01GPU clouds68
02Object storage32
03Application platforms28
04Virtual private servers28
05Managed databases25
06Edge and CDN20
07Managed Kubernetes17
08Bare metal and colocation13
09Cloud platform11
10Serverless functions10
11Hyperscalers9
12Serverless GPU6
13Marketplaces and aggregators5
14Neocloud / GPU cloud software1
GPU cloudsOPEN68
GPU capacity rented by the hour or the second, on machines the provider operates. Virtualised unless the listing says otherwise.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
Object storageOPEN32
S3-compatible buckets. Where weights, datasets and artefacts sit when they are not in a database.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
Virtual private serversOPEN28
A machine you administer yourself, billed hourly or monthly. The oldest and still the most common way to host something.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
Managed databasesOPEN25
Postgres, MySQL and Redis operated by the provider. General application data, not retrieval -- vector stores live in Data / RAG.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
Edge and CDNOPEN20
Static hosting and content delivery at the edge, where a model response is cached or a front end is served from.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
Managed KubernetesOPEN17
A Kubernetes control plane someone else operates, with nodes you size.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
Serverless functionsTHIN10
Per-invocation compute with no server to keep alive. Includes edge runtimes.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
SHOWING 10 OF 10 · ALL OF THEM
HyperscalersTHIN9
Amazon Web Services, Microsoft Azure and Google Cloud. The line is drawn at these three deliberately: every wider reading of the word is an argument rather than a test. A provider that runs its own general-purpose public cloud at any other scale belongs in cloud-platform.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
SHOWING 9 OF 9 · ALL OF THEM
Serverless GPUTHIN6
GPU compute billed per invocation, with nothing to keep alive between calls.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
SHOWING 6 OF 6 · ALL OF THEM
Marketplaces and aggregatorsTHIN5
Marketplaces and aggregators reselling other providers' capacity.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
SHOWING 5 OF 5 · ALL OF THEM
Neocloud / GPU cloud softwareTHIN1
Software licensed to a service provider so THEY can turn bare GPUs into a billed, multi-tenant cloud -- pooling, metering, billing, white-label UI. Distinct from GPU clouds (7): members here rent to nobody and are never a customer's point of sale. Run:ai, Rafay and Spectro Cloud are the same shape and not yet in this catalogue.
Ordered by attributes answered — the column we hold for most of them. Blank where we hold none.
SHOWING 1 OF 1 · ALL OF THEM