Skip to content
For Agencies

A machine for each client. Run by us, billed by you.

Give each client workload a dedicated home. Configure its key scope; you decide ownership and rebilling.

Book a callSee the machine

Your account or organization

  • Project AMachine-scoped key
  • Project BMachine-scoped key
  • Project CMachine-scoped key

Your account or organization

Project AMachine A

Machine-scoped keys limit access to the intended owned machine. Project labels don't create access boundaries.

Project labels illustrate your setup, not separate client accounts. Create a key scoped to the intended owned machine. An unscoped key can reach all machines owned by its account or organization.

Organization keys also require current membership and inference permission. A project name alone does not create an access boundary.

How it runs

You keep the client. We run the rack.

  1. Drop-in endpoint.

    Each machine serves an OpenAI-compatible API, so any OpenAI client works. Your client never sees a cloud console.

  2. Keys you hand out.

    Create a key scoped to the intended owned machine. A key without a machine scope can reach all machines owned by its account or organization. Project labels do not create separate client accounts.

  3. We run the building.

    Power, cooling and monitoring in Oregon, and each machine insured at replacement value while it's in our rack. Our SOC 2 Type II audit is underway with A-LIGN.

One line item per client.

Configure 4 to 8 cards for your client's workload. See the parts and starting price before you choose a build.

RTX PRO 6000 Server Edition

VRAM
96 GB GDDR7
per card
Bandwidth
1.6 TB/s
per card
Lead time
6–8 weeks

Starting at $99,971

Book a callConfigureEvery machine, from $7,880

Deployment planning

Plan your model deployment.

These catalog profiles can be sized to this machine's GPU memory and system RAM. Check the required configuration in each guide; compatibility and performance still need testing.

Llama 3.3 70B

ollama · 80 GB catalog GPU-memory threshold · 4-GPU planning configuration

gpt-oss 120B

ollama · 80 GB catalog GPU-memory threshold · 4-GPU planning configuration

Qwen3 32B

ollama · 32 GB catalog GPU-memory threshold · 4-GPU planning configuration

Qwen3 Coder 30B

ollama · 32 GB catalog GPU-memory threshold · 4-GPU planning configuration

Compare the alternatives.

B3IQ vs Hydra Host

Compare owned B3IQ machines with Hydra Host bare metal and AI Factory offers: procurement, control, costs and operating responsibilities.

B3IQ vs Vast.ai

Decide between buying a B3IQ GPU machine and renting Vast.ai marketplace capacity, with configuration, persistence and cost assumptions made explicit.

Browse model guides

Reviewed model identities, runtime profiles and source evidence.

Find hardware by memory

Estimate the memory budget for the workload you want to run.

Compare providers and GPUs

Compare purchasing models, operating responsibilities and hardware.

How we review model evidence

Catalog settings, memory estimates and measured evidence have different limits.

Questions agencies ask.

Either works. You own the machine and rebill hosting, or your client buys it outright and you operate it. We'll structure it on the call.

If you own the machine, it stays yours. Point it at the next client's workload, or sell it.

Use your account or organization dashboard for the machines it owns. Configure keys with explicit machine scopes for the intended workloads. Organization keys also require current membership and inference permission. Client/project names are your organizational labels, not automatic access boundaries.

Daryl XuViktoriya HyingSean GengMatt

Start with one client.

We size the first client machine and settle ownership, hosting and rebilling.

Book a callExplore machines

Or message us: iMessage +1 (332) 324-8327Telegram @b3iqbot

Book a call