Private AI for privacy-sensitive businesses
Private AI your whole team can use. Your data never leaves your control.
Private AI for businesses that handle private data. Your files stay on hardware you control. No token fees. Setup + support included.
No token limits. No per-seat fees. No surprise bills. Your AI runs on your hardware, so your team can use it as much as they want.
Forge AI Labs builds, installs and runs a private AI lab for your company: a GPU server running open-weight models, in your office or a data center. It works with your own documents, has no usage caps, and costs one fixed monthly amount instead of per-seat or per-use AI bills. We handle setup, support and upgrades.
- You own or control it. Buy, lease, or rent the hardware.
- Data stays with you. Prompts and files stay on your server.
- Fixed cost. No per-seat or per-use AI billing.
- Managed plan on every lab. We handle setup, support and upgrades.
Unlimited usage. No token fees.
Your AI runs on hardware your business owns or controls, so there's no per-token billing, no per-seat fees, no usage caps and no surprise invoices. Your team can use it all day. The only limit is how fast the system we size for you can answer, and we build it to fit your headcount.
Per-seat AI plans run about $20 to $30 per person a month, and premium plans about $100. For a 100-person team, that's about $2,000 to $3,000 a month on standard plans, or about $10,000 on premium plans, every month, with usage caps. A Forge lab for 100 people leases from about $2,721 a month, plus $1,995 a month for full support, and the hardware is yours.
All prices are estimates and may change.
Why Forge
A private AI lab that you own or control, set up and run for you.
Your data never leaves your control
Prompts and documents are processed on your own server, not sent to an outside AI service.
One fixed monthly cost
No per-seat or per-use AI bills. You know what AI costs before the month starts.
No usage caps
The whole team can use it. Capacity is set by your hardware, not by a plan limit.
Works with your own documents
Search and ask questions across the files you choose to connect.
You own or control the hardware
The server is yours, leased in your name, or rented and hosted for you.
Buy, lease or rent
Every option comes with a managed plan, so the lab keeps running after day one.
We handle everything
Setup, support and upgrades, including new open-weight models as they are released.
Built for privacy-sensitive industries
Law, healthcare, finance, accounting and real estate teams that handle confidential information.
How it works
Four steps from first call to a working private AI your team uses every day.
- 1
Assess
We learn how your team would use AI, how many people need it, and what data it should reach. Then we recommend a right-sized setup.
- 2
Install
Your hardware is built to order and shipped direct from our distribution partner (lead time varies). We set it up at your office or in a data center, load open-weight models, and connect it to your login and files.
- 3
Run
Your staff use a familiar chat interface and document search. Everything runs on hardware you control.
- 4
Support
Every lab includes a managed plan: monitoring, security patches, model updates, backups and user help, plus hardware upgrades when you need more capacity.
Industries served
Any business that handles sensitive information and wants AI without sending that information to an outside service.
Law firms
Review contracts, draft memos, and search case files while client documents stay on your own systems.
Healthcare
Help with notes, letters, and internal knowledge search on infrastructure you control. You set the policies.
Finance
Analyze reports and answer internal questions without uploading financial data to third-party AI tools.
Real estate
Summarize leases, draft listings, and search transaction files across your team.
Accounting
Work through client files, policy documents, and tax research with data kept in-house.
Insurance
Summarize claims and policies and search internal guidelines on your own hardware.
Running AI on your own hardware can make it easier to meet your own data-handling rules, but it does not by itself make you compliant with any regulation. Your compliance team decides what is required.
Forge Starter Box
Private AI for 5-15 people, preloaded and plug-and-play. Starting at about $12,195 one-time (estimate). Optional support from $349/mo.
Get the Starter BoxBuild your AI lab
Pick options to see a rough estimate. Estimated pricing
More build options
Each package can also be built on alternative hardware: to work around supply delays, to save money, or to plan for growth. Every option is priced in your quote. Estimated pricing
Forge Starter Box (5-15 users)
Alternatives to the NVIDIA DGX Spark, all built on the same NVIDIA GB10 chip.
- Acer Veriton GN100 (NVIDIA GB10, 128 GB, 2 TB)In stock
Our closest match to the DGX Spark: same NVIDIA GB10 chip, 128 GB of memory and NVIDIA DGX OS.
Priced in your quote.
- ASUS Ascent GX10 (NVIDIA GB10, 128 GB, 1 TB)In stock
A budget-friendly GB10 option. Best for teams that keep their documents on a network drive, since it has 1 TB of local storage.
Priced in your quote.
- HP ZGX Nano G1n AI Station (NVIDIA GB10, 2 TB)Backordered
For offices that standardize on HP hardware.
Priced in your quote.
GPU workstation (about 25 employees)
Alternatives to the 1x NVIDIA RTX PRO 6000 workstation.
- Workstation with NVIDIA RTX PRO 5000 Blackwell (48 GB)Check availability
A lower-cost build for small teams running smaller AI models (up to about 30 billion parameters).
Priced in your quote.
2-GPU server (about 100 employees)
Alternatives to the standard 2x NVIDIA RTX PRO 6000 server.
- Supermicro 5U server (2x RTX PRO 6000 Server Edition, dual AMD EPYC)In stock
A higher-end choice for growing teams: dual processors and room to add GPUs later, so you can scale up toward the 500-employee setup without replacing the server.
Priced in your quote.
- Matched pair of 2U 2-GPU servers (primary + failover)In stock
For teams that want two identical servers, one primary and one standby, so the lab keeps running if one goes down. Ordered as a pair.
Priced in your quote.
4-GPU server + failover (about 500 employees)
Alternatives for the largest setup.
- Matched pair of 4U 4-GPU serversIn stock
The same standard 4-GPU server, ordered as a pair. A good fit for multi-site deployments.
Priced in your quote.
- Configurable Supermicro 4U server (up to 8 GPUs)In stock
For teams that need more than 4 GPUs or a custom configuration. Your quote reflects the GPUs you choose.
Priced in your quote.
1-GPU do-it-yourself kit
Alternative graphics cards for the hardware-only kit.
- NVIDIA RTX PRO 6000 Blackwell Max-Q (96 GB, 300 W)Backordered
The same 96 GB of GPU memory at lower power: quieter and cooler for office use, and lets you fit two or more GPUs in one tower.
Priced in your quote.
- NVIDIA RTX PRO 5000 Blackwell (48 GB)Check availability
A lower-cost kit. 48 GB of GPU memory suits smaller AI models.
Priced in your quote.
Availability is checked with our partners and can change quickly; we confirm stock, lead time and price with your quote. Ask about a build option.
How we get it running
Every Forge AI Lab runs free, open-weight AI models on hardware your firm owns or controls. Pick the setup that fits your office:
Plug-and-play box
We load the operating system, AI model, chat screen and document search before it ships. You plug it into power and your network, and we finish setup remotely with a short onboarding call.
Remote install
Already have a GPU workstation or server, or bought hardware-only from us? We install and test the software remotely, then hand it over.
On-site install
We come to your office to set up the hardware, connect your sign-in and documents, and train your staff in person.
Hosted in a data center
Your lab runs in a secure data center. Your team reaches it through a private, secure login, with nothing to install in your office.
How your team uses it
Open a link in any web browser on the office network (or a desktop app), and sign in with your existing Microsoft or Google work account. Remote access is through your firm's VPN only.
Your documents, your rules
We connect only the folders you choose, and we keep group permissions so each team sees only what it should.
Kept up to date
With a managed plan, we monitor the system, apply security updates, test new models every month and back up your settings, all for one fixed monthly fee.
Pricing and setup options are estimates; we confirm them in your quote.
Buy, lease or rent
Every lab deal includes setup and a managed plan. Estimated pricing
Buy
Own the hardware outright. Paid at order. From (Starter) plus setup and a monthly managed plan.
Lease
Spread the hardware cost through our financing partner. Starting at /mo, rates may apply, plus setup and managed plan.
Estimated. Rates may apply and are subject to credit approval by our financing partner.
Rent
One monthly fee covering hardware, data-center hosting and managed support. Starting at /mo, rates may apply, plus setup and deposit.
Estimated. Rates may apply and are subject to credit approval by our financing partner.
Forge holds no inventory. Hardware is built to order and shipped direct from our distribution / system-builder partner; typical lead time varies with GPU supply and is confirmed with your quote.
Just want the hardware?
If your team will run its own AI lab, we can supply the GPU workstation or server only, built to order and shipped direct to you. Add our setup service if you want help getting it running.
Forge Starter Box
NVIDIA DGX Spark, 5-15 users. From (estimate). Optional setup .
GPU workstation
1x NVIDIA RTX PRO 6000 (96 GB). From (estimate). Optional setup .
2-GPU server
2x RTX PRO 6000, 192 GB GPU memory. From (estimate). Optional setup .
1-GPU do-it-yourself kit
1x RTX PRO 6000 parts kit you assemble and run yourself. From (estimate). Optional setup .
Estimated pricing. Hardware-only orders do not include a managed plan. Shipping, sales tax and lead time confirmed at quote. Price a hardware-only order in the configurator.
Questions, in plain language
What is "private AI"?
It is an AI assistant, similar to the chat tools you may already know, that runs on a server your company owns or controls. Questions and documents are processed on that server instead of being sent to an outside AI company.
What are open-weight models?
They are AI models whose files are published so anyone can download and run them on their own hardware. Examples include gpt-oss-120b, Qwen, Gemma and DeepSeek. Each has its own license terms, which we review with you.
Are they as good as the big subscription AI tools?
For many everyday tasks, such as summarizing, drafting and searching documents, open-weight models can work well. The most advanced cloud models may still be stronger on some tasks. We help you test with your own work before you commit.
Where does the server live?
Either in your office (you need suitable power, cooling and a secure spot) or in a professional data center, where the server sits in a rented cabinet with power and cooling provided.
Do I have to buy the hardware?
No. You can buy it outright, lease it through our financing partner, or rent it hosted by us. Lease and rental prices are estimated; rates may apply and are subject to credit approval by our financing partner. Every lab deal includes setup and a managed plan.
Do you keep hardware in stock?
No. We hold no inventory. Each system is built to order and shipped direct from our distribution / system-builder partner. Lead times vary with GPU supply; we confirm the estimate with your quote. The "In stock" and "Currently unavailable" labels on this site show whether our partners can supply that build right now.
Does this make us compliant with HIPAA or other rules?
Not automatically. Keeping data on your own hardware can help with your policies, but compliance depends on how the whole system is set up and run. Your compliance or legal team should decide what is needed.
Is this cheaper than per-seat subscriptions?
It depends on your size and usage. For small teams, standard per-seat plans can cost less. The case gets stronger with more users, heavy use, or rules that keep data off outside AI services. We will show you the numbers honestly.
Build my lab
Tell us a little about your company. Your configurator choices are included automatically.