QIT: Liquid-Cooled GPU Containers, Model API and Reserved Compute

Provider profile · Provider-submitted listing · Information checked: November 9, 2026 · Reviewed by the editorial team

QIT builds modular AI infrastructure around available power: factory-integrated, liquid-cooled GPU containers it says can be installed on a prepared site in 15 hours — plus a model API, serverless GPU pods and reserved hardware. Its one named reference project is, in its own words, a preliminary design. This profile separates what QIT publishes from what it has yet to show.

QIT at a glance

Four things to know before you buy

None of the following is hidden — each point is stated in QIT's own pages or follows from what they do and don't show. They matter more than the headline figures when you compare QIT with the providers in our directory:

1. The headline figures are vendor claims. A 20 ft container delivering 1 MW and 1,200+ GPUs, installed in 15 hours, is QIT's published specification. We have not verified a completed installation, and the 15 hours explicitly starts only when a prepared site with power, pad and connectivity is ready.
2. The Texas reference project is a design study, not an operating site. QIT Pod 001 is presented on QIT's homepage as "our first reference project", and the same page states, in QIT's own words, that it is a "preliminary design, issued for study". Read it as a design, not as a delivered, running deployment (details below).
3. Some application claims don't appear on the public pages we checked. QIT's listing application mentions, for example, $1.56 in free credits for new business emails. We could not find that offer on qit.energy's homepage or pricing page on November 9, 2026, so treat it as unverified until you see it on the site.
4. QIT is independent — by its own statement. Its FAQ says QIT is not an NVIDIA DSX POD or Crusoe partner and claims no certification or validated design from either. Its comparison pages (for example against Crusoe Spark) are QIT's own marketing, not third-party endorsements.

The container offer and deployment process

QIT's starting point is a site with power already available — stranded or underused electrical capacity. Around that capacity it develops modular AI data centers: liquid-cooled GPU containers, networking and supporting infrastructure in one coordinated deployment. Containers are integrated and tested in the factory, then shipped to the prepared site and, QIT says, set, connected and commissioned in 15 hours, with no construction project on site. The published deployment sequence is:

1. Scope & site — assess available power and connectivity; agree the workload, configuration, interfaces and responsibilities.
2. Factory build — integrate and test the container in the factory while the site team prepares power, cooling and connectivity.
3. Deliver — ship the tested container to the prepared site.
4. Install & go live — set, connect and commission on site, in the published 15 hours.

What the site must provide: a prepared location, appropriate power and network connectivity, cooling interfaces and delivery access, with the split of responsibilities agreed during planning. GPU platform, rack count, fabric and storage are defined around each workload. For buyers comparing this model with conventional builds, our inference serving guide and GPU cost calculator cover the demand side of the same decision.

QIT Pod 001, the Texas reference project

QIT's homepage presents Pod 001 as its first reference project: 1 MW on an energized Texas site — a liquid-cooled 40 ft hosting container for customer-owned racks, placed on an operating bitcoin-mining site and fed from power already on the property. Published specifications:

That last row is the one to remember: as of November 9, 2026, QIT's own site describes its flagship reference as a design issued for study. That doesn't make the engineering uninteresting — the spec is concrete — but a buyer should ask QIT what has actually been built and operated, and where, before treating Pod 001 as a track record.

Model API, serverless pods and reserved GPUs

Alongside the containers, QIT publishes a model API catalogue on its pricing page: several hundred priced entries spanning text, image, video, audio, 3D and LoRA training, billed per call — per image, per second of video, per million tokens or per request, depending on the model. The same page covers serverless GPU pods and reserved compute; QIT's application lists reserved and bare-metal GPUs including L40S, RTX PRO 6000, H100, H200, B200 and B300. Per-call example prices change with promotions and parameters, so we don't reproduce the table here — read the live list on qit.energy/pricing, and note the page's own warning that exact cost depends on your parameters.

Who runs QIT

The pages we checked (homepage, pricing, legal index) do not publish an operating company name, registration, headquarters address or team. Contact is hello@qit.energy, and the site's FAQ routes project proposals — a brief with location, available power, workload and desired deployment date — through its contact channel. For an infrastructure purchase at container scale, a buyer should expect to establish the contracting entity, references from completed deployments, and SLA terms during that first conversation; none of that is published on the site today.

Comparing providers? Return to the GPU cloud provider directory for the official pricing pages of all listed clouds, read how GPU cloud pricing works, and sanity-check any quote in our GPU cost calculator.

Related guides

• GPU Cloud Provider Directory: Where to Check Real Prices
• GPU Cloud Pricing Explained
• Choosing a GPU for Inference Serving