the $100 AI bill: what power users actually spend

Table of content

by Ray Svitla

$20 for Claude Pro
$20 for ChatGPT Plus
$10 for Perplexity Pro
$20 for Cursor
$30-50 for API credits (Claude API, OpenAI API)
$10 for misc tools (Anthropic workbench credits, experiments)

total: $110-150/month

that’s more than netflix, spotify, and my gym membership combined.

and it’s worth every dollar.

what I’m getting for $100+/month

Claude Pro ($20): unlimited conversations with Claude Sonnet 4, priority access during peak times, higher rate limits.

I use this for writing (articles, docs, emails), planning (project breakdowns, decision-making), research synthesis (reading papers, summarizing threads), and general thinking partner stuff.

ChatGPT Plus ($20): access to GPT-4, DALL-E, code interpreter, web browsing.

I use this less than Claude, but it’s good for: image generation, certain types of structured data tasks, and as a second opinion when Claude is being confidently wrong about something.

Perplexity Pro ($10): unlimited searches with citations, follow-up questions, deep research mode.

I use this instead of google for anything requiring synthesis across multiple sources. especially good for “what’s the current state of X” or “who’s doing Y in 2026”.

Cursor ($20): AI-powered code editor with Claude + GPT-4 integration, codebase-aware autocomplete.

I use this for all coding. it survived the shakeout for a reason. the @ mention system for pulling files into context is killer.

API credits ($30-50): direct API access to Claude and OpenAI models for custom agent workflows .

I use this for: automated content generation, batch processing tasks, custom skills that run server-side, experiments that need high volume.

why this is cheap

before AI, my productivity software spend was:

total: $77-97/month

and I got way less leverage.

grammarly fixed grammar. Claude rewrites entire drafts, changes tone, restructures arguments, adapts to different platforms.

github copilot did autocomplete. Claude Code orchestrates entire features .

notion stored information. Claude synthesizes information across all my files.

the ROI isn’t 2x. it’s more like 10x.

what $100/month buys in time

conservative estimate:

writing: AI cuts my writing time by 50%. I write ~20 articles/posts per month. that’s 10 hours saved. at $50/hour (low estimate), that’s $500 value.

coding: AI writes ~60% of my code. I code ~40 hours/month. that’s 24 hours saved. at $100/hour, that’s $2400 value.

research: AI synthesizes research that would take me 10-15 hours of reading/note-taking per month. condensed to 3-4 hours. that’s 10 hours saved, $500 value.

total time saved: ~44 hours/month
total value created: ~$3400/month
cost: $110-150/month

ROI: ~23x

even if my estimates are off by 50%, it’s still a 10x return.

what power users spend more on

I know people spending $200-500/month on AI. here’s where the extra spend goes:

$200-300 tier:

$300-500 tier:

at that level it’s not “personal productivity spend”. it’s business infrastructure.

what the $0 tier looks like

you can use AI for free. sort of.

the constraint is always: rate limits or quality.

free tier is fine for casual use. “help me write this email”. “explain this concept”. “debug this error”.

but if you’re trying to do serious work — build in public , write 700 skills , code with agents — you’ll hit limits fast.

and then you’re stuck waiting. or working around limits. or compromising quality.

at that point, $20/month is a no-brainer.

the psychology of AI spending

there’s a weird guilt around paying for AI that doesn’t exist for other tools.

people pay $50/month for adobe creative cloud without thinking. but $20/month for AI feels extravagant.

why?

I think it’s because AI feels like it should be free. like search. or email.

but AI isn’t infrastructure. it’s leverage. and leverage is worth paying for.

another factor: the uncanny valley of value.

when a tool saves you 5 minutes, that feels concrete. “I would have spent 5 minutes on this, now I didn’t, worth it.”

when a tool makes something possible that wasn’t before, the value is abstract. “I could never have written this analysis manually. so what’s it worth?”

that abstract value is harder to justify. even when it’s massive.

what I don’t pay for anymore

tools AI replaced in my workflow:

grammarly ($12/month) → Claude
grammarly caught grammar errors. Claude rewrites, restructures, and adapts voice. no contest.

otter.ai ($10/month) → ChatGPT voice mode
otter transcribed meetings. ChatGPT does that plus summarizes and extracts action items.

copy.ai / jasper ($40+/month) → Claude + custom skills
those tools were GPT-3 wrappers with templates. now I just have my own skills with better context.

readwise ($8/month) → custom research workflow
readwise aggregated highlights. I now have an AI that reads articles, extracts key points, and synthesizes themes across sources.

zapier ($20/month for pro) → agent automation
zapier connected apps. AI agents run multi-step workflows without needing pre-built integrations.

total saved: ~$90/month

net cost after replacement: ~$20-60/month

the spend ceiling

most people’s AI spend will plateau around $20-50/month.

one or two subscriptions (Claude or ChatGPT), maybe a coding assistant (Cursor or Copilot).

that’s the sweet spot for individual power users. you get enough capability to 10x productivity without thinking too hard about costs.

people who go above $100/month are:

going above $200/month is rare unless you’re making money directly from AI workflows.

the API vs subscription tradeoff

subscriptions are predictable. $20/month, unlimited usage (within rate limits).

APIs are variable. you pay per token. could be $10/month. could be $200/month.

when to use subscriptions:

when to use APIs:

most power users do both. subscriptions for interactive work, APIs for automation.

the hidden costs

$100/month is the direct cost. there are indirect costs:

learning time
figuring out how to use AI well takes time. context engineering , skill building , workflow design. that’s 10-20 hours upfront, 1-2 hours/month maintenance.

quality control time
AI output needs review. sometimes light editing. sometimes “this is wrong, start over”. that’s 10-30% overhead on AI-generated work.

infrastructure time
if you run local models or home lab setups , that’s hardware costs ($500-2000 one-time), setup time (5-10 hours), and maintenance (1-2 hours/month).

context anxiety
constantly wondering “is this context good enough?” or “should I add more examples?” or “why did it misunderstand that?” is a mental tax.

but even with those costs, the ROI is positive.

who should spend $0

if you:

then free tier is fine. no shame in that.

but if you’re hitting rate limits, or waiting for access, or compromising quality because the free model isn’t good enough — just pay.

$20/month is less than one billable hour for most professionals. if AI saves you even one hour per month, it’s break-even.

who should spend $100+

if you:

then $100+/month is probably worth it.

you’re not paying for AI. you’re paying for leverage. and leverage at 10-20x ROI is cheap at any price.

what I’d pay if prices doubled

if Claude Pro went to $40/month tomorrow, I’d pay it.

if Cursor went to $40/month, I’d pay it.

if APIs doubled in price, I’d optimize my usage a bit, but I’d still pay.

the value is that high.

the only thing that would make me stop: if free or cheaper alternatives got just as good.

right now they’re not. free models are 6-12 months behind. that gap is worth paying for.

what happens when everyone pays

right now, most people don’t pay for AI. they use free tier or don’t use it at all.

that’s changing. every month more people cross the threshold from “AI is neat” to “AI is essential to my work”.

when that happens, $20-50/month for AI will be as normal as $10/month for spotify.

and the people who figured out AI workflows early — who are already spending $100+/month — will have a 1-2 year head start.

that head start is worth something. maybe a lot.

the cost trajectory

my AI spend over time:

trend: up and to the right.

will it keep going? maybe to $150-200/month if I build more agent workflows. probably not beyond that unless I’m building a business.

the dinner party test

I spend $100+/month on AI tools.

at a dinner party, if someone asked “what do you spend money on?”, would I mention it?

probably not. because it sounds either:

but I should mention it. because it’s the best money I spend.

more impactful than:

less visible, more valuable.

the alternative cost

what if I spent $0 on AI and tried to get the same results manually?

I’d need to:

total alternative cost: $2000+/month + burnout

vs $110-150/month for AI.

the comparison isn’t even close.

the forward-looking question

will I still be spending $100+/month on AI in 2027?

probably. maybe more.

because the value compounds. the better I get at using AI , the more value I extract per dollar.

and as models get better, the ceiling on what’s possible keeps rising.

right now AI can write, code, research, analyze.

in 2027? who knows. maybe it designs, negotiates, strategizes, manages.

if that happens, $100/month will look like a steal.

I track my AI spending in a simple spreadsheet. subscriptions + API usage. every month I look at it and ask: “was this worth it?”

every month the answer is yes.

not because I’m rich or because I love spending money. because the leverage is undeniable.

if you’re still on free tier and you use AI more than a few times per week, try a month of paid. track the value you get.

my bet: you won’t go back.

what’s your AI spend? is it worth it? what would make you pay more or less?

Ray Svitla
stay evolving 🐌