning for data residency, on US and EU capacity

Cerebrium pricing examples

Cerebrium bills compute by the second while a container is running, with the GPU, vCPU and memory each priced separately, so a deployed app costs more than its GPU rate alone.

Platform plans sit on top of compute: a free tier, a flat monthly subscription, and custom enterprise pricing with volume discounts. Persistent storage is billed per GB-month.

Below are some example configurations and their estimated costs:

Example configuration	Estimated cost
VM Small	$22.73 / mo 1 vCPU, 1 GB RAM (CPU only)
VM Medium	$113.94 / mo 4 vCPU, 8 GB RAM (CPU only)
VM Large	$227.89 / mo 8 vCPU, 16 GB RAM (CPU only)
Block Storage	$5.00 / mo 100 GB beyond free allowance · First 100 GB included

Cerebrium GPUs

Here are some of the GPU configurations offered by Cerebrium:

9 GPU models

How this list works

Order

By GPU model: one card per GPU, the most looked-up models first, then the most widely offered ones.
By configuration: every configuration we track. What you can rent now comes first, and within that one row per GPU model before any repeats, so the opening rows cover the range rather than one model's sizes.

The price on a card

Each card leads with a per GPU per hour rate. We prefer configurations the provider reports as in stock, on-demand rates before reserved or spot, then take the lowest per-GPU rate among those.

Search

We match the GPU model, form factor, billing type, availability and instance name. Matching is partial and case-insensitive. Everyday words work too, so "interruptible" finds spot and "sold out" finds out of stock.

Transparency and funding

Ads and sponsors: Paid placements sit at the top of the list and are always labeled as sponsored content. Sponsorship never influences the organic ranking itself.
Affiliates: Affiliate links are marked. We may earn a commission if you click them, but commissions never affect the order.
Prices: Shown in USD, converted at daily reference rates where a provider publishes in another currency. A month means 720 hours. How we estimate costs.

Nvidia H100

4 configs 1x-8x checked <24h ago

On-Demand from $3.40

From $3.40 / GPU / hr On-Demand Visit website


1x H100 HOPPER_H100	80GB	--	--	On-Demand	$3.40	$3.40	Not reported
2x H100 HOPPER_H100	160GB	--	--	On-Demand	$3.40	$6.80	Not reported
4x H100 HOPPER_H100	320GB	--	--	On-Demand	$3.40	$13.59	Not reported
8x H100 HOPPER_H100	640GB	--	--	On-Demand	$3.40	$27.19	Not reported