A computer or server built around what you need to run.
Start with the job and the budget. Discuss the parts, setup and support that fit, whether you need a first computer, a small server or a machine for a specific AI workload.
The specification starts with the workload
Tell us which applications or services you want to run, what equipment you already have and what matters most: purchase cost, noise, physical size, performance, power use or room to expand.
A parts list follows from those needs. A more expensive machine is not automatically a more useful one.
Reference configurations
These are starting points for a discussion, not fixed-price products, industry standards or guaranteed performance levels. The final quote identifies the exact parts, variants, availability and tests for your project.
Starter automation server
A compact machine for light services and an assistant that uses a cloud AI provider. A dedicated graphics card is not assumed. The right capacity depends on the actual tasks and how many need to run at once.
Expanded services host
More memory, storage and processing capacity for several services, development work or a larger workload. The specification should leave a sensible route for expansion without buying unused capacity by default.
Local AI workstation
A machine assessed against a named model and intended use. Graphics memory, model size, context length and response-time expectations all affect the choice. Local installation alone does not establish that every part of the workflow stays on the machine.
Your first or next computer
A practical desktop build for the applications you want to use. If you want to take part in the build and understand the components, ask about learning sessions with Tyler.
Running costs
Compare electricity and the wider costs of ownership with a cloud option. Change the assumptions to match your tariff, expected use and actual quotes.
Electricity is only one part of the comparison. Hardware, installation, backups, software, AI services and support may also matter. The calculator can show either option costing less; it does not assume that buying a server is the right answer.
What the quote will make clear
- The exact components and whether any are customer-supplied.
- Assembly, configuration and tests included.
- Delivery, local installation or collection arrangements.
- Software licences and any ongoing supplier charges.
- Documentation, support and warranty-handling arrangements.
- How price or availability changes will be agreed before parts are ordered.
Already own a machine?
Tell us what it is and what you want it to do. A suitability review comes before a fixed installation quote. There may be a sensible upgrade or configuration change that avoids a new build.
Plan a build
Start with a short description of the job. You do not need to choose a processor or understand a server specification before making an enquiry.
Reference build details
Planning references checked 12 November 2026. Not quotes, measured consumption or benchmarks. Labour, travel, licences, backups, AI and continuing care are separate unless itemised.
Starter automation server
Intel N100 or N150 class, four cores and four threads · 16 GB starting memory · 512 GB to 1 TB SSD, with a separate tested backup destination
Not required for cloud API use, scheduling or a brochure website
Assumed wall power: 8–20 W. Predominantly idle or light automation with intermittent small jobs
- Start with one active browser/build job as an operational proposal, not a proven capacity figure
- Do not promise many simultaneous browser sessions or VM workloads
- No large local-model performance claim
- Public websites may remain externally hosted and independent of this machine
Expanded services and VM host
Modern eight-core Ryzen 7 class or an appropriately benchmarked Intel Core equivalent · 32 GB starting memory · 1 TB NVMe SSD; consider a second SSD for the required workload and recovery design
Optional; not required for ordinary website builds or cloud model APIs
Assumed wall power: 20–70 W. Moderate mixed services, builds and browser activity with idle periods
- Concurrency and VM counts require measurements with the actual software
- A second drive or mirror is not a substitute for a recoverable off-machine backup
- Consumer mini-PC specifications do not establish enterprise uptime or service guarantees
- If ECC, redundant power or on-site hardware support is required, quote a separate server class
Optional local-inference workstation
Modern six-to-twelve-core desktop CPU selected for the whole workload · 32 GB starting memory · 2 TB NVMe SSD as a starting allowance, adjusted for model and dataset storage
Candidate starting reference: NVIDIA GeForce RTX 5060 Ti with 16 GB VRAM; verify the 16 GB SKU, not the 8 GB variant
Assumed wall power: 40–180 W. Intermittent local inference mixed with idle/light work; sustained inference can exceed this illustrative average range
- Measure the exact model, quantisation, context length, latency and concurrency before promising capability
- RAM and VRAM capacities alone do not establish model quality or performance
- This hardware is optional and is not necessary to buy a brochure website or run an API-based control service
- A low-cost cloud VPS quote is not an equivalent GPU inference service
- Local inference does not by itself prove all data stays local; inspect every API, telemetry and storage path
- Larger GPUs, sustained workloads, cooling and PSU requirements require a new quotation and power assessment
What might your setup cost to run?
Your comparison
Changes power and upfront hardware only. Review the other inputs each time.
Starting rate: 26.32p, GB domestic direct debit, October–December 2026. Use your actual tariff.
Use average socket power while running. Include relevant UPS/network power once. This is not a power-supply rating or CPU TDP.
If your wattage already averages a full 24-hour day, enter 24 here.
Electricity: per average month · annualised · kWh per average month.
The £20 example is not a GPU-inference quote. Use comparable capacity and operating hours.
Include backups, AI, support, repairs and resale
A resale assumption reduces the total; it does not fund the upfront payment.
Check these inputs
| Cost | Own hardware | Cloud |
|---|---|---|
| Upfront payment | — | — |
| Ongoing per average month | — | — |
| Hardware repair/replacement budget | — | Provider hardware only; check terms. Add application recovery/support costs above. |
| Resale at period end | — | £0.00 |
| Total over — months | — | — |
This calculator sends nothing and saves no inputs. Use Reset illustration to restore the starting values.
How to use the result
- Use one consistent VAT basis: include VAT you cannot reclaim. The calculator does not calculate tax relief or VAT recovery.
- The tariff is a dated illustration checked on 12 November 2026: Ofgem’s October–December 2026 GB domestic direct-debit rate. Business contracts and later periods may differ.
- Each average month is 365/12 days. Tariffs, workload and recurring prices stay constant for the scenario. This is not an energy-price forecast or a discounted cash-flow model.
- Only add incremental broadband or standing-charge costs. Do not count hardware both upfront and as monthly depreciation. Do not add backups again if already included in your cloud quote.
- Include domain and public-website hosting costs where needed. Owning an automation server does not automatically replace those services.
- Assess outages, security, noise, maintenance effort and recovery separately. A cheaper total does not establish equivalent capability or reliability.
Monthly kWh = watts / 1,000 × hours per day × 365 / 12
Owned total = hardware + installation + months × (electricity + other + AI + care) + repairs − resale
Cloud total = setup + months × (compute + other + AI + care)
Compare running costs
Enter your own tariff, expected use and costs. Compare the money involved in owning a machine with a cloud option for the same intended workload.
Electricity is one part of the total
Average power, running hours and your electricity unit rate determine the electricity estimate. A power supply’s maximum rating is not the same as the machine’s average use. A measurement from the complete setup is more useful than a guess from one component.
Include the costs that apply to you
For ownership, consider hardware, installation, electricity, backups, extra connectivity, software, AI services, support and a repair or replacement allowance. For cloud, consider setup, compute, storage, traffic, backups, licences, AI services and management.
Use the same tax basis on both sides. Only include additional costs caused by this setup, and avoid counting the same charge twice.
Read the result as a scenario
The result uses the numbers you enter and holds recurring costs constant over the selected period. It is not a future-price forecast, a hardware quote or a performance guarantee.
Different setups can have different availability, support and recovery arrangements even when their processor or memory figures look similar. Those differences should be considered alongside cost.
A sensible next step
If a configuration looks promising, discuss the workload and a current itemised quote before buying. The right answer may be your existing machine, a new build or ordinary cloud hosting.