AI inference you can locate, trust and verify.
Starfield runs open models on AI factories built close to power and close to you. Every token is metered and recorded with where it ran.
STARFIELD IN 30 SECONDSWATCH ON YOUTUBE →
Inference that runs where you need it, with proof that it did.
Energy-first siting
We site AI factories where energy is already available, so capacity comes online faster than a new build.
Local by design
Inference runs near your users and your data, in the region or on the premises you choose.
Proof of origin on every token
Fabric records the site, hardware, model and policy behind every request.
One cloud, from the power line to the API.
The control plane that turns many sites into one cloud.
Fabric joins independently owned sites and mixed hardware into one network, places each request by your policy, then meters and records it.
Explore Fabric →The gateway to open-source AI.
A conversational interface and coding harness for your teams, and one unified API for your developers. Run, integrate and deploy open models across apps, products and workflows.
Explore Command →Standardized AI factories, from 200 kW to 150 MW.
Edge AI factories for on-prem and edge sites, and AI factory sites at grid scale. Built and operated by Starfield on partner power.
Explore AI Factories →Many sites, many owners. One cloud through Fabric.
AI capacity should not depend on a few campuses and a few owners. Starfield sites are owned and powered by different partners in different places. Fabric is what makes them work as one cloud.
Many owners, one network
Partners bring power and sites; Fabric joins them into one pool of capacity, so no single owner controls supply.
No single point of failure
If one site goes offline, Fabric is designed to re-place work on another site that meets your policy.
Mixed hardware, one endpoint
Different accelerators and inference engines sit behind the same API.
You set the rules
Choose where your workloads may run by country, latency, model and site, and get a record of where they did.
One gateway to open models. Two ways in.
Command is the gateway to open-source AI and inference. It pairs a conversational interface for your teams with a unified API for your developers, running on Starfield's inference infrastructure.
For your teams: chat and agentic coding
An intuitive conversational interface and a coding harness on open models, the workflow your engineers already know.
For your developers: one unified API
An OpenAI-compatible API to integrate open models into apps, products and automated workflows.
Priced per token, not per seat
One price book for chat, coding and API, per 1M tokens.
Coding is one of the most token-hungry things a company does with AI. Open models change the economics.
- Lower cost per tokenOpen models are typically far cheaper to run than closed frontier models, and the gap matters most at coding volumes.
- Pay for use, not seatsCost follows the tokens your engineers actually use instead of a fixed licence per person.
- Your code stays yoursRun the model in your region or on your premises, so source code never leaves the boundary you set.
- No lock-inSwitch to a better model as soon as one ships, or tune one on your own codebase.
- InspectableOpen weights can be evaluated, audited and pinned to a version, which compliance teams can sign off.
Built for the buyers the gigawatt clouds overlook.
Sovereign AI, with proof.
Keep models, data and inference in-country or on-prem, with an audit trail for every request. For energy, defense, healthcare and public sector.
Talk to us →Ship on open models.
Use the Command API to run leading open models close to your users, and its coding harness for your engineers. Start serverless, move to dedicated as you grow.
Get Command early access →Reserve capacity on contract.
Contracted token volume or GPU capacity, priced per 1M tokens or per GPU-hour, with verified delivery reports.
Reserve capacity →Turn your power into an AI factory.
Bring energized power and a site. We build, operate and sell the output, so you earn from AI without building a cloud.
Partner your power →Built with builders.
Starfield Labs is our builder community: hackathons, build days and open research.
Build with your agent
Weekend and week-long events, online and in cities, where people and their AI agents build side by side.
Private build days
Hackathons and workshops for your team or your developer ecosystem, run by Labs, with Command and open models built in.
Partner with Labs
Put your tools, credits, prizes or people in front of builders.
Model Watch
Dated model prices and repeatable speed runs on open models, with a public method and corrections log.
From energized site to verified token.
- 01PowerPartners bring energized sites and power.
- 02BuildStarfield deploys standardized AI factories, from 200 kW to 150 MW.
- 03ConnectFabric joins every site into one network and places each request by residency, latency, model and load.
- 04VerifyEvery token is metered and recorded with its origin.
A different shape of AI cloud.
| Gigawatt AI clouds | Token APIs | Starfield | |
|---|---|---|---|
| Built for | Frontier labs and hyperscalers | AI-native apps | Sovereign, regulated and edge buyers; AI builders |
| What you buy | GPU clusters, per GPU-hour | Tokens, per 1M | Tokens and reserved capacity |
| Footprint | A few gigawatt campuses | Rented capacity | Many sites near power and demand |
| Who controls capacity | A few campuses and owners | The provider decides where it runs | Decentralized: many sites and owners, your policy decides |
| Builder community | Enterprise sales | Developer programs | Starfield Labs hackathons and build days |
| Proof of where it ran | No | No | Yes, per request |
Inference that stays where you need it.
Tell us your model, region and volume. We'll come back with capacity and a price.
or contact us at hello@starfield.net