Services
Managed private AI, end to end
We deliver private AI as a complete business system — infrastructure, document intelligence, security governance and ongoing model care — sold as measurable workflow automation and privacy, not just hardware.
What We Deliver
Eight services, one outcome: AI that earns its keep — privately
1 · AI Readiness Assessment
A structured engagement covering workflow analysis, data‑sensitivity classification, security posture review and current AI spend. You receive a written business case — productivity estimates, risk analysis and a recommended deployment architecture — suitable for partner or board review.
2 · Paid Pilot Program
8–12 users, one or two high‑value workflows, a limited document repository and a 60–90 day evaluation window. We measure baseline productivity, test response quality against your standards, and validate security and permissions before you commit to the full platform. Charged pilots prove real demand — for both of us.
3 · Turnkey Deployment
The full system, installed and hardened: GPU server and redundant storage, identity and access management, encryption and audit logging, document ingestion and permission‑aware search, retrieval‑augmented generation, and connectors to Microsoft 365, SharePoint, CRM, ERP and file servers. Employee training included.
4 · Document Intelligence & RAG
Your firm's knowledge, finally searchable. Permission‑aware retrieval over internal procedures, client files, policies and correspondence. Draft proposals, reports and responses in your house style; summarize large document sets; extract structured data from invoices, forms and reports; produce first drafts of contracts and client letters.
5 · Security & Governance
Local hosting isn't automatically secure hosting. We implement network segmentation, role‑based permissions, endpoint security, secure backups, patch management, physical access control, incident response, and prompt‑injection and data‑exfiltration defences — aligned to the NIST AI Risk Management Framework.
6 · Model Management & Routing
No single local model is best at everything. We run model evaluation against your workflows, deploy the best fit for each task, and implement hybrid routing — local‑first processing with a secure, policy‑controlled fallback for unusually complex requests. Quarterly model refresh keeps quality improving.
7 · Training & Adoption
Technology only pays back when the team uses it. Role‑based training sessions, prompt guides for your workflows, office hours, adoption dashboards and an internal champions program — so 32 licenses become 32 power users.
8 · Managed Support & Model Care
24/7 monitoring, GPU and driver lifecycle management, search‑index health, backup and disaster‑recovery testing, helpdesk support and quarterly governance reviews. A recurring support contract keeps the US$150,000 asset performing — and keeps your support burden at zero.
Pricing
Three ways to buy — one that fits your balance sheet
Indicative pricing for a typical 32‑user deployment (GPU server, storage, software platform, integration and first‑year support). Final pricing follows the assessment and pilot.
Own It
Hardware purchase plus recurring service. The asset sits on your balance sheet; you control it outright.
From US$150,000installed · typical 32‑user platform
- GPU server, storage and networking — installed and hardened
- RAG platform, connectors and permission mapping
- Employee training and go‑live support
- US$30,000–60,000/yr software, support and security plan
- Optional integration projects from US$25,000
Managed Private AI
A subscription. We own and finance the infrastructure; you get the capability without the capital purchase.
US$7,500–12,500per month · 36‑month term · setup US$20,000–50,000
- Full platform deployed on your site, owned by TPU Systems
- 24/7 monitoring, updates and helpdesk included
- Standardized, continuously improved hardware platform
- Quarterly governance and model‑quality reviews
- Predictable cost — no per‑seat or per‑token fees
Hybrid
Customer‑owned local inference, our management layer, and a secure cloud fallback for unusually large or complex requests.
Custompriced by users, usage and support tier
- Sensitive documents never leave your hardware
- Policy‑controlled routing to cloud for heavy tasks
- Lower hardware entry point than full on‑prem
- Flexible monthly service fee
- Best for mixed sensitivity workloads
Every engagement begins with an assessment, and we recommend a paid pilot before any full‑scale commitment. We will not sell a platform we haven't proven against your workflows.
The Business Case
What a platform has to earn back — and how it does
| Value component | How it's realized | Typical annual impact |
|---|---|---|
| Avoided subscriptions | Consolidating per‑seat AI tools onto the private platform | US$8,000–38,000 for 32 users |
| Productivity | 32 users saving 20–30 minutes per workday at US$40/hr loaded cost | US$106,000–160,000 theoretical value |
| Risk & compliance | Eliminating shadow‑AI exposure; controlled, audited data flows | Avoided breach exposure — shadow‑AI incidents added ≈US$670,000 to average breach costs (IBM) |
| Workflow automation | Drafting, summarization and extraction replacing external spend or manual hours | Varies — measured during pilot |
Annualized platform economics: US$150,000 over a 4‑year life (≈US$37,500/yr) plus US$30,000–60,000 operating — so the platform needs roughly US$100,000–150,000 of annual measurable benefit to justify comfortably. Sources: IBM Cost of a Data Breach Report; NIST AI RMF.
Get Started
Start with an assessment, not a purchase order.
A free consultation maps your workflows, estimates your business case and proposes a pilot with measurable success criteria.