Is AI subscription poverty real? I built a spending line from household benchmarks, and a checker to go with it

In Part 1 I established that my AI spending is 44,408 yen a month, about 7.7 times the average for Japanese IT and web engineers (5,776 yen). On the amount alone, clearly a lot.

But one thing stayed stuck. "A lot" compared to whom, measured how?

The same 40,000 yen a month sits completely differently on someone taking home 300,000 and someone taking home a million. Comparing to an average tells you whether you're unusual. It does not tell you whether you're overpaying. For that you need a ruler based on the ratio to income.

I went looking, and there is no "AI spending should stay under X% of take-home pay" benchmark anywhere. If it doesn't exist, build one from the household benchmarks that do. That's this post.

Start by remembering the rent benchmark

Before AI spending, note that households already have a few "this is roughly normal" benchmarks.

• Rent is about 25 to 30% of take-home pay. A column from MUFG Bank gives that range as rent that won't leave you scraping. Some material suggests up to about a third for people living alone
• Food is about 15% of take-home pay. Card companies' example budgets often use rent 25–30%, food 15%, utilities 5%, communications 5%, savings 10–20%
• The 50/30/20 rule. Introduced by the US Consumer Financial Protection Bureau (CFPB): 50% of take-home pay to needs, 30% to wants, 20% to savings. The CFPB itself notes it isn't an absolute rule everyone can follow

If rent is 100,000 yen, working back from 25–30% implies take-home pay of roughly 330,000 to 400,000. On 200,000 take-home, 100,000 rent is heavy. On 400,000, it's normal. The same 100,000 means different things.

That "look at it as a ratio to income" idea carries straight over to AI.

AI spending lives inside the 30% you get to spend

Under the 50/30/20 rule, AI isn't a necessity, so it sits in "wants": the 30%.

Obviously, spending all 30% on AI would be strange. So, borrowing household benchmarks, here are the boundaries.

• Within 3% of take-home pay → ordinary. Little effect on the household
• 3 to 5% → AI enthusiast. Heavy user, but entirely plausible
• 5 to 10% → deep in it. Worth checking whether other hobbies or investments are getting squeezed
• Over 10% → the AI-poverty line
• Over 20% → you are now working for the AI

Why 10% for the poverty line: 10% of take-home pay is the common savings benchmark. Put another way, you're paying an AI exactly what was supposed to go into savings. 5% is a sixth of the 30% you spend freely — a reasonable ceiling for one hobby among several.

These are not official standards. They're lines this blog placed, based on the benchmarks above. Worth stating plainly.

Actual numbers make it suddenly concrete

Here are the 5% and 10% lines by take-home pay.

The same "I pay 30,000 yen a month for AI" is:

• 10% on 300,000 take-home. Heavy
• 5% on 600,000. Perfectly plausible if you use it for work
• 3% on a million. Not particularly abnormal

Which means "30,000 a month on AI, therefore too much" doesn't stand up on its own. It only means something as a ratio to income. Same as rent.

The scarier calculation, in reverse

Here's the real point. "If I want to spend this much on AI, what does my take-home pay need to be?" — worked backwards under the 5% condition.

If your AI spending is 50,000 yen a month, keeping it within 5% requires 1,000,000 yen a month take-home. At 100,000 a month, two million.

You were going to use AI to earn efficiently. Are you now working to pay for AI? A number that makes you pause.

So where do you land?

You can check that here, mid-post. Enter your take-home pay and the total you pay for AI each month. Nothing you type is sent anywhere; it's division happening inside this page.

The checker itself is Japanese-only for now → AI subscription poverty checker. The inputs are take-home pay (手取り) and monthly AI spending (AI課金).

An important exception: work use needs a different ruler

One key caveat. If you use AI for work and it directly produces revenue or saves time, this ratio isn't the right lens.

In that case it's a business expense, and the honest measure is ROI. If 30,000 yen of AI a month removes 100,000 yen of work, it looks expensive as a household ratio and is entirely rational as a business decision.

So the working definition of AI subscription poverty in this post is:

A state where personal AI spending exceeds 10% of take-home pay and squeezes savings, living costs or other discretionary spending.

If it pays for itself at work, that isn't poverty, it's investment. The dangerous case is saying "well, I use it for work too" while it isn't paying for itself. That would be me. Part 1's "how many hours does 44,000 yen need to buy?" was exactly that check.

So who is this for?

• Paying more than 10,000 yen a month for AI → check the ratio against take-home pay once. Under 3%, don't worry about it
• Saying "but I use it for work" → separately confirm whether that work recovers what you pay. If it does, it's investment; if not, it's a hole
• About to start paying for AI → using 5% of take-home pay as your ceiling keeps it from creeping

Once more, to be clear: "up to 5% / AI poverty from 10%" is not a standard set by any institution. It's an indicator this blog built, referencing the household ratios in the material below.

• Basis for the boundaries: J-FLEC "Money basics" (rent around a third of income), J-FLEC "Plan your life by subtraction" (income − savings = spending, one tenth of annual income), CFPB "Analyzing budgets" (the 50/30/20 rule)
• Supporting Japanese household benchmarks: MUFG Bank on what share of take-home pay rent should be, Sumitomo Mitsui Card on saving while living with family, and on rent when living alone

Household benchmarks vary a little by period and source. The figures here are the general benchmarks as checked in September 2026.

Next: what a ratio to take-home pay can't see

This ruler was "what percentage of take-home pay". Simple and clear, and it has a weakness. On the same 300,000 take-home, someone living alone in Tokyo and someone living with family have completely different amounts of free money. The same 10% is not the same weight.

So next time, using the Family Income and Expenditure Survey and the Housing and Land Survey, I generate "your standard household" from your family structure and prefecture, subtract living costs and savings, and judge what percentage of the money you can actually spend freely goes to AI. And I confess my own result.

• Part 1: Can you actually go broke on AI subscriptions?
• Part 2: this post
• Part 3: Is that AI spending actually expensive for your life?