How to cancel your AI subscriptions without losing the capability

In May 2026 an essay titled, roughly, "The solution might be cancelling my AI subscription" made the rounds, and the comments agreed faster than usual. People were tired of optimizing their day around a meter. There's even a popular thread on Hacker News about scheduling sleep around Claude usage limits. Cancelling isn't a tantrum; for most of what you do, it's just the right call. Here's how to do it without losing a thing you actually use.

Step 1: Add up the bill

Open your card statement and list every line with "AI" in it. The stack creeps because each charge looks small on its own. A typical loadout: ChatGPT Plus, Claude Pro, Cursor, GitHub Copilot, Perplexity. Five tools, none of them outrageous alone, and you're past $90 a month, north of a thousand dollars a year for software you half-forget you're paying for. We did the full breakdown in the $90/month AI subscription stack; the number surprises almost everyone.

This isn't only a consumer feeling. Axios ran a piece called "AI sticker shock" on May 28, 2026; a few days later, on May 30, came "Corporate America Is Starting to Ration AI as Cost Skyrockets." And on May 17 an industry post warned that "AI subscriptions are a ticking time bomb for enterprise." When companies are rationing the thing and rate-shopping their vendors, a solo subscriber paying full retail for five overlapping tools should at least add it up.

Step 2: Find the part that doesn't need the cloud

Now sort what you actually do with these tools into two buckets. The first is the everyday stuff: rewriting an email, summarizing a long thread, fixing a function, asking what a config option does, turning notes into a draft. Be honest about the ratio; for most people that's something like 90% of their AI use. The second bucket is the genuinely hard 10%: novel architecture, a gnarly proof, the reasoning task where you'd notice a weaker answer.

Here's the thing nobody selling you a subscription will say out loud: the first bucket doesn't need a frontier model, and it definitely doesn't need a meter. A model running on your own Mac handles drafting, summarizing, ordinary coding, and Q&A without a round-trip to anyone's server. On a 54-prompt comparison, Outlier's Core 27B matched Claude Opus on 98.9% of rubric checks; the full method and results are published here. That's not a claim that local is smarter — it's that for the work that fills your day, you stop being able to tell the difference, and you stop paying per use.

Step 3: A clean cancellation checklist

Don't rage-cancel on a Tuesday and lose six months of chats. Do it in order:

1. Export your data first. Every major provider has a data export in account settings. Download your conversations and any saved projects while the subscription is still active, because access often ends when billing does.
2. Note what each tool actually did for you. One line per tool: "Cursor: inline edits across files," "Perplexity: sourced answers." This is how you'll know whether the replacement covers it.
3. Install the replacement. Set up the owned tool that's going to absorb the daily 90% before you cancel anything, so there's no gap.
4. Run a week in parallel. Keep both for seven days. Throw your real work at the local tool and only reach for the cloud when you truly need to. You'll see your own ratio fast.
5. Then cancel — and check the off-ramp. Some subscriptions make leaving deliberately annoying. The US FTC fined Shutterstock $35M over hard-to-cancel subscriptions; if a cancel flow buries the button, that's the tactic, not your mistake. Keep clicking.

Outlier fits the "install the replacement" step cleanly: one signed download, no account, no terminal, no Docker, and it works with Wi-Fi off. The full walkthrough is in install local AI on a Mac.

Step 4: What to keep (honest)

Cancelling everything isn't right for everyone, and pretending otherwise would be a worse sales pitch than the truth. The cloud still wins the hardest reasoning, and it's faster: local Core 27B runs about 20.7 tok/s on an M1 Ultra where cloud flagships sit around 80–100. If your job regularly lives in that hard 10% (research at the edge, the proof that has to be right), keep one cloud subscription on purpose. The move isn't "cancel all AI." It's "stop renting five tools to do the work one owned tool does for free, and pay for the frontier only when you actually use it."

That's the real pitch, and it isn't about IQ. It's ownership and no meter: the everyday model is a file on your disk, it can't be retired or repriced mid-project, and its marginal cost is your wall socket instead of a per-token tab. If you want the longer version of that argument, it's in what owning your AI actually means. Keep what earns its keep. Cancel the rest. Your statement will be shorter and your work won't notice.

Frequently asked questions