Which cloud is best for an AI product?

Training or inference: which are we buying for?#

Buy for how the product runs, not how the model was built. Training teaches a model, and it's a project. Inference runs the finished model for your users, and it's the product, every day it's live.

The two want different things. Training wants a block of GPUs, well linked, for a stretch. Inference wants steady GPUs close to your users, next to the rest of what the product runs on.

Ask whether you're renting GPUs or renting a model. Some clouds sell access to ready-made models, so you never touch a GPU. Others rent you the GPUs to run your own.

The product lives next to its data and its users, not only next to GPUs. The app, database and sign-in need a home too. Pick for the GPUs alone and the rest has nowhere to live.

Here's who plays. AWS is a public cloud with GPU servers and AI chips of its own. Akamai sells cloud GPUs, with inference close to users.

Megaport rents bare-metal GPU servers and links clouds together privately. That's a sample of the cloud providers we work with, and Clectiv AI runs the process for you: start a project.

Data, users and rules narrow the list. GPU access picks from what's left. That part changes with your ask, so it belongs in the plan.

Before you call anyone#

What the product does, where its data lives and who uses it go into Clectiv AI before you pick a cloud. After that, Clectiv AI runs the process, bids it out to the providers that fit, has them compete, and lets you choose. Pick one first and you've chosen your platform from a single sales pitch.

Start a project

Does our data pick the cloud for us?#

Often it does, because big data is slow and costly to move. Compute tends to go where the data already is.

Four questions to answer first:

• Where does the data live today? Which cloud, which region, or your own data center.
• How big is it, and what does moving it cost? A little data moves freely. A lot doesn't, and data leaving a cloud is billed.
• Can you link privately? A private link connects a data center to a cloud directly, and Megaport sells them.
• Do the GPUs sit near the data? A fast model on a distant copy of your data isn't fast.

Those are some of our cloud providers for this, and Clectiv AI runs the process with them for you: start a project.

You don't start from scratch. Clectiv AI runs the process we ran for companies in your position, so you start where they left off. It carries the plan straight to these providers, after the best pricing, terms and contract we can get.

Map it with Clectiv AI before you choose. Your data and your GPUs may not sit in the same place, and the plan says whether that's fine.

Moving the data is a choice, not a default. Sometimes the GPUs are worth moving data toward, and sometimes the GPUs come to the data. Price both ways.

What if the data has to stay in the country?#

Then the cloud needs a place in that country, and the rules go in the plan. This is data residency, or sovereignty: your data stays in one country, under its laws. A region is the cloud's set of data centers in one area.

The rule can come from a customer's contract, not only from a law. Ask your buyers what they've required.

• Is there a region there, and are there GPUs in it?
• Who runs it, and who can reach your data?
• Do backups and support leave the country?
• Can you show it? Your buyers or regulators may ask where the data sits.

AWS runs regions around the world, and Rackspace sells private cloud, run on infrastructure that's yours alone. All of them are cloud providers we work with, and Clectiv AI runs the process for you: start a project.

One hint: get the answers in writing. A pin on a map isn't a promise about where your data sits. The plan asks the rest.

Common questions#

How does Clectiv AI get this done?#

Put it in Clectiv AI. Tell it what your product does, where its data lives and what rules it runs under. It scopes your AI platform with you and writes your plan: training or inference, the data, the region and the GPUs.

Then the Clectiv team takes your plan to the cloud and GPU providers that fit, all at once, through our partner network of 620+ providers. Each one bids, they compete on the same plan, the team pushes on terms, and you choose.

You inherit what others learned. Clectiv AI runs the process that got them there, so you start where they left off. It takes the plan straight to the providers that fit, for the best pricing, terms and contract we can get.

It's free to you, and it gets moving in days, not months. The rest of what we source in cloud is on the cloud page.

If GPU capacity is the open question, read how to compare GPU cloud providers.