Home / Learn / Data & AI / Offline AI / Lesson 16.6
All lessons

Module 16: Cost: local AI against subscriptions · Lesson 16.6 · about 20 min

The hybrid budget most people should use

Four sample budgets (student, professional, small team, heavy builder) that mix free local AI with just enough cloud, in India, Canada, the USA and the UK, and how to review them every few months.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 16 · about 20 minutes · read first: 16.5. You’ll be able to set an AI budget that mixes local and cloud, and know when to change it.

For most people the best value is a local model for the daily, private and high-volume work plus one modest cloud plan for what local cannot do, reviewed every few months.

1. Why this matters

For most people the cheapest and best answer is not “all cloud” or “all local”, but a mix. A local model does the daily, private and high-volume work for the cost of electricity, and a modest cloud plan covers what local cannot do. This lesson turns that into numbers.

It is like a household food budget: cook at home most days, and eat out on purpose.

2. Four sample budgets

Prices from Lesson 16.1, checked 19 November 2026. Usage is made up.

Who Local part Cloud part Monthly (USA / India)
Student Existing laptop with Gemma 4 E4B or 12B (Module 9) Free tiers; one month of a standard plan in exam season about US$0 to 2 / ₹0 to 170 a month on average
Professional 32 GB Mac mini or 16 GB graphics card, 12B to 27B model One standard plan (Claude Pro or ChatGPT Plus) about US$20 + power / ₹2,000 + power
Small team of 8 One 64 to 128 GB office server (Lesson 5.5) Two standard plans for the hardest work, an API key with a cap for scripts about US$40 to 60 + power, after the server
Heavy builder 128 GB box or a big graphics card for agents and batch jobs One heavy plan or API with a cap about US$100 to 200 + power

In Canada and the UK, use the same structure: most plans are charged in US$ plus tax (Lesson 16.1).

Before. Priya, a data analyst in Pune, pays ₹1,999 for ChatGPT Plus and ₹1,950 for Google AI Pro: ₹3,949 a month.

Move the routine work. SQL help and report drafts go to a local 12B model on her 16 GB laptop. Her work data now stays on the laptop.

Cancel one plan. With the routine work gone, she does not need two plans. She cancels Google AI Pro and keeps Plus for hard problems: ₹1,999 plus a few rupees of power.

The saving. ₹1,950 a month × 12 = about ₹23,000 a year, and her Plus allowance now lasts longer, because it only gets the hard jobs.

3. Rules that keep it under control

  1. Local by default. Every new job tries local first (Lesson 15.6).
  2. One cloud plan, not three. Pick the one whose strengths you use most.
  3. Caps on every API key in the provider’s dashboard, so a runaway script cannot run up a big bill.
  4. Track it: a one-line monthly note of plan costs, API spend and power.
  5. Review every 3 to 6 months: new open models (Lesson 7.3) and new prices may change the answer.
monthly AI cost=hardware share+power+plans kept+API spend
hardware share
the machine's real cost spread over the months you keep it (Lesson 16.3)
power
the electricity it uses each month
plans kept
the cloud plans you still pay for, ideally one
API spend
pay-per-token use by scripts, with a cap set on each key
Physical meaning
Two pipes fill the same bucket. The local pipe is fixed and cheap however much flows through it; the cloud pipe is charged by the litre. A good budget sends the steady flow through the local pipe and opens the cloud pipe only for what the local one cannot carry.
Maths and principle

Principle: spend cloud money only where it buys something local cannot.

Equation: monthly AI cost = local hardware share + power + plans kept + API spend; aim to keep the cloud part for jobs that fail the local quality test.

Analogy: a household food budget: cook at home most days, eat out on purpose.

Try it: a team with 8 plans at US$20 drops to 2 plans and an office machine costing US$60 a month over its life. What is the saving? (160 − (40 + 60) = US$60 a month, plus the privacy.)

“More plans means better coverage.” Two or three overlapping plans mostly buy the same thing twice. Once routine work runs locally, one plan for the hardest jobs usually covers the rest, and its usage window lasts longer because it only gets the hard jobs.

4. Worked example

Made-up person, real list prices. Priya, a data analyst in Pune, paid ₹1,999 for ChatGPT Plus and ₹1,950 for Google AI Pro (₹3,949 a month). She moves routine SQL help and report drafts to a local 12B model on her 16 GB laptop, cancels Google AI Pro, and keeps Plus for hard problems. New spend: ₹1,999 plus a few rupees of power, saving about ₹23,000 a year, and her work data stays on her laptop.

5. Practice

  1. Write your own budget in the table’s format.
  2. Which plan would you keep if you could keep only one, and why?
  3. Put a date in your calendar for your first review.
Answers
  1. Your answer; include the date you checked the prices.
  2. The one you use for the work local fails at most often.
  3. Three months from today is a good start.

New words: hybrid budget, spending cap, review cycle.

  • The usual best value: local for routine, private and high-volume work, plus one standard plan for the hardest jobs.
  • Monthly AI cost = hardware share + power + plans kept + API spend.
  • Put a monthly spending cap on every API key.
  • Keep a one-line monthly note of what you spend, and review the mix every 3 to 6 months.

Check yourself

1For most single professionals, what is usually the best value?
2Why review the budget every three to six months?
3What should a team set on every API key?

Sources

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Prices and model names change fast: check the official source before you buy.