Infrastructure Solutions
One fixed monthly fee on a 3-year plan, with hardware, AI models, the always-on Faraday Arc, support and updates included. No seat limit and no per-token charges.
Starter
Great for 2-9 people
Open-weight models it runs
Models up to about 60B parameters, such as:
- Qwen3.8-27B ~17 GB · 4-bit Fast coding, multilingual work, image and video understanding
- Faraday Arc (Adaptive Reasoning Companion), always on 24/7/365
- Mac Studio with M5 Max chip and 128GB unified memory
- Private AI computing
- Expandable infrastructure
- End-to-end encryption
- Basic support, updates & maintenance for the full term
- No seat limit or per-token charges
Growth
Great for 10-19 people
Open-weight models it runs
Models up to about 150B parameters. Everything in Starter, plus:
- Nemotron 3.5 Super VL ~80–87 GB · 4-bit NVIDIA’s 120B model for images, documents and agent tasks
- GLM-5.3-Flash ~93–109 GB · 1–2-bit Business analysis and documents, as a compressed build
- Faraday Arc (Adaptive Reasoning Companion), always on 24/7/365
- Mac Studio with M5 Ultra chip and 256GB unified memory
- Private AI computing
- Expandable infrastructure
- End-to-end encryption
- Enhanced support, updates & maintenance for the full term
- No seat limit or per-token charges
Scale
Great for 20-29 people
Open-weight models it runs
Models up to about 350B parameters. Everything in Growth, plus:
- GLM-5.3-Flash ~180 GB · 4-bit Business analysis, documents and image input
- MiMo-V2.6-Flash ~170 GB · 4-bit Efficient all-round assistant
- MiniMax M3 ~240 GB · 4-bit Long-context and multimodal agent tasks, a tighter fit at 428B
- DeepSeek V4.1 Flash ~250–350 GB · 1–3-bit Agentic coding and long documents, as a compressed build
- GLM-5.3 ~223–239 GB · 1–2-bit The full GLM-5.3 flagship, as a compressed build
- Faraday Arc (Adaptive Reasoning Companion), always on 24/7/365
- Mac Studio with M5 Ultra chip and 512GB unified memory
- Private AI computing
- Expandable infrastructure
- End-to-end encryption
- Priority support, updates & maintenance for the full term
- No seat limit or per-token charges
- GDPR & HIPAA compliance
- Industrial spy protection
Faraday Arc Adaptive Reasoning Companion
Included in every plan. Arc runs on the Mac Studio in your office 24 hours a day, 7 days a week, 365 days a year, working through your company files in the background so the answers are ready when you ask.
- Always on, 24/7/365Runs around the clock on your own Mac Studio, with nothing sent to the cloud.
- Always-on forecasts and analysisForecasts and checks re-run as new files arrive, so your numbers are never last month’s spreadsheet.
- Instant search and interpretationAsk about anything in your company files and get an answer that points to the source.
How the monthly plan works
- 36-month term, billed monthly in US dollars.
- We provide the hardware. Faraday Machines buys, installs and maintains the Mac Studios; support, updates and hardware replacement are included for all 36 months.
- No seat limit and no per-token charges: your whole team can use it as much as it needs.
- A fixed commitment, with a 30-day guarantee. If it isn’t right for you in the first 30 days, you get your money back. After that, like a professional services contract, the plan runs for the full term and can’t be cancelled early.
- At the end of the term, return the hardware or renew the plan.
Sizes are approximate. A model needs room to work on top of its file size, so as a rule of thumb from our own testing, models of about 30–60B parameters need 128 GB, 60–150B need 256 GB and about 300–350B need 512 GB. 4-bit builds stay close to the full model. Compressed 1- to 3-bit builds, such as Unsloth’s dynamic GGUF releases, let larger models fit smaller tiers but give up some quality: Unsloth reports its builds keep roughly 70–88% of the full model’s accuracy, and the loss shows most on long or precise work. The larger models use most of a tier’s memory, so they run one at a time. A 4-bit DeepSeek V4.1 Flash (about 400–445 GB), Kimi K3 and MiMo-V2.6-Pro need a larger setup with more than one machine. Compare the models in our open-source model comparison, see what they cost per task in Dollars per Task, and check their terms in the open-weight license tracker.
What a 10-Person Team Pays Over 3 Years
Cloud AI Subscriptions
Top-tier plans such as ChatGPT Pro or Claude Max at $200 per person per month × 10 people × 36 months
You rent access. Prices increase. Data leaves your premises.
$599/month × 36 months, with hardware, support and updates included
No seat limit. Price fixed for the term. Data stays private.
Public statements and announcements
Canada Is Building Sovereign AI
Over the past year, Ottawa and Canada’s largest telecom companies have made keeping AI and data under Canadian control a national priority. Their projects are data centres for governments, researchers and large enterprises.
-
Evan Solomon, Minister of Artificial Intelligence and Digital Innovation. Speaking at the All In conference in Montreal, warned that if Canada does not build AI capacity at home it will have to “rent” it from its neighbours and follow their rules. He described sovereignty as domestic growth plus partnership with countries that share Canada’s values.
Source: The Globe and Mail, BNN Bloomberg
-
Bell AI Fabric, Cohere, Hypertec and BUZZ HPC. Announced a sovereign AI compute deal to run Cohere’s models on Canadian infrastructure at Bell’s facility in Merritt, B.C. HIVE, the parent of BUZZ HPC, reported the three-year contract at about US$220 million.
-
Prime Minister Mark Carney. Launched AI for All, Canada’s national artificial intelligence strategy, built on three principles: trust, opportunity and sovereignty. CBC reported more than $2 billion in funding.
Source: Prime Minister’s Office, CBC News
-
Government of Canada and TELUS. Announced work on a proposed sovereign AI factory cluster in British Columbia under the federal initiative for large-scale sovereign AI data centres.
Source: Innovation, Science and Economic Development Canada, TELUS
-
Darren Entwistle, President and CEO of TELUS. Opening Canada’s first sovereign AI factory in Rimouski, Quebec, said businesses, researchers and governments should not have to rely on foreign-controlled systems to advance their AI work.
Source: RCR Wireless News
Faraday Machines applies the same principle at the size of an office: the AI model and your data stay on hardware you control, in your building.
These are public statements and announcements by governments and companies, summarized from the sources linked. None of them refers to or endorses Faraday Machines.
Scale With Your Team Growth
Each Mac Studio delivers 50 tokens per second on a frontier model. As your team grows and AI demand increases, add more units to maintain fast response times.
Start with one Mac Studio
Every plan is a single Mac Studio, sized to the models you need: 128 GB, 256 GB or 512 GB of unified memory.
Add units as needed
When you need extra performance, simply add more Faraday units. Each additional unit improves overall capacity and performance.
Stack compute
For heavy workloads and ultimate performance, upgrade up to 4 stacked Mac Studio units for highest cost-performance in the world.
Unlock more performance as needed
1 Mac Studio = ~50 tokens/sec • 2 x Mac Studios = ~100 tokens/sec • 4 stacked Mac Studios = 200+ tokens/sec
Your Hardware Gets Smarter Over Time
AI models improve up to 10x in capability every 18 months. Your Faraday hardware benefits from these advances. Learn how on-premises AI creates a flywheel of innovation.
Frequently Asked Questions
What is included in Faraday Machines pricing?
Faraday Machines plans are a fixed monthly fee on a 36-month term, in US dollars: Starter is $299 a month for a Mac Studio with the M5 Max chip and 128GB of unified memory, Growth $599 for an M5 Ultra with 256GB, and Scale $999 for an M5 Ultra with 512GB. Faraday Machines buys, installs and maintains the hardware, with preloaded open-weight AI models, end-to-end encryption, and support, updates and replacement for all 36 months. There is no seat limit and no per-token charge. Every plan also includes Faraday Arc (Adaptive Reasoning Companion), always on 24/7/365, with continuous forecasts and analysis and instant search and interpretation across all company files.
What is Faraday Arc (Adaptive Reasoning Companion)?
Faraday Arc (Adaptive Reasoning Companion) is the platform included in every Faraday Machines plan. It runs always on, 24/7/365, on the Mac Studio in your office: it keeps forecasts and analysis up to date as new files arrive, and lets your team search and interpret all company files instantly, with answers that point to the source. Starter, Growth and Scale all include it at no extra cost. See it at work in The AI CFO That Lives in Your Office.
Which Faraday Machines tier is right for my team size?
Faraday Machines sizes its tiers by team size: Starter ($299 a month, 128GB) suits teams of 2–9 people, Growth ($599 a month, 256GB) suits 10–19 people, and Scale ($999 a month, 512GB) suits 20–29 people. There is no seat limit on any tier. Each plan is one Mac Studio: more memory runs larger models, and more machines can be added for bigger teams or heavier workloads.
How does the 36-month plan work?
The plan runs for 36 months, billed monthly in US dollars. It starts with a 30-day money-back guarantee; after that it can’t be cancelled early: like a professional services contract, the full term is payable. Faraday Machines buys, installs and maintains the hardware, and support, updates and hardware replacement are included for the whole term. At the end of the term you return the hardware or renew the plan.
What AI models run on each Faraday Machines tier?
Each tier runs the open-weight models that fit in its memory. As a rule of thumb from our own testing, models of about 30–60B parameters need 128GB, 60–150B need 256GB and about 300–350B need 512GB. At 4-bit quantization, Starter runs Qwen3.8-27B, Growth adds NVIDIA’s Nemotron 3.5 Super VL, and Scale adds GLM-5.3-Flash, MiMo-V2.6-Flash and, as a tighter fit, MiniMax M3. Compressed 1- to 3-bit builds, such as Unsloth’s, let larger models fit with some loss of quality: GLM-5.3-Flash on Growth, and DeepSeek V4.1 Flash and the full GLM-5.3 on Scale. Kimi K3 and MiMo-V2.6-Pro need larger setups.
Does Faraday Machines offer a money-back guarantee?
Yes. Every Faraday Machines plan is backed by a 30-day money-back guarantee. Before signing up, you can also see or try a demo system at the Faraday Machines office in Toronto, or have it brought to yours. The demo mimics the expected performance of a full deployment, so you can test it with your own data and workflows before committing.
What happens if a Faraday Machines unit fails?
Faraday Machines replaces it. Every plan includes updates, maintenance and hardware replacement for the full 36-month term, on top of the standard hardware warranty. If a unit fails, Faraday Machines ships a replacement and walks you through the swap, typically resolved within 48 hours. No IT staff is needed: initial setup is handled remotely, and a management dashboard covers day-to-day administration.
How much does on-premises AI cost compared with cloud AI over 3 years?
For a 10-person team over 3 years, top-tier cloud AI plans such as ChatGPT Pro or Claude Max at $200 per person per month cost $72,000. Faraday Machines Growth costs $599 a month, or $21,564 over the 36-month term, with hardware, support and updates included and no seat limit: a saving of about $50,000.
Own Your AI Infrastructure
One fixed monthly fee, unlimited use, no seat limit. Talk to us about the right configuration for your team and lock in your costs for three years.
Schedule Consultation