Skip to content
For Startups

Own your inference. Make the bill predictable.

A metered API prices every token. A machine you own prices the month.

Book a callSee the machine
A Python request showing the B3IQ endpoint, API key and served model, centered over a sunlit landscape.
app.py3 line change
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://b3iq.org/v1/api",
    api_key=os.environ["B3IQ_API_KEY"],
)

reply = client.chat.completions.create(
    model="qwen3.8:27b-bf16",
    messages=[
        {"role": "user", "content": "Hi"}
    ],
)

Replace the example model with an ID from client.models.list(). It must be served by your machines and available to this key.

Quickstart

Your key, your hardware.

  1. Your applicationYour app | OpenAI SDK
  2. B3IQ gatewayKey and machine scope checked
  3. Your machineRun on your machine
Response returned to your app

Same code at any size.

Add machines. The endpoint doesn't change.

Your account, your machines.

Keys route only to machines owned by their account or organization. Narrow a key to the machines that should serve that app.

Sized to your model.

On the call, we size the build to the models you run: VRAM, throughput and card count.

What do you run today?

What you pay for

Every token

With B3IQ: The machine, plus Managed Hosting monthly

Where it runs

Their servers

With B3IQ: Your machine in Oregon

Who else is on it

Every other customer

With B3IQ: Nobody

Who fixes it at 3 a.m.

Their on-call team

With B3IQ: Our engineers on site, for the hardware

The whole machine.

8 cards and 1,128 GB of GPU memory in one chassis.

H200 NVL

VRAM
141 GB HBM3e
per card
Bandwidth
4.8 TB/s
per card
Lead time
6–8 weeks

Starting at $329,032

Book a callFull specsEvery 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 · 8-GPU planning configuration

gpt-oss 120B

ollama · 80 GB catalog GPU-memory threshold · 8-GPU planning configuration

Qwen3 32B

ollama · 32 GB catalog GPU-memory threshold · 8-GPU planning configuration

Qwen3 Coder 30B

ollama · 32 GB catalog GPU-memory threshold · 8-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.

Before you switch.

If spend is small and stable, stay on a metered API, and we'll tell you so. Owning makes sense when usage is sustained, spiky, or growing fast enough that a predictable month matters more than the next token's price.

Your customers' requests run only on machines you own. Our SOC 2 Type II audit is underway with A-LIGN, and b3iq.org/security lists our controls.

Yes. The machine is paid for in full, and title transfers to you on acceptance, when your machine is racked, tested and online. There is no financing and no balance.

Daryl XuViktoriya HyingSean GengMatt

Bring your token bill.

We size a machine against your real usage and tell you whether owning beats the API yet.

Book a callExplore machines

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

Book a call