Get in touch

AI data storage in three US regions

Store the data your AI work reads and writes: training sets, checkpoints, model weights, document stores for retrieval, logs and outputs. It's S3-compatible object storage in Ashburn, Chicago and Los Angeles, close to users and compute on both coasts and in the middle of the country. Run your pipeline on your own dedicated servers or VPS in the same building, and read the same data again and again: reads inside your plan cost nothing extra.

Get an AI storage quote How our S3 storage works
3 regions East, Central and West Coast
7 ms New York to our Ashburn site
14 ms Toronto to our Chicago site
<1 ms Across Los Angeles to our LA site

Why store AI data with us

AI work reads the same data many times. Storage should not charge you each time it does.

Get an AI storage quote

Read it as often as you need

Training runs, evaluations and inference servers pull the same files over and over. Public clouds bill every read out: Amazon S3 lists $0.09 per GB for the first 10 TB a month sent to the internet from US East (AWS price list, checked October 15, 2026). With us, reads inside your plan cost nothing extra.

Three regions, one provider

Ashburn on the East Coast, Chicago in the middle, Los Angeles on the West Coast. Put data near your users, your team or your compute.

Near the clouds and the people

Ashburn is in the same metro as AWS us-east-1. Chicago reaches New York in about 19 ms, and Los Angeles is our fastest site for Asia-Pacific users, with Tokyo at about 110 ms.

Your data in a building you can name

Datasets, customer documents and model weights sit in a US data center you can name, with buckets and keys that are yours alone. Need drives no other customer uses? We'll build a dedicated storage server. We're SOC 2 Type II certified.

Keep the exact dataset a model saw

Bucket versioning keeps every version of a file, so you can go back to the data a model was trained or evaluated on. Object lock stops anyone deleting it before its retention ends.

Servers for the work around the model

Cleaning, deduplication, tokenizing, small embedding models, vector databases and ETL run fine on CPU. Put them on our dedicated servers or VPS next to your storage. We don't run a GPU cloud; keep your GPUs where they are, or ask about a custom GPU build.

Workloads

What AI teams store with us

  • Training data: raw and cleaned datasets, often tens or hundreds of terabytes, read in full on every epoch.
  • Checkpoints: large files written during long runs and read back when a job restarts or resumes.
  • Model weights: the files your inference servers load each time they start or scale up.
  • Retrieval data: the documents, chunks and embeddings behind an assistant that answers from your own files.
  • Logs, evaluations and outputs: prompts, responses and scores you keep to measure and improve a model.
  • Backups of all of it: with object lock on, so a bad script or ransomware can’t delete them.

All of it sits behind the S3 API, so code that reads from S3 today, using boto3, rclone or a data loader built on them, points at our endpoint with new keys.

Latency

Pick the region closest to your work

Measured from users on local internet providers to each site, using RIPE Atlas probes (median round trip).

RegionClose toMeasured round trip
Ashburn (East), Equinix DC3New York, the Northeast, AWS us-east-1, EuropeNew York about 7 ms; London about 80 ms
Chicago (Central), Arlington HeightsThe Midwest, Toronto, most of the US in one hopToronto about 14 ms; New York about 19 ms; Los Angeles about 48 ms
Los Angeles (West), Digital Realty LAX10California, the Pacific Northwest, Asia-PacificAcross LA under 1 ms; Seattle about 28 ms; Tokyo about 110 ms

The full table covers 29 cities: see our network latency page.

An example

What a quote looks like

Say your team keeps 100 TB of training data, checkpoints and model weights in Chicago. Two dedicated servers in the same building clean and tokenize new data, and your GPUs elsewhere pull about 20 TB of it a month.

  • Storage: 100 TB on our shared cluster, with room to grow.
  • Servers: the two dedicated servers for the CPU work, in the same building as the storage.
  • Bandwidth: reads inside your plan cost nothing extra. Past a metered allowance it’s $0.02 per GB.

You get one monthly price for all of it. On Amazon S3, sending those 20 TB a month to GPUs outside AWS would cost about $1,800 at its list price (US East, checked October 15, 2026), before storage.

Get an AI storage quote

Tell us how much you'll store, how much your jobs read back each month and where your compute runs. It goes straight to our sales team, and we'll reply within 1 business day with a configuration and a monthly price.

Prefer to talk first?

    1About you

    2Your storage

    Which setup? *

    3Anything else

    By submitting this form, you agree to GigeNET's Privacy Policy. We'll use your information to respond to your quote request.

    AI data storage FAQs

    Related: Egress fees explained · S3-compatible object storage · Dedicated storage servers · Network latency · Dedicated servers

    Do you have GPUs?

    We don’t run a GPU cloud. Most teams keep their GPUs where they already are and store the data with us, or run the CPU parts of the pipeline on our dedicated servers. If you need a GPU server, ask about a custom build.

    How much does it cost?

    Shared storage starts at 10 TB, and it’s quoted per deployment, because the price depends on how much you store, how much you read back each month, the region and whether you want shared storage or a server of your own. Fill in the quote form and we’ll price it.

    Is the storage shared?

    The storage cluster is shared, and your buckets and access keys are yours alone. The servers that read it are your own dedicated servers or VPS. If you need drives no other customer uses, we’ll build a dedicated storage server or cluster.

    Will my AI tools work with it?

    If the tool can read from Amazon S3, it can usually read from ours once you change the endpoint and keys. That covers boto3, rclone and data loaders built on them.

    How fast can my jobs read the data?

    It depends on distance, your server’s network port and how many other jobs read at the same time. Within one region, network round trips are a few milliseconds. Tell us the read rate you need on the quote form and we’ll tell you what we can do.

    Where is my data stored?

    In the region you pick: Ashburn, Virginia (Equinix DC3), Chicago (our Arlington Heights facility) or Los Angeles (Digital Realty LAX10). It stays there.

    Can I keep old versions of a dataset?

    Yes. Bucket versioning keeps earlier versions of each file, and object lock can stop them being deleted before a retention date.