Local Ollama vs ChatGPT: cost over 3 years
This page may contain Amazon affiliate links. A purchase after a click can earn the site a commission, at no extra cost to you. The advice stays independent. How affiliation works.
Choosing between ChatGPT Plus (OpenAI’s online service) and a local AI with Ollama (a free app that runs models on your PC)? This page compares the cost over 36 months for a household. The prices are indicative ranges read in October 2026, plus named scenarios you can retune with the hardware selector and the electricity budget. This page does not invent a catalog price.
Two ways the money leaves
| Approach | What you pay | What you own |
|---|---|---|
| Cloud (ChatGPT Plus) | A subscription that renews every month | Access to models OpenAI hosts, with updates included |
| Local (Ollama) | The hardware purchase (money up front) plus electricity, and upgrades if you choose them | The machine, the models you downloaded, and your conversations if you keep them off the public internet |
Finance people often call that up-front hardware purchase capex. Here it is simply the hardware you buy at the start.
A 36-month comparison (order of magnitude)
Shared assumptions: one cloud seat of the ChatGPT Plus type. On the local side, a mini PC or a tower already sized for Ollama and GGUF models. GGUF is the file format that stores the model’s weights, compressed, for local inference.
| Line | Cloud, indicative | Local, indicative |
|---|---|---|
| Subscription, 36 months | ~€720 (about €20 × 36, public list) | €0 (Ollama is free) |
| Hardware | €0 (the cloud’s machines stay on their side of the bill) | €600–1,500 depending on the profile |
| Electricity, 36 months | Folded into the subscription on the user’s bill | ~€50–200 depending on average watts and the price per kWh |
A nuance on cloud kilowatt-hours. Your ChatGPT bill does not show kilowatt-hours, and the service does use energy in data centers. You fund that through the subscription. At home the link is direct, on your meter. That is why this page points at electricity and the home AI budget for realistic watts (a mini PC left on, versus a GPU tower used now and then).
Scenario A: a moderate household
- Use: 2 adults, daily questions, email summaries, simple homework help.
- Cloud: ChatGPT Plus alone, about €720 over 36 months, and €0 of hardware.
- Local: a mini PC with 32 GB of RAM, a “balanced” profile around €850, plus about €90 of electricity (assuming about 25 W average, and about €0.25/kWh). About €940 over 36 months.
- How to read it: the totals sit close together. Local adds privacy and a household chat with no vendor usage cap.
Scenario B: a heavy household
- Use: 3 or 4 people, long sessions, Whisper transcription, and sometimes local images.
- Cloud: one subscription can become the limit (queues, quotas). Some households add a second tool, and the cloud budget goes past €720.
- Local: a tower or a mini PC plus a mid-range RTX GPU, about €1,200, plus about €150 of electricity. About €1,350 over 36 months, with much more room for VRAM and SSD storage.
Answer quality: cloud and local
Flagship cloud models are often ahead of a local 7B–8B on hard reasoning. For summarizing, rephrasing, sorting ideas, or playing with the kids, a local model from choosing a first model is enough for many households. The private family assistant is built on that compromise.
Hardware: where the money goes
| Profile | Hardware range, October 2026 | What it suggests |
|---|---|---|
| Budget | €500–800 | A 32 GB mini PC, CPU or iGPU, small GGUF files |
| Balanced | €800–1,300 | 32–64 GB. See the mini PC comparison |
| Comfort | €1,300–2,500 and up | A Mac mini M4 or a GPU with 12–16 GB of VRAM |
Costs that hide
- Setup time. A first local AI in 15 minutes, then Open WebUI. Count an evening, and a month is the wrong unit.
- Disk. A library of models. See SSD and storage.
- Noise and power. A mini PC left on if the machine lives in the living room.
- Updates. Windows, macOS, or Linux, and GPU drivers. Ollama itself has no software subscription.
Privacy and the rules that still apply
Local hosting does not erase every duty (children’s data, health). It does keep each conversation off a third-party server. Read privacy when the AI stays home before you load medical or school PDFs.
A quick decision table
| Lean toward… | When… |
|---|---|
| Staying on the cloud | You want the newest model, you would rather skip PC admin, and your use is light |
| Moving local | Sensitive data, heavy family use, Home Assistant, or n8n |
| A mix | Cloud for the occasional creative job, Ollama for the private everyday |
What to do next
- Estimate your real kilowatt-hours on the electricity budget.
- Cross that with the hardware selector (total cost at 12 and 36 months).
- Install a proof of concept: Docker for Ollama and Open WebUI, or a native install.
- Compare the chat with LM Studio if you prefer a graphical app.
How the numbers were built
We do not read your real bill. We multiply public ranges by 36 months. On the cloud side, the reference is OpenAI’s ChatGPT Plus list, checked in October 2026. On the local side, the hardware comes from the profiles in choosing hardware and the mini PC comparison. Electricity follows “average watts × hours × price per kWh” as written on the electricity budget. If your price per kilowatt-hour is higher (late 2026), raise the local column to match.
A worked electricity example
A machine at 35 W average, on 24 hours a day: 35 × 24 × 365 ≈ 306 kWh/year. At €0.25/kWh, about €77/year, or about €230 over 36 months. The table’s €50–200 band is where you land if the machine sleeps often, or runs less than continuous GPU inference.
What this comparison leaves out
- Technical support. The cloud has a help center. At home, the getting started guides and the community are the help.
- Aging hardware. A GPU bought in 2026 will still run quantized models. A cloud subscription hands you the new models without a hardware upgrade.
- Beyond chat. ComfyUI / FLUX or Whisper push local use past chat alone. The cloud would mean other subscriptions.
A decision you can make this week
| Question | Move |
|---|---|
| “I want to try this without paying” | 15 minutes on the PC you already have |
| “I want a total at 12 and 36 months” | The selector plus this page |
| “I want family privacy” | The private assistant plus a quiet machine |
A mid-range hardware example (Amazon partner): NVIDIA RTX cards on amazon.co.uk. Check the current price (October 2026).
Useful links, in this section and next door
- Choosing hardware for home AI
- 32 vs 64 GB of RAM
- GPU, iGPU, eGPU, and OCuLink
- Local coding with Continue
- Install Ollama on Windows
Appendix: keep the budget after you decide
Revisit this comparison every 12 months: the ChatGPT price, the home price per kilowatt-hour, and what the hardware would resell for. A mix: keep Plus for a quarter while Ollama comes up to speed, and count both columns. Reselling a mini PC: the 2026 second-hand market moves around. The hardware selector uses a conservative band. Children: on a local machine the extra cost per child is zero, against several cloud accounts. Whisper and images: add the SSD and GPU lines (SSD, GPU). Support: a parent’s admin time is a hidden cost. If you compare with turnkey online software (SaaS, software as a service), value something like 2 hours a month × 36 months.
Outside a household: tax, a small business, a professional resale
This page is for households. It does not offer a small-business tax deduction, and it does not give accounting advice on depreciating hardware. Professionals should match their own rules with an accountant. What a mini PC resells for after 36 months stays a conservative estimate in the selector. It is not a guaranteed second-hand price.
Common questions
What does ChatGPT Plus cost per year in France?
On OpenAI’s public price list (October 2026), count about €20 a month tax included, or about €240 a year if you stay on monthly billing. Check the pricing page when you subscribe. A yearly plan is sometimes offered at a discount.
When does the local hardware pay for itself?
Often between 18 and 36 months if you buy a machine around €800–1,000 and several people use it regularly, including an electricity estimate from our budget page.
Is electricity included in the cloud price?
For you as a subscriber, yes: the cloud provider pays for its data centers. At home, the kilowatt-hours are on your bill. See the home AI electricity budget for typical watts of a mini PC or a GPU tower.
Can a household use both?
Yes. Many households keep ChatGPT for very new models and use Ollama for sensitive data. The total is then the subscription plus the hardware spread over time.
Sources
Narrow it down with the hardware selector · French version of this section