# B3IQ: full guide

> A neocloud that belongs to its customers. B3IQ builds, hosts and operates every node as a single AI cloud. Owners finance the fleet. Teams rent the compute.

Catalog refreshed: 2026-10-09 (UTC)
Canonical site: https://b3iq.org
Docs: https://docs.b3iq.org
Concise version: https://b3iq.org/llms.txt

B3IQ builds, hosts and operates every node as a single AI cloud. Owners finance the fleet. Teams rent the compute. Run your models on bare metal. Put spare capacity to work when you choose.

Estimate optional idle-compute earnings with attributed external market observations and explicit assumptions. Rates, utilization and operating costs can change; projections are not guaranteed.

## Machine lineup

| Machine | GPUs in starting configuration | VRAM per GPU | Bandwidth per GPU | Starting price per machine | Availability | Mode |
|---|---|---|---|---|---|---|
| [H200 NVL](https://b3iq.org/machines/h200) | 8 | 141 GB | 4.8 TB/s | $329,032 | built to order | earning |
| [RTX PRO 6000 Server Edition](https://b3iq.org/machines/rtx-pro-6000) | 4 | 96 GB | 1.6 TB/s | $99,971 | built to order | earning |
| [RTX 5090](https://b3iq.org/machines/rtx-5090) | 1 | 32 GB | 1.8 TB/s | $7,880 | built to order | private |
| [Workstation RTX 5090](https://b3iq.org/machines/rtx-5090-workstation) | 1 | 32 GB | 1.8 TB/s | $12,506 | built to order | private |
| [B300](https://b3iq.org/machines/b300) | 8 | 288 GB | 8 TB/s | $560,000 | preorder | sales |

Prices are our landed component cost plus one 5% Build & AI-Guarantee fee, with no markup on parts. Configure the exact build on the next step.

## In the press

- Fortune: [Startup B3IQ has found a new AI niche: Supplying rent-to-own machines to privacy-conscious university researchers](https://fortune.com/2026/08/11/ai-startup-b3-has-found-a-compute-niche-with-a-rent-to-own-model-aimed-at-academics-clients-include-stanford-and-nyu-researchers/)
- Pantera Capital: [Why AI Builders Need to Own Their Compute with B3](https://www.youtube.com/watch?v=9cKS6FH99Xo)

## Ownership terms

Pick the GPU, memory and storage. One charge pays for the machine, with no balance left on it. You can make idle capacity available for hourly rental. If it rents, your share is 85% of gross rental revenue, less Managed Hosting. Demand is not guaranteed. In Earn mode, inference and SSH access are unavailable. Rack machines. It runs your workloads and nobody else's. You pay a flat monthly hosting fee: our cost plus a margin. Every workstation and tower, and racks too. Yes. You pay in full, and title transfers to you on acceptance, when your machine is racked, tested and online. There is no lien, no financed balance and no condition attached to it. It is your equipment, sitting in our facility because that is where it earns. Yes — it is your machine. Give us notice and we will drain its running work, decommission it cleanly and crate it. On rent-to-own, the terms set on the call decide when it can ship. Shipping and any reinstallation are at your cost, and it stops earning the day it leaves the rack. No. Private use and hourly Earn are optional, separate whole-machine modes. During Earn, inference and SSH are unavailable. Ask our team to change modes after running work finishes. Our 15% applies only to rental income; Managed Hosting remains a separate cost. Projected from clearing rates at our modeled occupancy, after the projected annual rate decline. Your machine's actual income depends on demand and is not guaranteed. Our 15% of income: offtake: contract sourcing, demand matching, routing, billing, SLA, marketplace operations and payouts. Managed Hosting: electricity and cooling, rack space in Oregon, 24/7 monitoring & on-call, physical security, machine insurance, remote-hands labor, networking & bandwidth. Rent-to-own. Select units only. We set up rent-to-own on a call.

Projected from clearing rates at our modeled occupancy, after the projected annual rate decline. Your machine's actual income depends on demand and is not guaranteed.

| Config | Machine price | Paid to you (modeled month, after fee and hosting) | Projected payback |
|---|---|---|---|
| 8× H200 NVL | $329,032 | $8,352/mo | ~84 mo |
| 4× RTX PRO 6000 | $99,971 | $2,278/mo | ~105 mo |

## Who it's for

- [AI startups](https://b3iq.org/solutions/ai-startups): Make the token bill predictable.
- [Private AI](https://b3iq.org/solutions/private-ai): Sovereignty for regulated teams.
- [Local AI, scaled](https://b3iq.org/solutions/scale-your-local-ai): Your homelab setup, datacenter-grade.
- [Agencies & consultancies](https://b3iq.org/solutions/agencies): Dedicated serving for client models.

## Financing FAQ

### Do I actually own the machine?

Yes. You pay in full, and title transfers to you on acceptance, when your machine is racked, tested and online. There is no lien, no financed balance and no condition attached to it. It is your equipment, sitting in our facility because that is where it earns.

### Can I have the machine shipped to me?

Yes — it is your machine. Give us notice and we will drain its running work, decommission it cleanly and crate it. On rent-to-own, the terms set on the call decide when it can ship. Shipping and any reinstallation are at your cost, and it stops earning the day it leaves the rack.

### Do I have to let it earn?

No. Private use and hourly Earn are optional, separate whole-machine modes. During Earn, inference and SSH are unavailable. Ask our team to change modes after running work finishes. Our 15% applies only to rental income; Managed Hosting remains a separate cost.

### Is my machine insured?

Yes. While it is racked in our facility it is covered at replacement value against physical loss or damage from fire, theft, flood and electrical events. It is on our policy for as long as it is in our rack. Coverage does not extend to data or to lost income.

### How do you handle security?

Staff access to production runs through single sign-on, and every administrative request is logged. Our SOC 2 Type II audit is underway with A-LIGN, and our written security policy set is available on request.

### What if rental rates fall?

Your income falls with them, and the machine takes longer to cover its price. We assume rates decline every year and floor them well below today's, and every payback figure on this site is computed after that decline. Nothing about your income is guaranteed. On rent-to-own, lower rates slow the machine's path to being fully yours; the terms are set on the call.

### Why is a single RTX 5090 usually shipped rather than hosted?

A single 5090 does not earn enough for hosting it with us to be worth your while. That class is a great machine to own and run yourself, so we build it and ship it to you.

## Owned-compute field notes

### You should own your compute.

B3IQ lets AI teams own a physical NVIDIA GPU machine that B3IQ builds, hosts and operates, or ships to them. Their API keys route only to their own machines, with no per-token B3IQ inference charge. Optional Earn mode sells machine time on external GPU marketplaces by the hour; demand and net income vary.

Canonical article: https://b3iq.org/blog/you-should-own-your-compute

- **What does owning compute mean on B3IQ?** It means owning the physical GPU machine rather than renting a slice of shared capacity. B3IQ procures, racks, hosts, and operates the machine under the owner's account.
- **Is inference on an owned B3IQ machine metered per token?** No. Inference on a machine you own does not carry a per-token B3IQ charge. The machine is a fixed asset, while Managed Hosting and other applicable services remain separate costs.
- **Can a B3IQ machine earn when its owner is not using it?** Earn is an optional per-machine mode for selling time on external GPU marketplaces by the hour. It withdraws the machine from B3IQ inference and SSH and boots a separate encrypted volume. Demand varies; serving inference itself produces no per-request earnings.

### How B3IQ works: from bare metal to sovereign AI.

B3IQ turns a physical GPU server into managed AI infrastructure with a host agent on the machine, an orchestration control plane, and an OpenAI-compatible gateway. Machines use cryptographic identity and outbound-only connections; account routing is limited to that account's machines.

Canonical article: https://b3iq.org/blog/how-b3iq-works

- **What runs on each B3IQ machine?** A host agent inventories the hardware, establishes machine identity, installs models and runtimes, runs benchmarks, serves a local OpenAI-compatible endpoint, and reports signed machine data.
- **Does a B3IQ node require an inbound internet port?** No. Machines connect to the control plane through outbound connections, including the tunnel used for managed SSH access.
- **How does B3IQ keep inference on the owner's machines?** The hosted gateway applies an own-nodes-only routing rule: API keys are scoped to an account and eligible inference targets are limited to machines on that account.
- **What is B3IQ's SOC 2 status?** Our SOC 2 Type II audit is underway with A-LIGN, and the report is expected in 2027. We've completed the HECVAT (full) and keep a written security policy set. Ask us for either.

## API

OpenAI-compatible gateway at https://b3iq.org/v1/api. API keys are scoped to your
own machines (a key's /v1/models lists the models your machines serve). Swap
the SDK base URL and go:

```python
from openai import OpenAI
client = OpenAI(base_url="https://b3iq.org/v1/api", api_key="<your B3IQ API key>")
```

GET /models is owner-scoped; anonymous requests have an empty model list. POST /chat/completions requires a suitable gateway key and an eligible machine on that account. Inference has no per-request charge; usage is recorded for attribution. There is no third-party provider fallback or shared public pool. GET /stats describes public aggregate telemetry, not capacity reserved for a caller. GET /pricing is retained settlement-policy metadata, not a commercial inference quote.

Additional authenticated endpoints include GET /billing/jobs/{job_id}, POST /private-inference/jobs and GET /receipts/{receipt_hash}; scopes and limits are documented in the contract.

Full machine-readable contract (OpenAPI 3.1), including request and response schemas:
https://b3iq.org/openapi.json

## Key pages

- Machines, configure & buy: https://b3iq.org/machines
- Ownership, how it works: https://b3iq.org/ownership
- GPU earnings calculator: https://b3iq.org/gpu-earnings-calculator
- Hardware & specs: https://b3iq.org/hardware
- Reviewed model requirements: https://b3iq.org/models
- Model fit and benchmark methodology: https://b3iq.org/models/methodology
- Provider and GPU comparisons: https://b3iq.org/compare
- B3IQ field notes: https://b3iq.org/blog
- HIPAA compliance, when B3IQ is a business associate, what the BAA covers, and which surfaces are out of scope for PHI: https://b3iq.org/hipaa
- Security, data retention, routing, encryption and vulnerability reporting: https://b3iq.org/security
- Trust center and evidence requests: https://b3iq.org/trust
- SOC 2 Type II: In progress. Target: January 2027. Audit underway with A-LIGN. The report has not been issued. January is a target, subject to completion of the audit. See https://b3iq.org/trust.
- HIPAA readiness: In progress. Target: January 2027. Readiness work is in progress. We already sign BAAs for agreed workloads; an executed BAA and a review of your deployment's scope are required before handling PHI. See https://b3iq.org/trust.
- ISO 27001: Planned. Timing to be confirmed. Planned for our compliance roadmap. Certification work has not started, and no certificate has been issued. See https://b3iq.org/trust.
- Accessibility statement, the WCAG 2.1 Level AA target (not a certification), known limitations, and how to report a barrier: https://b3iq.org/accessibility
- API contract (OpenAPI): https://b3iq.org/openapi.json
- Docs: https://docs.b3iq.org
- Privacy: https://b3iq.org/privacy · Terms: https://b3iq.org/terms